Start here
This is the method we use when we work with a team. It's free, and there's no catch hiding on page 40. We give the process away. We charge for running it with your team as the neutral outsider in the room, and for helping build what comes out of it.
Who this is for
The person who wants to lead it. An owner, an executive, a manager, or the person on the team who is excited about all this.
Any size. A whole company, one team, or a team of one. For the most part, these processes are size agnostic.
Anyone feeling the pressure. You're considering AI, you've deployed it with poor results, or you're paying for tools nobody uses. Possibly (probably?) all three.
What you'll have at the end
A goal your people agree on.
The numbers to pay attention to that will show whether things got better.
A map of the root causes of the problems you face, built by the people who live with them.
Every AI dollar you spend checked against that map.
A roadmap that takes you from your current reality to the way you want things to be.
At least one real change made, and a measurement to track to show whether it worked.
Ideally, buy-in* from the rest of the team.
*In our experience, this is the part that can be the most challenging from the inside and is one of the top causes of failure. You must pay attention to this, even if you think that some people are "just being difficult". Ignored disagreement leads to resentment, and resentment leads to tools and processes that are discarded. Also, the people with the most resistance often have the most useful insights to give!
How it's organized
We like to call our method Pattern Interrupt. The idea is to purposefully disrupt your normal way of looking at things in order to build something better.
The process has two halves.
Aim: Get clear on what matters and see what's really going on.
Strike: Build a 6-month roadmap, and act on it with your people.
There are nine steps in all. Aim is steps 1 to 4. Strike is steps 5 to 9.
Aim
Get clear on what matters and see what's really going on.
- 1Goal
- 2Ground
- 3The Wall
- 4Root causes
Strike
Build a 6-month roadmap, and act on it with your people.
- 5Roadmap
- 6People
- 7Deploy
- 8Check
- 9Adapt
Before the steps, there are three short sections: principles, mindset, and what we mean by AI. Please don't skip them. Every step along the way is simple on purpose, but doing them with the wrong mindset is how they fail.
You'll notice this playbook goes a long way before it helps you pick an AI tool. That's on purpose. Where AI fits, and how to choose tools, is covered in the Roadmap and Deploy steps. Everything before that is the work that makes those choices good ones.
This playbook will help you even if you never use AI. Good system design is good with or without it. AI is an option, and it can make a well-designed system much better. Good system design is the foundation for getting a return on anything you add, AI included. Start automating a system that isn't well defined, and there are a lot of ways for it to go wrong.
We don't ignore AI until the end, either. Before the steps, there's a short section on what AI is, so everyone starts with the same understanding. In step 2, you'll ask your people how they already use it. In steps 5 and 7, you'll choose tools, once you know what problem they're for.
Two ways to use it
On paper. Print it. Use pens, sticky notes, and a wall. This is still the best way to do it with a group. In our experience, sticky notes remain undefeated for having a fun, insightful time as a group working on identifying root causes and building a plan for the future.
With an AI. There's a second version of this playbook: a set of instructions you can hand to an AI, so it walks you through each step. Use the best model you have access to. This can be paired with in-person activities. We like to do analog work as a team, then have AI help keep it organized well digitally.
Either way, one rule holds. Think for yourself, and let the AI assist. Don't ask it to do the thinking for you. More on that in the Mindset section.
How long it takes
The short version, as little as an afternoon (more time is better): agree on what you're trying to achieve, get everything that's going wrong out of heads and onto a wall, take a quick pass at what's really causing it, and check your AI spending against what you find. Bring a list of your software and AI tools and what they cost. You'll know whether your AI money is aimed at the right thing.
The full version, two to four weeks: a few hours a week. You'll finish with a roadmap, a first change, and a first measurement. If you finish it in a weekend, you either run a very small business or skipped something.
A word on where we're coming from
We've worked on business process design from both the inside and outside of businesses of all sizes. One thing that is remarkably consistent across different types of teams and businesses is that even the most skilled operators develop blind spots that are obvious once someone points them out, and almost impossible to see from inside the firefighting. That's what running a business does to your field of view, however good you are at it.
We've also found that the personal dynamics of being coworkers can dramatically affect what gets communicated. This is why a third-party facilitator can be helpful in conducting these activities. People will tell us things that they won't tell a supervisor, and sometimes those insights are the ones that generate the big unlock.
If you're running this yourself, plan for that. Have people write before they talk, collect the touchy answers without names, and let someone other than the boss lead the candid parts.
Principles: how we think about all this
These are some of the things we believe. We've earned them the usual way, often by being wrong first. Take what holds up in your business and argue with the rest.
1. People come first.
The point of all of this is to empower your people to solve your biggest problems, not replace them with an AI tool. We don't take work that's only about cutting payroll, and this playbook isn't built for it.
We acknowledge that there may come a day when AI can do all of it on its own. If that day comes, we'll all have to revisit the nature of work and meaning. In the meantime, the right move seems clear enough to us: help people do better work and live better lives. Don't stress them out or make their lives worse.
2. The answers should come from your people.
Your people already know where the business is stuck, even if they aren't saying anything. Often nobody has asked them. A plan they help build is a plan they'll actually use. A new-fangled plan handed down from the leadership team without feedback is one that will be abandoned in a few months (if it ever gets acted on at all).
3. AI is the easy part. People are the hard part.
This is related to the first two. Most tools are actually pretty easy to set up and use, once you know what you should be doing. What fails is the rollout nobody asked for, aimed at a problem nobody had, handed to people who weren't consulted.
Force it on people and the outcome is simple. Your people will leave. Some will leave the building. Some will leave the tool. Some won't say anything but will actively try to sabotage the entire project.
4. The resistance is the way.
When people push back, drag their feet, or quietly stop using something, it's tempting to treat that as a thing to get past so the work can get done.
Go looking for the resistance, and lean into it. It usually points at something real: a problem nobody has named, a fear nobody has answered, or a flaw in the plan. The people who resist the most often have the most useful things to tell you.
Ignore it, and it turns into resentment. Resentment turns into tools and processes that nobody uses.
5. Protect what's already working.
Most of what works in a business is cheap, simple, and solves a real person's real problem. Protect it. It also sets the bar for what a fix should cost.
6. Most AI spending speeds up the wrong work.
Every business has a few things that limit how much it can get done, and usually one that limits it most. Speed up something else, and the work piles up in front of that spot. You've bought a faster treadmill. You are not getting anywhere sooner.
So find what's holding you back most before you decide what to buy, if anything. We call that the constraint.
It isn't always easy to tell which one matters most, and there can be more than one. A well-reasoned answer that you can test is enough to start.
7. Process must come before AI.
By process, we mean how a piece of work actually gets from start to finish: who does what, in what order, with which tools, and what happens when something goes wrong. Process design is deciding that on purpose instead of letting it happen by accident.
A messy process with AI on top is a faster mess. If the work isn't written down, agreed on, and done roughly the same way twice, there's nothing solid to automate yet.
So before any AI, look at the process itself. Is it effective, meaning it produces what it's supposed to? Is it efficient, meaning it gets there without wasted steps, waiting, or rework? Would the people who do it describe it the same way? If you can't answer those, the next purchase is a guess.
Designing a process has a process of its own. See how the work happens today. Agree on what it's for. Cut what shouldn't be there. Simplify what's left. Only then decide what to automate. The rest of this playbook is that, one step at a time.
Good news: fixing the process is often the whole win. Picture a team that asks for an AI quoting agent when what it needs is a sign-off rule. (The rule was cheaper.)
None of this depends on AI. Good system design is worth doing with or without it, and it's the foundation for getting a return on anything you add later.
8. Delete before you automate.
This one is closely related to number 7. Good process design mostly is deleting before you automate.
Business processes grow over time. Each step was a response to what was needed at one moment, built with the tools and people available at that moment. If it works well enough, it keeps going, even after the reason for it is gone. That's how you end up with redundant steps, outdated tools, and inefficiencies that could be replaced quickly.
One of the worst and most flagrant wastes of money in all of AI deployment is automating a step that doesn't need to exist anymore.
The best step in any process is the one you no longer do. Cut the steps that shouldn't exist, then simplify what's left, then automate. In that order.
9. Know your numbers before you build.
If nobody can say what number should move, and what it is today, you're buying a demo. Demos are lovely. They don't make payroll.
A surprising number of businesses don't know what their most important numbers are. That's not an insult. A business can run for decades knowing two things: whether enough money is coming in, and whether the books balance. Everything between the first lead and the final dollar is a blur.
Sometimes the number exists, but it lives in one person's head. Sometimes the number everyone watches is a stand-in for the one that matters, because the real one is hard to get. And sometimes the data is there, but nobody trusts it.
It's okay if you don't know your numbers yet. Finding them is part of the work, and it comes before building anything. Without them, you can't tell what's working, what isn't, or whether the thing you just paid for made a difference.
10. Most important, most effective fix first.
Start with the fix that will do the most good, measured against your goals. Sometimes a quick win comes first to build momentum, the way some people pay off their smallest debt first. That's a fine choice when you make it on purpose.
11. The goal is to be more effective, not busier.
Point AI at the wrong place and it creates infinite work. More drafts, more messages, more things to review. The goal is to get more of what matters done, ideally with less stress.
And hours saved only count if they go somewhere. "We saved 200 hours" is the start of a sentence, not the end.
12. The future isn't guaranteed.
There's a range of outcomes with this technology, for a person, a business, and a society. Some are very good. Some are bad, and those are real. We'd rather work toward the good ones on purpose than hope.
That starts with us. Be effective, stay human, and keep your relationships intact. Then help your people do the same.
13. We might be wrong.
The method is built on well-tested ideas. But your business has nuances we've never seen. Where this playbook and your reality disagree, believe your reality and tell us about it.
Mindset: how to think while you do this
The steps are the easy part. (You may be noticing a theme.) What decides whether they work is how you think while you do them.
Interrupt the pattern
A pattern interrupt is purposefully generating a fresh perspective. You need one because the daily firefight narrows what you can see.
We're not against the pattern. Habits, familiar workflows, and the tools you already know are how anything gets done. Nobody can rethink every task every morning, and nobody should. Routine is what lets a team move fast.
The same goes for norms, both the ones you agreed on out loud and the ones that grew on their own. "The owner signs off on every quote" was a good idea once. Every one of these is there because it solved something.
The catch is inertia. The better a routine works, the harder it is to see that there's another way to do it. It stops being a choice and becomes "how it's done." Interrupting the pattern is how you make it a choice again.
Here are a few reliable ways to interrupt a pattern:
Change what you focus on. Ask and answer difficult questions. That's what this guide is all about. There are specific processes you can use to generate a fresh perspective.
Bring in another person. Someone who doesn't know why things are done this way, and is willing to ask.
Physically change your surroundings. A different room counts. So does putting paper on a wall instead of a screen.
Use as many as you can. At a minimum, try not to do this at your desk, squeezed in between emails. Your desk is probably where the patterns are the strongest.
Who's in the room matters for the same reason. Some of the most valuable people to have there are easy to overlook:
The one who looks at things differently. Most teams have someone everybody knows for it. Invite them on purpose.
The one who isn't afraid of the awkward pause, or the awkward conversation.
The newest person on the team. They haven't yet stopped noticing what the rest of you no longer see, and some of the best ideas come from them.
Then pair them with the people who have the tenure. Experience and leadership know why things are the way they are. Fresh eyes ask whether they still should be. You want both in the room.
Think for yourself. Let AI assist.
An AI will happily write you a goal, a root cause, and a six-month roadmap in forty seconds. It will be well formatted. It will also be a guess about a business the AI has never set foot in.
So use this order:
Write your own answer first, even a rough one.
Then ask the AI to question it, find the holes, and offer what you missed.
Decide for yourself what to keep.
The value of this work is that you and your people understand your own business better at the end. You can't outsource that part.
Describe what's happening, not what to buy
"We need a better CRM" is a solution wearing a problem's clothes. "Customers wait four days for a quote" is a problem. Focus on the specific problems people are actually experiencing. That includes your customers and the people doing the work.
When you're mapping problems, resist the temptation to jump to the solution. Jumping is how you end up paying for a tool that fixes a symptom of something deeper. It's also how you end up paying for a tool when a simpler, more elegant fix was available.
We'll show you exactly how to write a problem down when we get to that step.
Be the dumb person in the room
Someone needs to ask the questions that seem like they have obvious answers. We'll go into more detail about this in the specific processes, but it is very easy to yada yada over the parts where the biggest insights and improvements can be made.
"Why do we do it that way?" "What happens after that?" "Who told you that was the rule?" Somebody has to ask these questions. It's a surprisingly enjoyable job once you stop worrying about looking smart.
Make it safe to say the true thing
None of this works if people can't be candid. A few habits help:
Write first, talk second. Everyone writes silently before anyone speaks. The loudest voice doesn't get to set the agenda.
Rank stays outside. If the most senior person speaks first and everyone agrees, you've learned nothing. Senior people go last.
Name the fear. If people are worried AI will replace them, say it out loud early. Unspoken, it comes back later as quiet resistance. Get all the concerns out early so they can be worked into the plan.
Thank the bad news. The person who tells you what's broken is doing you a favor. Act like it.
Treat your map as a hypothesis
When you think you've found the real problem, hold it up and ask, "Is that accurate?" Then listen. Being corrected by the people who do the work is the method working, not failing.
No matter how exciting an insight may be, it is also good practice to sit with it for at least a little while before acting. It can be tempting to rush off to execute on a fresh insight, but sometimes an even bigger insight is right around the corner. You obviously don't want to get stuck in a planning loop, but once you've decided to do the work of the pattern interrupt, milk it for all the value it can give you.
Don't take on too much at once
This work will produce a long list of problems, projects, and ideas. That's a good sign. It's also a risk, because a long list worked on all at once is a list where nothing gets finished.
So you need a way to decide what comes first, and there's more than one good answer. It's a lot like paying down debt. Some people go after the balance with the highest interest rate, because that's where the math says the money is. Others pay off the smallest balance first, because a quick win builds the momentum to keep going.
In a business, the first approach is fixing the thing that holds everything else back. The second is fixing something small and visible, so your people see that this time it's real. Both work. What doesn't work is having no approach at all.
Pick an approach on purpose, then work the list methodically. When more than one thing is moving at once, it should be because each one has its own owner, and nobody is carrying three.
You can't keep up with the tools, and that's okay
AI tools are changing faster than any business can track, especially a business that also has to do its actual work. That isn't about to change. If anything, it's speeding up.
So if the idea of picking an AI tool leaves you completely overwhelmed, that's normal. To a point, you should embrace it. Nobody is keeping up, including the people who say they are.
It's also why the order of this playbook matters. You don't want to be paying for a tool that isn't solving a mission-critical problem. And you don't want to be locked into a year-long contract for something that may not be the best option in three months. Knowing what you need, and staying free to switch, beats knowing every tool.
Hold the outcome loosely
You'll start this process expecting a certain outcome. Maybe it's a tool, a project, or a particular fix. Hold that loosely.
What we hold on to tightly is fixing the core problems. This work, and the mapping most of all, has a way of turning up things you weren't looking for. Sometimes a side quest turns out to be the main quest.
That's okay. Be ready for it.
A note on how we like to work
We think serious work goes better when people enjoy it. We also know that most team-building exercises are terrible. There is nothing worse than being stuck in a room full of people pretending to have fun when you know there is important work you could be getting done. So the games in here are the work itself, and each one ends with something your team will use. Games make a room candid and keep it awake. If a step in here feels like a compliance exercise, change how you're running it. When we facilitate these exercises, we like to build games around them.
What we mean by AI
Before you decide where AI belongs, make sure everyone means the same thing by it.
Why this comes before the steps
"AI" has turned into a catch-all word for all sorts of different tools. Put ten people in a room and you'll get ten pictures of it: a chatbot, a feature in the email program, a robot that takes jobs.
If you aren't clear about what it is, it's really easy to invest poorly. So spend a few minutes here first.
This section doesn't pick tools. That happens in steps 5 and 7, once you know what problem a tool would be for. It's here so everyone can follow the conversation until then.
A few ways to think about it
AI is a broad term, and no single definition covers all of it. Here are two ways to think about it that we find useful.
More intelligence built into software, and software that's easier to use. It can read, write, summarize, make sense of messy information, and take instructions in plain language. That changes what software can do for you, and which software you need at all.
A capable new assistant who doesn't know your business yet. It knows a great deal in general, and it works fast. It knows nothing about your business until you tell it.
Neither is the whole picture. Together, they're enough to follow the rest of this section.
The forms it comes in
You'll run into AI in a few different forms. Most businesses end up using more than one.
A chat assistant. You ask, and it answers. It drafts, summarizes, explains, and analyzes. A person starts every task and reads every result. This is the form most people have tried.
AI inside software you already have. Your email, documents, spreadsheets, and accounting software have probably added AI features. You may already be paying for these without using them.
An assistant set up for one job. A chat assistant that's been given your instructions, your documents, and your examples, so it does one job your way each time. Answering questions from your handbook is a common one.
An agent. AI that takes actions as well as answering. Given a goal and access to your systems, it works through several steps on its own: looking things up, updating records, sending messages.
Automation with AI inside it. A workflow that runs by itself when something happens, like an order arriving. Ordinary rules handle the predictable steps, and AI handles the ones that need reading or judgment.
AI that builds things. AI can now write working software. A small custom tool, a database, or a report that once needed a developer or an expensive subscription can often be built quickly.
The kinds you've used for years. Forecasts, recommendations, fraud alerts, and turning speech into text are AI too. They've been inside your software for a long time.
What it's particularly good at
Reading a lot of text quickly, and summarizing it.
Writing first drafts.
Pulling information out of messy documents and putting it in order.
Sorting things: what kind of request this is, and who should get it.
Answering questions from your own documents.
Working with data. A lot of expensive software exists mainly to make databases easier for people to use, and AI is very good at that.
Being a thinking partner: offering options, finding holes, and checking your logic.
Where the biggest risks live
It can be confidently wrong. It makes things up at times, and it sounds exactly the same when it does. Anything that matters gets checked by a person.
It's uneven. In one study of 758 consultants, people using AI did better and faster work on tasks it suited. On a task it didn't suit, they were 19 percentage points less likely to get the right answer. You often can't tell which kind of task you have by looking at it, and the line keeps moving as the tools improve.
It doesn't know your business. It only knows what you give it. Without your context, you get answers that would fit any company.
Data security. Know where your information goes, who can see it, and what the tool's provider keeps. Take the most care with customer, employee, and financial information. Decide what never goes into an AI tool before anyone starts using one.
Agents can be tricked. An agent that can take actions for you opens the door to new kinds of attack. The best known is called prompt injection. Someone hides instructions in an email, a web page, or a document, and the agent reads them and follows them as if they came from you. That doesn't mean you need to be scared. It means you need to be ready. Give an agent as little access as the job needs, and keep a person on anything involving money, customers, or records.
It can create more work. If you're not careful, AI simply produces more: more drafts to check, more messages to read, more to keep up with.
What it might unlock for your team
Keep these in the back of your mind as you work through the steps. You'll spot candidates as you go.
Time back from the boring tasks.
Faster answers for customers.
Information that's no longer stuck in one person's head.
A simpler, cheaper set of software.
Things you couldn't afford before, like a custom tool or a proper analysis.
Where we often start
A lot of this playbook is about finding the right places to use AI in your own business. That work matters. We're putting one place here, ahead of the steps, because we think it pays off for almost any team.
Keeping good notes, keeping records, and keeping track of projects takes a lot of time. That's true for everybody.
It's also work that some people just don't do well, however skilled they are at everything else. Notes don't get taken. Records don't get kept. Paperwork runs late. Tasks, projects, and deadlines get lost.
This is why one of the highest-leverage things AI can do for you is help your people manage their projects better. Set up well, it can give even the most disorganized person superb organization skills. Keeping things up to date can be as simple as leaving a voice note.
It doesn't take much. You can set it up in a spreadsheet, with very little investment. Your people keep doing the work. The AI helps them keep track of it.
Even the most well-organized people and teams, and people who specialize in admin, can benefit from this. If you have someone who is stellar at it, they're a good person to look after it: making sure the AI is right, and keeping everything up to date. It doesn't have to be one person.
We think this is one of the most common constraints in any business. It limits how much time each person gets to spend on the work they're best at, and the work they enjoy most. Give that time back, and each person can do more. Because people work together, the gain multiplies. When one person's notes, tasks, and records are in order, everyone who depends on them moves faster too.
It helps with this playbook, too. As you work through the steps, AI is good at keeping your notes, lists, and maps organized.
So you don't have to wait for your map before you start here. The steps will still show you what else is holding you back, and where to go next.
We're not married to it
We aren't doing this because we're desperate to add AI. We're doing it because AI might help, and because it's easy to spend badly when you aren't clear about your own system first.
Sometimes the right fix has no AI in it at all. It's a deleted step, a changed rule, or an honest conversation. That's a good result.
There's a lot AI could do for you. We'll get to it once you know what it's for.
If you're leading this
Early on, ask everyone to write down what they mean by "AI" and what they've already used. Compare the answers, then walk through this section together. Plan on about fifteen minutes.
Before you start
An hour of setup saves a week of confusion. Do these five things first.
1. Pick the scope
Decide who this is for: the whole organization, the leadership team, or a specific team. Any of these can work, and you can always run it again somewhere else.
If you don't yet know where the real problem is, start wide. Run it with the leadership team, or with people from across the business. Pick one team too early and you risk a local win. The sales team generates a ton of new leads, the real constraint was somewhere else, and now there's a bigger pile in front of it.
Go narrow when you have good reason to believe that's where the constraint is, or when one team is all you have the authority to change.
Then answer three questions in writing:
What triggered this now? A new initiative, a failed rollout, a budget decision, a vendor pitch, a nagging feeling.
What decisions does this need to produce? If you can't name one, you're planning a nicer staff meeting.
Has this been tried before? If a tool or a consultant already came and went, write down what happened. People remember.
2. Pick the people
A leader with the authority to change how the work gets done. Without one, you'll produce a lovely map and nothing else.
The people who do the work, front line included. The owner's description of a process and what really happens are often two different processes.
A group of any size. This can work with a few people or a full company retreat. The bigger the group, the more it matters to split into good subgroups for some of the exercises, so nobody gets drowned out. Past about twelve people, start splitting.
Working alone? It still works. Where a step says "the team," find two or three people who'll give you honest answers by phone or email. Customers count.
3. Make it clear what you're doing
The last thing you need is for these exercises to become one more source of fear and uncertainty. People notice when leadership starts asking questions about how the work gets done. If nobody explains why, they'll write their own explanation, and it won't be a kind one.
So before you start, the leader tells everyone involved what this is and why you're doing it. Something like: "We're going to figure out where, how, and why to use AI here. We're doing it people first. You know this work better than anyone, so we're starting by asking you."
Be open about what you don't know yet. You can't say exactly what AI will be used for, because finding that out is the point. What you can say is how the decision will get made, and that your people are part of making it.
Make sure everyone means the same thing by "AI," too. Walk through the "What we mean by AI" section together, or send it around. People can't weigh in on something they each picture differently.
Only say it if it's true. If it isn't true, this is the wrong playbook, and your people already suspect.
4. Take the pulse
We recommend a simple, anonymous survey of everyone involved. You're about to ask your people a lot of questions. Before you get into the deeper ones, take the pulse and see where things stand, so you aren't working from assumptions about how people feel. Send it along with your explanation of what you're doing and why. Three statements, each scored 1 to 5:
| Question | Score before | Score after |
|---|---|---|
| I understand where we're going with AI. | ||
| I have a say in how my work changes. | ||
| I'm willing to change how I work. |
These don't have to be the exact statements. Change them to fit your organization, but ask something like them. Add a space for anonymous comments, and invite anyone with a specific concern to come and talk it through.
You'll ask again at the end, and that's where it earns its keep. Things may not be as bad as you assumed. And the work isn't done until these scores show meaningful improvement. It turns "are people on board?" from a feeling into a number.
5. Gather the gear and block the time
We're about to shift into the actual processes. When you're ready to run them with your team, here's what you need.
On paper: sticky notes in two colors, markers that work, a big blank wall, and this playbook printed.
Remote: a shared whiteboard everyone can edit.
Either way: coffee and snacks. A caffeine deficit and low blood sugar is a great way to ruin these exercises.
Rough times for the full version:
| What you're doing | Time | Who |
|---|---|---|
| Goal: Agree on what you're trying to achieve, and note what's already working | Half a day at minimum | Leader and team |
| Ground: See the work as it actually happens, and check the data | 3 to 5 hours, spread over a week | Whoever is leading it, plus short conversations with the team |
| The Wall: Get everything that's going wrong out of heads and onto the wall | 2 hours | Everyone |
| Root causes: Map how the problems connect, and find what's really holding you back | 4 to 6 hours, spread over at least 2 days | Everyone |
| Roadmap: Build the action plan from your current reality to the way you want things to be | 2 to 3 hours | Leader and team |
| People: Engage the people who will do the work | 1 to 2 hours, then ongoing | Leader and owners |
| Deploy: Make the change that matters most, and finish it | Days to weeks | Owners |
| Check: See whether the numbers moved | 30 minutes each, at about two weeks, then one, two, and three months | Leader and owners |
| Adapt: Adjust as you learn, then go after the next problem | 1 hour each quarter | Leader and team |
Getting the problems onto the wall and mapping what's causing them are the heart of it. Give them uninterrupted chunks of time, and don't cram them into one day. Spread across a few days, you get a break and come back with a fresh perspective. A conference room is fine.
Aim: steps 1 to 4
Get clear on what matters and see what's really going on.
Step 1: Goal
What we're trying to achieve, and what's already working.
Why it comes first
A problem is only a problem compared with a goal. "Quotes take four days" is a disaster if you're demand constrained, and irrelevant if you're supply constrained.
It's also where many AI projects go wrong. They start with the tool, then go looking for something to point it at.
Nobody gets to skip this one
We've found that the smaller the business, the more likely it is to think it can skip goals. The same goes for the older ones. In our experience, this is often where the problems start. It doesn't matter how small you are, whether you're a local service business, or whether you've been at it for fifty years.
A business without goals runs on autopilot. The goal doesn't have to be about money. It can be quality of life, getting ready for a sale, or a better place to work. What matters is being as explicit as you can about what you're trying to do.
It's also more motivating. If you or your people seem checked out and going through the motions, there's a good chance the mission isn't compelling enough or the goal isn't exciting enough. Getting clear on both starts to change that right away.
What you need a good sense of
Before you go looking for problems, you need a good sense of six things:
Why you exist, who you serve, and what you value.
What's already working and worth protecting.
Where you want to be a year from now.
Your goals: what matters most right now, and what has to be true for each one.
How you'll measure them, and where you stand today.
Whether your people agree.
Start with the five-minute check
Do this first, whatever framework you use. Everyone taking part answers four questions alone, in writing, with no conferring:
What is our mission? Why do we exist?
Who are we trying to serve, and what do they need from us?
What are our most important goals right now?
How will we know we're getting there?
Then read the answers out. You're finding out three things. Do we have good goals? Does everybody understand them? Is everybody working in the same direction?
Where the answers match, good. Where they differ, you've found the first real piece of work. Don't smooth it over.
You'll run the same check again at the end of this step.
If you already have this
If your organization has a mission statement and a goal-setting framework you like, and together they cover those six things, keep using them. We're not here to replace something that works.
Bring what you have, run the check, use the parts below to fill any gaps, and move on.
If something's missing
If what you have falls short, or you're starting from nothing, here's how we think about each one. Take as long as you need. For a team doing all of it, half a day is the minimum, not the maximum.
1. Why you exist, who you serve, and what you value
The mission. If you have a mission statement, read it aloud. Is it still true? Would a customer recognize you in it? If you don't have one, write two sentences on why the business exists.
The customer. Who are they? What do they hire you to do? What do they value most about how you do it? If you aren't sure, the honest move is to ask a few of them.
Your values. Name the few things you won't trade away to hit a goal.
Your principles. The working rules you want everyone to act on. If you already keep a list, bring it. If you don't, start one here.
A note on principles. Ray Dalio makes a useful distinction in Principles. Values are what you consider important. Principles connect those values to what you do. They're guidelines you can apply again and again when a similar situation comes up.
He built Bridgewater Associates this way. In 1982, a bad call cost him nearly everything. He let almost all of his people go and borrowed $4,000 from his father to pay the bills. After that, he started writing down the reasoning behind each decision, so he could go back and see what held up. Over the years those notes grew into a list of about 200 principles that the whole firm worked from.
A handful is plenty to start. Write down the ones you already work by, in plain words. Add one when you learn something the hard way. Revisit the list when you revisit your goals, and change what no longer holds.
2. What's already working
Everyone lists the tools, habits, people, and small fixes that carry the business now. What would each person fight to keep? Aim for ten or more.
Mark anything that only works because one person holds it in their head. That's an asset and a risk.
This list does real work later. It's where "what must not get worse" comes from.
3. Where you want to be
Ask: "It's a year from today. What's true that isn't true now?"
Everyone writes five to ten statements alone, then you compare. Describe where you've ended up, not how you got there.
This is a first pass. You'll build on this picture in the Roadmap step, where you map everything that has to change to reach it.
4. Your goals
From that picture, choose what matters most right now.
How many goals? Different frameworks set different maximums. Lencioni's thematic goal is a single rallying goal for the whole team. The 4 Disciplines of Execution says one or two. OKRs usually run three to five objectives. EOS runs three to seven priorities a quarter. We recommend an absolute maximum of seven per quarter. Every organization and team is different, so look into these when you decide how to frame yours. They're all in the reading list at the end of this step. If you have twenty goals, that's a problem.
Whatever number you land on, be able to say which one matters most.
They don't have to be financial. Less stress, a lighter workload, being ready to sell, and a better place to work all count.
Then, for each goal:
Write three to five things that must be true for it to happen.
Look back at your what's-working list and write down what must not get worse while you chase it.
5. How you'll measure them
What gets measured gets managed. So each goal needs a number, or a small group of numbers. This is harder than it sounds, and it's fine if it takes a while. Here's how to find one:
Ask what you'd see. "If this goal were going well, what would we see more of, or less of? What would a customer or an employee notice?"
Then count it, time it, or score it. Count things, like jobs finished, callbacks, or complaints. Time things, like days to send a quote or hours of overtime. Score things you can't count, with a simple 1 to 5 question asked on a regular schedule.
Scoring is how you measure the human goals. If the goal is less stress, ask everyone to rate their week from 1 to 5 every Friday, and track weekends worked. It isn't precise, but it's far better than guessing.
Once you have a candidate, you're in one of three places:
You have the number. Write the goal as "from X to Y by when."
You have a stand-in. Use it, mark it as a stand-in, and write down what the real number would be. "Hours saved times salary" is a stand-in. So is "people logged into the tool."
You have nothing yet. That's okay. The goal stays in words for now, and your first task is to measure a baseline for about a month.
Then add one to three numbers your team can influence week to week, because the main number usually moves slowly.
Give each number an owner. One person, by name, who'll still be watching it in three months.
6. Check again
Run the five-minute check a second time with everyone who took part. Then take the result to the people who weren't in the room, and let them push back. Change it if they're right.
This step is done when there's broad agreement and understanding, not when the worksheet is full.
Worksheets
Give each of these a full page, with real room to write. Here's what goes on each one.
| Page | Who writes on it | What's on it |
|---|---|---|
| The five-minute check | Each person, at the start and again at the end | The four questions, with space for a few sentences each |
| Why we exist, who we serve, and what we value | The group | The mission, the customer, what they hire us to do, and our values |
| What's already working | Each person, then the group | Ten numbered lines, and room for more |
| A year from today | Each person, then the group | Ten numbered lines |
| Our goals | The group | Each goal, what must be true for it, which one matters most, and what must not get worse |
| Our numbers | The group | For each goal: the number, what it is today, what we want it to be and by when, who owns it, and whether it's real, a stand-in, or not measured yet |
The five-minute check
Each person, at the start and again at the end
The four questions, with space for a few sentences each
Why we exist, who we serve, and what we value
The group
The mission, the customer, what they hire us to do, and our values
What's already working
Each person, then the group
Ten numbered lines, and room for more
A year from today
Each person, then the group
Ten numbered lines
Our goals
The group
Each goal, what must be true for it, which one matters most, and what must not get worse
| Goal | What must be true for it | Matters most | What must not get worse |
|---|---|---|---|
Our numbers
The group
For each goal: the number, what it is today, what we want it to be and by when, who owns it, and whether it's real, a stand-in, or not measured yet
| Goal | The number | Today | What we want it to be, and by when | Who owns it | Real, a stand-in, or not measured yet |
|---|---|---|---|---|---|
If you're doing this remotely
The five-minute check works well on a call. Everyone types their answers privately, holds them, and posts at the same moment, so nobody copies. Keep the goals and their numbers in a shared document or on your board, where everyone can see them.
Check yourself
Could three people on the team state the goals the same way, without looking?
Is each number a real measure, or a stand-in?
Are the goals something people care about, or something written for a slide?
If you kept your own framework, did you still run the check?
Watch out for
The goal that's really a solution. "Implement AI across the company" is not a goal. Ask what it's for, and keep asking until you hit something a customer or an employee would notice.
"We already have goals." Maybe so. Run the check anyway. It takes five minutes, and it's the only way to know whether everyone has the same ones.
Skipping what works. It can feel soft, but it protects good things from being "transformed" away, and it makes the list of problems that follows more honest.
The number that takes over. A single number, chased hard, crowds out everything else. That's what the "must not get worse" list is for.
Working alone?
Answer the four questions yourself. Then ask two people who know your work what they think your goals are. The gap between your answer and theirs is worth an afternoon of thought.
Where this comes from, and where to go deeper
None of this started with us. Our process is informed by the collective wisdom of a lot of people who have thought hard about goals. If this is a part of the work you're excited about and want to research further, these are some of the best business goal-setting frameworks in existence:
The Five Most Important Questions, Peter Drucker. Mission, customer, and what the customer values.
EOS, from Traction by Gino Wickman. A two-page vision and quarterly priorities.
The 4 Disciplines of Execution, McChesney, Covey, and Huling. One or two goals, a scoreboard, and weekly follow-through.
OKRs, from Measure What Matters by John Doerr. Goals in words, paired with measurable results.
The thematic goal, from The Advantage by Patrick Lencioni. One rallying cry for the whole team.
The Goal Tree, from The Logical Thinking Process by H. William Dettmer. The goal, and everything that must be true to reach it.
Principles, Ray Dalio. Write down the principles you work by, and keep adding to the list.
The 7 Habits of Highly Effective People, Stephen Covey. Begin with the end in mind.
Hoshin Kanri. Setting goals with every level of the organization.
V2MOM, Marc Benioff. Vision, values, methods, obstacles, and measures.
Jobs to Be Done, from Competing Against Luck by Clayton Christensen. What your customers really hire you for.
Appreciative Inquiry, David Cooperrider. Starting from what works.
Step 2: Ground
See the work as it actually happens, and check the data.
Why it matters
You can't find what's holding a process back if you've only heard it described by the person who designed it. Every business runs two versions of each process: the one on the org chart, and the one people really do at 4:45 on a Friday. This step is about the second one.
It has five parts. Do them over a week or so, in any order.
Part 1: Listen
Have short, private conversations with five to eight people across levels, front line included. Twenty to thirty minutes each. Promise that what they say won't be pinned on them, and keep the promise. Check back with each person a couple of days later. People often remember a step they forgot, or an exception that turns out to be common.
Good questions:
What's the most frustrating part of your week?
Where does work pile up, or wait on someone?
What workaround do you use that nobody officially knows about?
If the amount of work doubled next quarter, what would break first?
Which numbers don't you trust until you've checked them yourself?
Where does your information come from, and where do you send it?
What do you do that you suspect nobody needs?
What do you rely on every day that you'd fight to keep?
What have we tried before that didn't stick? What happened?
What are you worried AI will do here?
Is there anything you're excited about trying AI for?
Are you already using AI for anything you think is working really well? (Nobody's in trouble. Write these names down too.)
If one part of your job took half the time, which part would make the biggest difference?
Who do you go to when you're stuck with a tool? (Write these names down. You'll want them in the People step.)
Mostly, be quiet and take notes. You'll be tempted to explain why things are the way they are. Resist that urge. This is about collecting what is actually happening, and the more you try to interject here, the more likely people will filter their feedback more than you want.
Pay special attention to that last AI question. We've found that the people using AI best in a business are often more or less invisible to everyone else, including leadership. Get a good picture of how people are already using it, especially where they think they're doing something interesting. It's some of the most useful information you'll collect.
Part 2: Follow the work
Pick at least one real item, like an order, a quote, a new hire, or a support request. Walk it through every person who touches it, start to finish, with those people.
How many you walk depends on the time you have. Do it at least once, and start with whatever feels most important. But this isn't a one-time exercise. The more you get in the habit of doing it across all your workflows, the more likely you are to uncover the insights that matter.
For each step, write down who does it, where the information lives, how long it takes, how long it waits, and what goes wrong.
| Step | Who does it | Where the info lives | Time it takes | Time it waits | What goes wrong |
|---|---|---|---|---|---|
Then ask the question that finds the good stuff: "What happens when something weird comes in?"
Where the item waits is usually where the problems are. Pay special attention to the handoffs between people and teams.
Make a game of it. Before you start, have everyone write down a guess: how many steps, how many handoffs, and how many days from start to finish. Then walk it and see who came closest. If the leader's guess is far from the real answer, that gap is worth talking about.
You can also hand one person the item itself, a real order or a printed ticket, and have them carry it from desk to desk while someone keeps time.
Part 3: The ten most boring tasks
Everyone lists the ten most boring things they do each week, with a rough guess at the hours. It's funny, it's honest, and it shows where the time is hiding.
Don't skip this one. It's the shortest exercise in the playbook, and it does far more than it looks like it should.
Why it matters
It gets people talking. Nobody is nervous about admitting which parts of their job are boring. It's the lowest-stakes honest thing you can ask. People who would stay quiet about bigger problems will happily tell you about the report they hate.
It builds buy-in. You're asking people what they'd like off their plate, not telling them what's about to change. When a fix later comes from someone's own list, it's their fix.
It takes the fear down a notch. "What's boring?" is a very different question from "What could a machine do instead of you?" It tells people the point is to give them better work.
It shows you things you didn't know. Expect surprises: tasks that were supposed to have ended years ago, reports nobody reads, and the same information typed into three places.
It points you at the right work. Boring, repeated tasks are the first place to look for something to delete, simplify, or automate.
How to run it
Write alone. Give everyone ten minutes and a sheet with ten lines. For each task, write what it is and a rough guess at the hours it takes each week.
Read out the top three. Go around the room. Laughing is allowed. If someone's list runs to fourteen, let them. They've been waiting to be asked.
Put them all on the wall. Group the ones that match. When five people list the same task, you've found something.
Add up the hours. It'll be rough. It's still the first real number you have for where the time goes.
Ask two questions about each group. Who uses the result of this? What would happen if we stopped?
You are not promising to automate these. Some will turn out to matter a great deal, and some will turn out to be things nobody needed. Both are worth knowing.
What to do with the lists
Keep them. They come back three times. They go on the wall when you collect everything that's going wrong. They help you decide what to delete, simplify, and automate when you build the roadmap. And they're a baseline: ask again in six months and see what's left.
Part 4: Check the data
Pull the numbers that matter. Start with the numbers for your goals from the Goal step, then two or three that feed each one. Check them against what people assume. Teams are regularly surprised by what a thing really costs or how long it really takes.
List your whole technology stack. Every piece of software, subscription, AI tool, pilot, and project, with what it costs. Not only the AI. You'll check this list against your constraints at the end of Aim.
Why everything, and not only the AI tools? Because "AI" is a catch-all term. What it really means is more intelligence built into your software, and that can dramatically change what you need in order to do good work.
A lot of the most expensive business software exists to make it simpler for people to work with a database. That happens to be something AI does very well. So some of the biggest savings come from changing what your technology stack needs to look like, while making it easier and more enjoyable to use than it is today.
Plenty of what's on your list is used every day and working to some degree. That doesn't put it beyond questioning. Take an honest look at all of it.
| Software, tool, or project | Cost per month | Who uses it, and how much | What it's for, and how well it does it |
|---|---|---|---|
Be honest in the last two columns. Some lines will be used every day and doing their job well. Some won't. Both are normal, and the point is to know which is which.
Part 5: Find where the information lives
Notice where information lives. One source of truth, or scattered across inboxes, spreadsheets, and people's heads? If it only lives in someone's head, no tool can use it, and neither can a new hire.
This is one of the biggest pain points in most businesses. Information ends up spread across a dozen places, or held by one person. When one person holds it, they become a bottleneck of their own. Work waits for them. It stops when they're on vacation. And it leaves when they do.
It isn't always a problem. Sometimes knowledge sits with one person because they're the expert, they're the right home for it, and they like it that way. That's fine. What you're hunting for is the other kind: information that sits with one person, or in six places, because nobody ever organized it. This part helps you tell the two apart.
An exercise: "Ask Jim"
This one is optional. We'd do it anyway. It takes about twenty minutes, it's fun, and it's usually enlightening.
"Jim" is a placeholder for wherever people go to find something out. Jim might be one person. Jim might be a different person for every process. Jim might be a shared drive, a system, or a sticky note on the monitor by the door. Sometimes there is no Jim at all. The point is to find out where people turn, and whether they all turn to the same place.
Write the questions. Put ten to fifteen everyday questions on cards. Use real ones. Where's the current price list? What did we promise this customer? How do I process a refund? Who approved this? Where's the login for the supplier portal?
Ask fast. Read a card aloud. Everyone writes down, in five seconds, where they'd go. The answer can be a person, a place, a system, a document, or "no idea."
Reveal together. Everyone shows their answer at once.
Read the pattern. There's no right result. Each one tells you something different.
Everyone named the same place. A system, a folder, a document. Good. That's a real source of truth.
Everyone named the same person. That person is carrying it. Find out whether they want to.
A different person for each question, and everyone agrees who. That can be healthy. Each area has its expert. Check that each of them is happy with it, and that there's a backup.
Everyone named somewhere different for the same question. Nobody really knows where it lives.
"No idea." Write that down too.
Tally the answers. Count the people and the places. You may find one person carrying far more than anyone realized. You may find it's spread evenly. You may find nobody is carrying it at all.
This exercise tends to bring three things into the open: people who are shouldering work they never asked for and don't want, plain disorganization, and experts who are exactly where they should be. All three are worth knowing about.
Whatever you find, aim for one thing by the end. For every question on those cards, everyone should know where to turn. You don't need a new system for that, as long as the answer is clear.
A good rule to aim for is one home for each piece of information. It gets created and updated in one place, and everyone else reads it from there. When the same thing can be changed in two places, the copies drift apart, and people stop trusting both.
Draw the map
Turn what you found into a simple map of who and what holds the information the business runs on. One line for each thing people need to know:
| What people need to know | Where it lives, or who holds it | Why it lives there | Okay, or a risk? |
|---|---|---|---|
| Example: what we've promised each customer | The account manager's inbox and memory | It's always been that way | A risk. Nobody else can answer when she's out. |
Talk to the people whose names came up
If some names came up a lot, talk to those people privately, and be kind about it. Often they never asked for the role. They just happen to be the one who knows. The person everybody asks is usually the person who has been holding things together. Ask:
Why does this live with you?
Do you like it living with you?
What happens when you're out?
What would it take for someone else to find it without asking you?
Then sort each item into one of three:
Keep it. They're well suited to it, they like it, and the business would cope for a week without them.
Back it up. It belongs with them, but it's a risk. Write it down, or train a second person.
Give it a home. It's with them by accident. Move it somewhere everyone can find it.
The goal is to leave knowledge with the experts, and to clear up the disorganization around them.
If you're doing this remotely
Listening conversations work fine by video, one person at a time. To follow the work, have the person who does it share their screen and walk through a real item while you take notes. The boring-tasks lists and the "Ask Jim" answers can go into a shared document or onto your board. Have everyone hold their answers and post at the same moment.
Check yourself
Did you hear from people across the business, in private, and mostly listen?
Did you follow at least one real piece of work from start to finish, with the people who do it?
Does everyone have a boring-tasks list?
Do you have today's numbers for your goals, and a list of your whole technology stack?
Do you know where the important information lives, and who holds it?
Does the way the work was described to you match what you saw when you followed it? Where it matches, good. That's worth knowing. Where it differs, write down the difference without deciding who's right.
Is a spreadsheet doing a core job next to a system you already pay for? That can cut either way. Maybe the spreadsheet needs a better home. Or maybe the spreadsheet is enough, and it could replace the system.
On your technology list, do any tools do the same job? Do they share information, or does someone carry it between them?
Did the boring-tasks lists show anyone building the same report by hand every week?
When something went wrong in the work you followed, where did it start? It's often in what comes in: free-text boxes, missing fields, and three date formats.
Do you know who is already using AI on their own, and what's working for them? It shows you where the real demand is, so there's no reason to stamp it out.
After the "Ask Jim" exercise, does everyone know where to turn for each question on the cards?
A note on data
You don't have to get all your data in order before you start. You only have to fix the data your first change touches. "We need to clean up our data first" has delayed more good work than any technology problem. It's part of the journey, not the toll booth.
Most data gets cleaned up by fixing the process that produces it. So think of it as two passes.
Now: clean up what you can with the information you already have. It won't be perfect, but it will be enough to see the outline of what's going on.
As you go: some data can't be cleaned up, because it was never collected properly in the first place. You can't see what nobody wrote down. That takes a change to the process, so the information becomes visible from here on.
Those changes are usually small. Tag every invoice with the kind of work it was for. Record where each new lead came from. Write down the date a job was promised as well as the date it was finished. A simple tag added today is a number you can trust in three months.
So as you work through this step, keep a running list of what you'd need to track to have the visibility you want. That list goes on the roadmap.
If you'd like help working out what to track and how, that's a conversation we're glad to have.
Working alone?
When you're the one doing all the work, you can't watch it from the outside. Most of it lives in your head. So the job here is to get it onto paper, then look at it as if it belonged to someone else.
Pick one kind of work you do often. A quote, a client project, or a monthly invoice run.
Write the steps from memory. Every step, from the first request to done. Don't tidy it up.
Keep a log the next time you do it for real. Each time you touch it, note what you did, how long it took, and what you were waiting on before you could do it. A note on your phone is enough.
Compare the two. You're looking for the steps you forgot, the waits you didn't know were there, and the things you do twice.
Ask what a team would ask. Where does it wait? Where do I redo things? Which steps exist only because of how I set things up years ago?
Then do the other parts on yourself: your own boring-tasks list, your whole technology list, and where your information lives. If the answer to that last one is "in my head," write that down.
If you're comfortable with it, one outside view helps. This is optional. Pick a customer or collaborator you get along with and ask one easy question: "I'm tidying up how I work. Is there any part of working with me that takes more effort on your end than it should?"
Step 3: The Wall
Every problem you can think of, out of heads and onto the wall, where everyone can see it, sort it, and add to it.
Why it's called the Wall
This step is called the Wall because that's what you do. You put every problem you can think of up on a wall. Then, in the next step, you map how those problems connect.
We prefer sticky notes and a real wall. If you're doing this remotely, a digital sticky-note board like Miro can work too. We like the real thing because it's more flexible, people deal with each other more directly, and the canvas is much, much bigger. Ideally you have a big blank wall that sticky notes actually stick to. Test that first. Some walls have opinions.
Why it matters
To us, this step and the next one are the most important thing in the whole playbook. So we're going to slow down.
One of the biggest mistakes people make is attacking symptoms. They see a problem. They know it's a problem. They feel it in their gut. So they assume that if they solve it, they're making progress.
The trouble is that it's hard to tell by feel which problems are root causes and which are symptoms. Symptoms sit downstream of root causes. And when you map them, you often find loops: a few problems that feed each other, round and round, throwing off new problems the whole time.
If you spend your effort on the problems coming out of a loop like that, you're wasting time, energy, and money, because the loop keeps printing. You have a factory that produces problems. Solving them one at a time is inefficient, and it wears people down.
Break the loop, or fix the root cause underneath, and something different happens. Instead of working down a list of a hundred problems, you address one thing that wipes out twenty of them in one go.
None of that can start until the problems are out where everyone can see them. Right now they live in people's heads. Everyone on your team is carrying a private list of what's broken. They mention items from it at lunch, in the parking lot, and to their spouse. They rarely mention them in the meeting where something could be done. The wall puts all of those lists in one place.
How to run it
This works in rounds, and you can go back to an earlier round at any time.
One rule holds from start to finish: anyone can add a sticky note at any time, without asking. And anyone can flag something on the wall that looks wrong to them.
Round 1: Brainstorm
Put your goals at the top of the wall. Everything that goes up is measured against them.
Everyone writes. Give each person a stack of sticky notes and a marker. One problem per note, as many as they can fill. Ten to fifteen quiet minutes.
Don't aim for the perfect note. That comes later. Vague is fine. Half-formed is fine. A solution in disguise is fine for now. The goal of this round is to get ideas out of heads and onto notes.
Add what you learned in Ground. Whoever did the listening adds notes from those conversations, without names. Add what went wrong in the work you followed, too.
Round 2: Put them up and refine them
Everything goes on the wall.
Go through the notes one at a time and talk about each. The aim is to make the note better. Nobody has to defend it, and you don't need to debate whether it's true.
Make it more specific. What happens, to whom, and how often?
Make sure it isn't a solution. When a note says "We need a CRM," ask why. What's happening that makes us say that? Keep asking until you reach the problem underneath.
Replace notes freely. If the group finds better wording, write a new note and swap it in.
Add what it triggers. Talking about one note will remind people of others. They write them and stick them nearby. No permission needed. Expect the wall to grow in this round.
Here's what refining does to a note:
| Before | After |
|---|---|
| We need a CRM. | We lose track of leads after the first call. |
| Sales overpromises. | A third of jobs start with a scope the crew didn't agree to. |
| Nobody uses the AI tool. | We pay for 40 seats and 4 people log in. |
| Communication is bad. | Customers hear about delays from us only after they call. |
A refined note is something you could go and observe. It describes what's happening, not what to buy and not who to blame. If you can add a number, add it. A guess is fine. Mark it as a guess.
Make sure these get on the wall
People hold back on a few topics. Ask for them by name:
Fears about AI. "Is this going to replace me?" "Is this another thing being done to us?"
What past rollouts left behind. Tools nobody uses, workarounds that came back, the project nobody mentions.
Things leadership does. If the wall has nothing about the people in charge, the room isn't safe yet. The leader can go first with one about themselves. It works wonders.
People leaving, or checking out. If good people have left or gone quiet, that goes up too.
Round 3: Sort
From here there are two ways to go. If the wall is small enough to take in, you can move straight to mapping in the next step. If you have a lot of notes, sort them first. This is a way we've run it in person that works well:
Break into small groups. The groups take turns at the wall, moving notes that seem related next to each other.
Name the groups as they form. They can be by team, by process, by customer, or by whatever fits. There isn't one right way.
Put each note where it connects most strongly.
Treat disagreement as a good sign. If people disagree about where a note belongs, don't fight over it. Write the same thing on a second note and put it in both places, or in every place it belongs.
If the wall is very large, you can also mark the heavy ones. Give everyone five dots to put on the notes that hurt your goals most. Bosses dot last. The dots tell you where to start mapping. They don't tell you what the root causes are.
If you're doing this remotely
The rounds are the same. The wall is a shared digital board, like Miro.
Brainstorm alone first. Make sure everyone takes at least ten minutes to list problems before anyone discusses them. They can add sticky notes straight onto the board, or type a bulleted list and send it in.
Get every item onto its own note. If people sent lists, one person puts each line on a note. An AI agent that can work in your board can do this for you.
Refine and sort together on a call, the same way you would in the room. Anyone can still add a note at any time.
Check yourself
Did notes come from every part of the team, not only the people in charge?
Did everyone get notes down before the talking started?
After refining, is each note something you could observe?
Did the wall grow while you refined it? If nobody added a note, ask what's missing.
Is there anything everyone knows and nobody wrote? Ask. Then wait. The silence is doing work.
Watch out for
Solving too early. Someone will try to fix the first note. Thank them, and write "later" on a flip chart. You'll get to solutions. Fixing now means you fix whatever came up first, not whatever matters most.
Polishing too early. Chasing the perfect note in round 1 slows everyone down and loses ideas. Rough first. Refine second.
The tidy wall. A wall with twelve polite notes isn't finished. A big, messy wall is a good sign, and a hundred notes or more is fine.
Working alone?
Set a timer for twenty minutes and write everything that's going wrong, one per note. Don't polish. Then go through them one at a time and refine each one, the same way a group would.
If you're comfortable with it, ask one or two people you trust: "What's one thing about working with me, or my business, that you'd fix?" Add their answers. This part is optional.
Step 4: Root causes
Map how the problems connect, and find what's really holding you back.
Why it matters
Your wall may have forty problems on it. It may have hundreds. Either way, they aren't all equal. A few are root causes. Most are symptoms, sitting downstream of something else. And some are caught in loops that keep producing new problems.
This step is how you tell which is which. It's the difference between solving problems one at a time, forever, and fixing the few things that make most of them go away.
There's a second payoff, and it's about your people. When the problems are one big mess, it's easy for each person to believe their problem is the most important one, or that it's being ignored. Once every problem is on the wall and connected to the others, people can see that theirs is there, and where it sits.
That changes how disagreement works. It gets easier to say out loud, easier to settle, and easier to let go of. Mapping is one of the things that helps most with getting buy-in, because the map does much of the persuading a leader would otherwise have to do.
How the mapping works
The tool is simple. You connect problems using the words "if" and "then." It comes from the theory of constraints, and people have been using it on problems like yours for decades.
If every quote needs the owner's sign-off, then customers wait four days for a quote.
The cause goes underneath. The effect goes above it. An arrow points up from the cause to the effect. As you add more, the wall takes on levels:
At the top: the things people feel. These are the effects your customers and your team notice.
In the middle: causes that are themselves caused by something else.
At the bottom: root causes. Nothing else on the wall causes them, and they're within your control or influence.
Loops: places where following the arrows brings you back to where you started.
How to run it
We'd encourage you to try this on paper, with sticky notes on a real wall. Doing it on a digital board is totally fine too.
Pick a place to start. Use one of the groups from your sort, or the notes with the most dots. You'll work across the rest of the wall from there.
Ask why. Take one note and ask, "Why does this happen?" Look for the answer on the wall first. If it isn't there, write a new note. Put the cause underneath, with an arrow pointing up.
Say it out loud. Read every link to the room as "if, then." "If quotes need two sign-offs, then customers wait four days." Do this for every single link. It's the most important habit in this step.
Check for what's missing. After each link, ask: "Is that true? Or does something else need to be said to make it true?" Often there's an assumption hiding in the middle. "If we get more leads, then we close more sales" is only true if someone has time to follow them up. Write the missing piece on a note and add it. This is where the blind spots come out.
Keep going down. Ask why again. Several levels is normal.
Connect across. When one note explains effects in different groups, draw an arrow from it to each one.
Look for loops. If the arrows lead back around, so that A causes B, B causes C, and C makes A worse, mark it clearly. A loop is a problem factory.
Stop when you reach something you can change. A root cause is the deepest cause that's still within your control or influence. "The economy" is not a root cause. "We quote every job from scratch" is.
Count what sits above. For each root cause and each loop, count the notes it feeds. The ones with the most above them are where one fix removes the most problems. Hold each one up and ask the room, "Is that accurate?"
This is iterative. Go back and add notes, re-sort, or reword at any point. Anyone can flag a link they think is wrong, and anyone can add a note, at any time.
Save the map, and let AI help
However you built it, the finished map should end up on a digital board like Miro, or a similar tool. The wall is a great place for it while the work is active, so leave it up as long as you can. A board is its long-term home. It can be kept, shared, commented on, and updated, and everything you do in the later steps builds on it. The specific tool matters less than picking one your people will really open.
If you worked on a real wall, here's how to move it over:
Take photos before anyone touches it. Clear, well-lit, and covering every part of the wall.
Have an AI rebuild it. Send the photos to an AI agent and ask it to recreate the notes and the arrows on your board. As long as the handwriting isn't completely terrible, it's pretty good at this.
Give one person the job of checking it. Every note and every arrow, against the wall or the photos. Their job is to make sure it's an accurate copy and that nothing got lost in translation.
If you worked remotely, the map is already on the board, and the same person checks that it says what the group agreed.
Once the map is digital, an AI can be very good help with it. Explain how the process works, give it your own problems, links, and thinking, and ask it to flag where the logic is weak. That means links that aren't really true, places where you've assumed something without writing it down, and links that cover too much ground, where one or more notes belong in between. It can also point out causes you may have missed and loops you didn't spot. This is worth doing whether you mapped as a team or on your own. It's useful here because it's working from your material and not making things up about a business it has never seen. You still decide what's true.
Two small examples
Both are invented, to show the shape.
One cause, three effects.
Effect: Customers wait four days for a quote.
Effect: Two salespeople have quit this year.
Effect: The new quoting tool is barely used.
Cause of all three: Every quote needs the owner's sign-off, and the owner is on job sites all day.
Why: Two years ago a bad quote lost money, and the rule was never revisited.
No software fixes that. A new rule does: quotes under a set amount don't need sign-off. Total cost: one awkward conversation.
A loop.
If quotes go out late, then we lose jobs to faster competitors.
If we lose jobs, then money gets tight.
If money is tight, then we can't hire help in the office.
If there's no help in the office, then the owner writes quotes at night.
If the owner writes quotes at night, then quotes go out late.
Every pass around makes it worse, and each link throws off problems of its own. Fixing any one of those side problems leaves the loop running. Breaking one link stops it.
Find the constraint
Now test each root cause and loop you marked against your goals. A constraint is something that limits how much the whole system can do right now. You're looking for the one that limits it most.
If this improved tomorrow, would your goal numbers move? If not, it's a real problem, but it isn't what's holding you back most. Park it.
Where does work pile up? Look back at the work you followed in Ground. The pile sits in front of the constraint.
Is it a rule, not a resource? Constraints are often a policy, an approval, or a measure that rewards the wrong thing. Those are the cheapest to fix and the hardest to see.
Is it trust? If people don't believe the change is for them, no tool will work until that's dealt with. It moves to the front of the plan.
Be ready for a surprise. The constraint may be somewhere you didn't set out to look. Sometimes it's the offer itself: what you sell, who it's for, or how you describe it. If a side quest turns out to be the main quest, that's okay. Go back to the Goal step and update your goals with what you know now.
You may find more than one. It can be hard to tell which constraint limits you most. That's normal. Choose the one your team has the best reasons for, and treat it as a working answer. The Check step will tell you whether you were right. If two or three look equally important and each can have its own owner, you can work on them together.
Some constraints show up almost everywhere. The time people lose to notes, records, and keeping track of projects is one of them. We covered it in "Where we often start," and you can begin on it whenever you're ready.
Before you spend anything on the constraint, ask two more questions:
Are we wasting it? Is the constraint idle, interrupted, fed bad inputs, or doing work someone else could do? Fixing that is usually free.
Is everything else lined up behind it? If other parts of the business produce more than the constraint can absorb, you get piles, rushing, and firefighting. Slow them down to match.
After those two, ask whether to add something: a hire, a tool, or an AI agent. How to make that choice is covered in the Roadmap step.
The technology check
This is the part that pays for the time you've spent. Take the technology list from Ground. For each line, ask one question: is this aimed at a constraint you found?
| Software, tool, or project | Aimed at a constraint? | Keep, fix, or cut |
|---|---|---|
| Yes / No / Don't know | ||
| Yes / No / Don't know | ||
| Yes / No / Don't know |
Yes: keep it, and make sure it's working.
No: it may still be worth having. Be clear about what you expect from it, because it's less likely to move your goal numbers.
Don't know: that's an answer too, and a common one.
If most of your spending lands on "no," you're in large company. That's the whole reason this playbook exists.
Worksheet
| Your answer | |
|---|---|
| The root causes and loops with the most problems above them | |
| The constraint: what limits your goals most right now. List more than one if you need to. | |
| How we're wasting it today | |
| What needs to slow down or line up behind it | |
| What we'd add only after that |
Check yourself
Was every link said out loud as "if, then"?
For each link, did someone ask whether anything was missing in the middle?
Did you look for loops?
Would the people who do the work agree with the map? Have you asked them?
Is each root cause within your control or influence?
Did you land on the cause you expected before you started? Be a little suspicious. Check it twice.
Has a person, an AI, or both checked the map for weak links, unstated assumptions, and missing steps?
Is the map saved somewhere people will look?
Watch out for
Skipping the out-loud part. A silent arrow hides assumptions. Saying the link is what exposes them.
Chasing the loudest symptom. The note with the most dots is often a symptom. Dots show where it hurts, not where it starts.
Blame. "Dave" is not a root cause. Ask what about the system makes a reasonable person act that way.
The shopping reflex. When everyone wants to buy something, ask the constraint question again. A purchase can be the right move. Check that it's aimed at something that's holding you back.
Working alone?
Do the same thing on your own. We'd encourage you to try it on paper, with sticky notes on a wall or index cards on a table. Doing it on a digital board is totally fine too. Then walk one trusted person through your map, link by link, out loud. You'll hear the weak links yourself before they say a word. Then have an AI check the map too, as described under "Save the map, and let AI help." The ideal is to ask both.
Want a hand with these two steps?
The Wall and the mapping are where an outside facilitator helps most. People put franker notes on the wall when the boss isn't the one running the session, and a neutral person can keep asking the obvious questions. We run both with teams, in a single day or as part of a longer engagement. See "When to bring someone in" at the end.
Strike: steps 5 to 9
Build a 6-month roadmap, and act on it with your people.
Step 5: Roadmap
Build an action plan that maps what needs to be done to go from your current reality to the way you want things to be.
Why it matters
You now know what's really in the way. That is more than most businesses know before they spend money, so take a moment. Then get moving, because a map nobody acts on is wall art.
The mistake here is jumping straight to a task list. Tasks feel productive. But a list of tasks written before you know what has to be true is how you end up busy and in the same place.
The rule: conditions first, then tasks
Map what must be true before you map what to do. Then translate each condition into tasks.
How to run it
It would be easy to read the list below, nod, and move on. Please don't. Reading it does nothing. The value comes from actually doing each step, in order, in writing, with your people. Give it real time.
Describe the way you want things to be. Start with the picture of a year from now that you wrote in the Goal step. Then take your root causes, your loops, and the worst effects on the wall, and flip them. "Customers wait four days for a quote" becomes "Customers get a quote the same day." Write three to five of these. Be specific enough that you'd know it if you saw it.
List what stands in the way. For each one, ask, "What stops this from being true today?" Write every obstacle down. Let the skeptics lead this part. They've been waiting all week.
Turn each obstacle into a condition. An obstacle says what's wrong. A condition says what must be true instead. "The owner approves every quote" becomes "Salespeople can approve quotes under an agreed amount."
Put the conditions in order. Which must be true before another can be? Draw arrows. You now have a path.
Translate conditions into tasks. For each condition, what actions make it true? Now you may write tasks.
Sort the steps of the work itself. For each workflow you plan to change, put every step on a card and sort each into one of four piles: delete it, handle it with simple code or a rule, hand it to an AI agent, or keep it with a person. Do the piles in that order. Most teams are surprised by how big the first pile gets.
Give every task an owner and a date. One named person each. Not a team, not "IT," not "everyone."
Choose the first change. It can be the one that matters most to your goals, or a quick win to build momentum. How to choose is in the Deploy step.
When it feels impossible
Some of what you write here will sound impossible the first time you say it out loud. "Customers get a quote the same day." "Salespeople approve their own quotes." Somebody will laugh.
Write it anyway. It doesn't have to feel possible yet. Your job at this point is to state it clearly, not to believe it.
A lot of what looks out of reach only looks that way because of assumptions about how things have to work. The way through tends to appear once you start questioning them.
That's what these steps are for. Doing the work of restating an obstacle as a condition often unlocks a way of looking at the problem that you couldn't see before, and didn't know was possible.
How far out to plan
Plan six months in outline and the first 90 days in detail. Beyond 90 days, a detailed plan is mostly guesswork. You'll redo it in the Adapt step anyway.
If you already work in quarters, with OKRs or quarterly priorities, use them here. The first 90 days of your roadmap becomes next quarter's set. Each condition you're working toward is an objective, and its number from the Goal step is the result you measure.
If you don't work that way yet, this is a good moment to start. A goal in plain words, paired with a result you can measure and checked every quarter, is a simple habit, and it's the one nearly every goal-setting framework agrees on.
Worksheet
| The way we want things to be | What must be true (conditions) | Tasks | Owner | By when |
|---|---|---|---|---|
The first change:
| Your answer | |
|---|---|
| What we'll change first | |
| The workflow it touches | |
| Who owns it | |
| The number it should move, and what that number is today | |
| When we'll check |
Where AI fits, and how to choose tools
This is the part most people wanted to start with. You can do it well now, because you know which problem you're solving, what's holding you back, and which number should move.
The forms AI comes in, what it's good at, and where the risks are were covered in "What we mean by AI," before the steps. This section is about matching it to your own work.
Where AI tends to fit
When you sorted the steps of the work, one pile was "hand it to an AI." A step is a good candidate when:
It's mostly reading, writing, summarizing, sorting, or searching.
What comes in is messy language, like emails, notes, calls, and documents, more than clean numbers.
A person can check the result quickly.
A mistake is cheap to catch and fix.
It's a poor candidate, for now, when:
It has to be exactly right every time, like arithmetic or record-keeping. Use a rule or simple code for that.
It commits money, makes a promise to a customer, or touches safety or the law with nobody reviewing it.
Nobody could tell whether the answer was wrong.
The process underneath isn't agreed yet. Fix that first.
Start with what you already have
Before you buy anything, find out what your current software and a basic AI subscription can already do. It's usually more than people expect, and it costs you nothing new to find out.
How to choose a tool
We aren't going to name tools here. Any list would be out of date before you read it. These rules hold up no matter what's on the market:
Start from the problem and the number. Look at demos after that.
Try it on your real, messy work before you pay. A tool that only works on tidy examples doesn't work.
Stay flexible. Prefer month-to-month. Be wary of a year-long contract while things are moving this fast.
Keep your information portable. Can you get your data out? Keep it in files and systems you control.
Own the process, not just the tool. Write down how the work gets done and the instructions you give the AI. Then you can move to a better tool without starting over.
Give it an owner, a number, and a review date. If the number hasn't moved by then, stop paying.
Look again every six months. What was the best choice in the spring may not be by the fall. The Adapt step is where you do this.
Where we can help
You don't have to track all of this yourself. It's a big part of what we do. At any given time, we can help you understand the current state of AI tools, where they're likely headed over the next six months to a year, where the value is for a business like yours, and how to set things up so you can adapt as the landscape changes.
Do it yourself, or get help?
For each task, be honest about what it needs. A lot of this you can do yourselves, including a lot of the AI.
Get real expertise when a task touches a system of record, customer or payment data, or anything regulated. The same goes for anything that has to run every day without someone watching it.
If you're doing this remotely
Build the roadmap on the same board as your map. Give people a few minutes alone to list what stands in the way before you discuss it, the same way you did for the Wall. Then put the conditions in order together on a call.
Check yourself
Does every task trace back to a condition, and every condition to a goal?
Is the plan aimed at the constraints you found? Or has it filled up with things people already wanted before you looked, like a tool someone had their eye on?
Can everyone find their own problems on the map, and see what they connect to?
Did everyone get to disagree before you decided? If someone still disagrees, can they point to the link on the map they'd change? That's a far easier conversation than arguing over whose problem matters most.
For each step you plan to automate or hand to AI, have you decided who checks the result? And have you marked the decisions that still need a person's sign-off, like anything involving money, a promise to a customer, or someone's job?
Could you start the first change on Monday? If it needs a purchase, did the work point you to it?
Does each change you're starting have an owner with the time to see it through?
Has a person, an AI, or both checked the roadmap for assumptions and missing steps?
Watch out for
The roadmap an AI wrote from scratch. It's tempting. Ask a model for a six-month AI roadmap and you'll get one in under a minute, with phases and everything. It will fit any company, which is the problem. Build yours from your wall. Then, by all means, get an AI to help. Once you've explained the process and given it your own problems, links, and conditions, it's very good at this. Ask it to flag steps that depend on something you haven't written down, assumptions you're making, and jumps that are too big. Keep the roadmap on the same board as your map.
Automating the work the way it's done today. Before you automate any workflow on your roadmap, write down how it's done now, delete the steps that don't need to exist, and simplify what's left. Then automate. Otherwise you carefully automate things that never needed doing.
Too many changes at once. How much you take on depends on your situation. Some changes are just a decision: you agree to do it differently, and it's done. Others are projects. If you found five root causes and five people each take one, that can work. What to guard against is getting distracted and taking on more than people can finish. Each change needs an owner with the time to see it through. Everything else goes on the roadmap.
Working alone?
Same steps, smaller table. Send your finished roadmap to one person and ask only this: "What's missing, and what's wishful?"
Then ask an AI too. Give it your map and your roadmap, explain how the process works, and ask it to flag three things: steps that depend on something you haven't written down, assumptions you're making, and jumps that cover too much ground, where a step or more belongs in between. The ideal is to ask both a person and an AI.
Step 6: People
Engage the people who will do the work, both internally and externally as needed.
Why it matters
This is where many AI efforts quietly die. Sometimes the tool is wrong, but oftentimes it's just that the people never use it.
If you've followed the playbook, you're already ahead. The people who'll do the work helped to find the problem and build the plan. This step finishes the job.
The map is a big part of why. People who can see their own problems on the wall, connected to everyone else's, don't have to fight to be heard. This is why you need to keep it where people can see it, review it, add to it, and comment on it over time. Point to it when disagreements come up.
In our experience, the wall is a great place for the map while the work is active. Leave it up for as long as the project is running, and encourage people to keep using it.
For the long term, give it a digital home as well, so people can look at it, reflect on it, comment on it, and keep building on it over time. This is where a tool like Miro can be really valuable.
One caution. If your people don't want to use Miro, it can easily become one more tool that was built with good intentions and never used. If you take the time to put the map somewhere, it has to be looked at and used. The specific tool matters less than this: store the map in a place, and in a way, where people know to look for it and will.
We also highly recommend running this whole process again on a regular basis, such as once a year, as a yearly pattern interrupt. It works well as part of a retreat. Building a retreat around this work is our favorite way to do it.
Find the people who are already excited
One of the highest-leverage moves is to find the people most excited to build with AI and put them in a position to guide the change for everyone around them. Think of them as little nuclei for your AI transformation.
You never know where you'll find them. They aren't always where you'd expect, and they're often not the most senior or the most technical. Look for:
The names that came up in Ground when you asked, "Who do you go to when you're stuck?"
People who are already using AI on their own. They're often invisible to the rest of the business, so go by what you heard in Ground, not by who talks about AI the most.
Whoever has been oddly cheerful through this entire process.
Give them time, a real role in the roadmap, and public credit. Then let them teach. People learn a new way of working faster from the colleague two desks over than from any training.
How to run it
Name an owner for every change. One person, by name, with the time, the authority, and the interest. Don't assign a change to someone just because they weren't there. Find a person who understands why it matters, wants it to happen, and has a real stake in making sure it does.
Connect each person to their own win. Everyone affected should be able to name something that gets better for them. If they can't, expect resistance, and they'll have a point.
Deal with the fear directly. Repeat what you said at the start about why you're doing this. Say plainly what happens to the time that's freed up. Vague reassurance sounds exactly like bad news. Be optimistic, and also show that you understand the risks and pitfalls. A big one is that AI, if you're not careful, simply creates more work. The more you show you understand that, the more people will trust the rest of what you say.
Bring in the people who weren't in the room. Hold at least one session to walk them through the wall, the map, and the plan, and ask what you got wrong. Let the team present it, not the boss.
Train inside the real work. Skip "AI in general." Teach each person the new way of doing their own job, with their own tools and their own messy examples.
Go first. If the leader doesn't follow the new process, nobody will. This one isn't negotiable, and everyone is watching.
When to bring in outside expertise
The change should come from your people. That doesn't mean they have to do all of it alone. Bring in subject-matter experts when the work is complex, when sensitive data is involved, or when something has to run reliably every day.
The experts support your people's plan. They don't replace it.
Worksheet
| Change | Owner | Champion who'll help others | Who's affected | What gets better for them |
|---|---|---|---|---|
Check yourself
Does every change have one named owner who wants it?
Is there a champion in every team that's affected?
Has anyone been made to feel replaced? Address it in person, today.
Is the leader using the new process themselves?
Does every owner understand why their change matters, and have a real stake in it?
Have we gone looking for resistance, and listened to what it's telling us?
Is the map somewhere people can see it, add to it, and comment on it?
Signs people are checking out
Watch for quiet non-use, old workarounds coming back, and complaints in the hallway instead of the meeting. A common one is the old spreadsheet someone keeps "just in case." It usually means they weren't shown the new way well enough, or the new way doesn't handle something they deal with. Don't treat these as bad behavior. Treat them as new notes for the wall, and find out what's causing them. Go looking for resistance like this, and lean into it. Don't tell people they need to get past it so the work can get done. The resistance is the way.
Working alone?
Your "people" are your customers, your collaborators, and you. Tell the people affected what's changing and why. Then be honest about whether you're following your own new process. You are the hardest person on your team to manage.
Step 7: Deploy
Delete what shouldn't exist. Then make the change that matters most, and finish it.
Why it matters
A plan changes nothing until one real thing is different on a regular Tuesday. Your team has seen initiatives announced before. They have not always seen them fulfilled.
So the first change does two jobs. It improves something, and it shows your people that this time the work gets finished. That's a good reason to make it visible and to see it all the way through.
This is also the step where AI gets added, if it belongs. You picked the candidates in step 5. We aren't trying to automate everything. We want to automate the parts of the work people don't like or aren't especially good at, and leave them the parts they like and do well.
Choose what goes first
There's more than one good way to choose, the same way there's more than one good way to pay down debt.
The most important fix first. Go straight at the thing that's holding you back the most. It has the biggest payoff, and it can take longer to show.
A quick win first. Finish something fast to build momentum and belief, then take on the bigger one.
Either can be right. It depends on your team, how much trust there is, and how much time people have. Choose on purpose, and tell people why.
The order of operations
Whatever you chose, work on it in this order.
Delete. Remove steps nobody needs: approvals nobody reads, reports nobody opens, the form that asks for a fax number.
Simplify, then standardize what you'll automate. Make what's left as simple as you can. Then, for the steps you plan to automate, agree on one way of doing them and write it down. A step with no agreed way of doing it can't be automated well. Not every step has to be standard.
Fix what comes in. Every piece of work starts with something arriving: a request, an order, a handoff from another team, a file, a phone call. If it arrives incomplete, unclear, or different every time, every step after it struggles, and it's easy to blame the people or the tool. Look at what's coming in first. Often the fix is small. Agree on what a complete request looks like, and ask for it that way.
Add rules or simple automation. A checklist, a template, a rule in software you already have, or a simple automation handles a lot of steps with no AI at all.
Where it fits, add AI. Go back to "What we mean by AI" and to "Where AI fits, and how to choose tools" in step 5, and pick the form that suits the step. If nothing fits, that's a fine answer.
Many changes are finished by the third of these. That's a good result.
What a good first change has
How much you take on depends on the size of your team and how much time people really have.
A clear focus. Everyone involved knows what's changing and why. It can be one workflow or several changes at once, as long as each has an owner with the time for it.
An owner. One person leading the change, and an owner for each part of it. They come from the People step.
Numbers. Today's values, written down before you touch anything.
A quick start. Started in days, even if the whole change takes longer.
Someone who can feel it. A customer, or the people doing the work, should be able to tell that something changed. Some changes show up fast, and some take time. If yours will take a while, agree on the first sign you expect to see, and roughly when.
If the change involves AI or automation
"Adding AI" can mean very different things, and the habits worth having depend on the form it takes. The forms are described in "What we mean by AI." Find yours below. None of them require an engineer to understand.
Whatever form it takes
Try it on real work first. Use last week's actual emails, orders, or records, and leave the clean demo examples aside.
Decide who checks the results. And decide in advance which decisions stay with a person.
Give it an owner. One person who knows how it's being used, and who notices when it stops helping.
If you're turning on an AI feature in software you already have
Find out what you already have. Check what's included in what you pay for today before you buy anything new.
Check what it can see. Know which of your information the feature reads, and where that information goes.
Show people how you use it. A feature nobody was shown doesn't get used. Ten minutes inside the real work beats a memo.
If you're giving people their own AI assistant
Agree on what never goes into it. Decide this before anyone logs in, based on what the plan you're paying for does with your information. A one-page AI policy is enough: which tools are approved, what information can't go in, and when a person has to review what comes out.
Don't stop at handing out logins. Teach each person to use it on their own job, with their own examples. Keep it going with a regular practice session, such as an hour a month, where people rebuild a task they already do every week.
Share what's working. Your best users are often invisible. Give them a regular place to show the rest of the team what they've figured out.
Save the good instructions. When someone works out how to get a job done well, write the instructions down, so the result doesn't depend on who's typing.
The person still owns the work. Whoever sends it is responsible for it, whatever wrote the first draft.
If you're bringing in a new AI tool
Use the rules from step 5. Try it on your own work before you pay, prefer month-to-month, and make sure you can get your information back out.
Name what it replaces. If nothing gets turned off or stopped, you've added work and cost.
If it runs on its own: an automation or an agent
These matter most here, because nobody reads each result as it happens.
Test it on the odd cases. The strange order, the half-finished request, the customer who replies to the wrong email. Those are where automations break.
Run it alongside your people first. Let it work in parallel for a couple of weeks while the team does the job the normal way. Read every case where they disagree. You'll learn more from those than from anything else.
Collect past cases your best people already judged. That's your answer key. Check the new thing against it, and check again whenever something changes.
Earn trust in steps. First it suggests. Then it acts with a person's approval. Only later, where the stakes are low, does it act on its own.
Read first, write later. Give any AI tool read access before write access, the way you would with a new hire.
Keep a person on the important decisions. Decide in advance which ones.
Make it shout when it fails. A named person hears about a failure right away. Silent failure is the expensive kind.
Write the backup plan on day one. One page: what we do when it's down.
Keep a log people can open. What it touched, and when.
If you bring in a vendor or builder
At the other end of all this, you might find that what you want to build is more than your team is ready to take on. In our experience, it's often simpler than you expected. When it isn't, or when you just want to be sure you get it right, bringing someone in is completely reasonable.
You're in a much better position to hire well now. Because you've done the work up to this point, you know what problem you're solving, what you need built, and what it should change. That makes it far easier to tell who is worth paying, and what you need them to do.
These are the kinds of things to ask before you sign anything:
Can we see something you built that's running for another customer today? Can we call them?
What will it cost to run each month, in writing?
Where does our data go, who can see it, and what does the model provider keep?
Who fixes it when it breaks, and who owns it when you're gone?
Can we walk away after the first month if nothing has changed?
Can you explain your recommendation to us out loud, right now, without the slides?
We can help with this, too. See "When to bring someone in" at the end.
Document it, and keep it going
Once a change is agreed, it needs to be documented and tracked.
Documenting it means recording what the change is, what it replaced, who owns it, what's due and when, and which numbers it should move. Tracking it means keeping that record current as the work happens.
We recommend letting AI do most of this, with the kind of system described in "Where we often start." Set it up so that once a change is agreed, an AI keeps the record current, reminds owners what's due, and flags what's slipping. Make sure someone is checking that what it says is right. That can be each owner for their own changes, or one person for all of them.
You don't need AI for this. You can keep the record yourself, in a document or a spreadsheet. Either way, the record has to exist, and someone has to keep it current.
This kind of tracking is one of the weakest areas in most businesses. It's also one of the things AI does well, at a very low cost.
This is one of the things we can help with. See "When to bring someone in."
Check yourself
Did we choose what to do first on purpose, and say why?
Did we delete before we automated?
Have we left people the parts of the work they like and do well?
Are today's numbers written down?
Does the owner know they're the owner, and exactly what that entails? Could they tell you what they're responsible for, what they can decide on their own, and how they'll know it's working? (Ask. You'd be surprised.)
If something that runs on its own breaks at 4:45 on a Friday, who finds out, and what do they do?
Is the change being tracked somewhere, and is someone checking that the record is right?
Working alone?
Same order: delete, simplify, fix what comes in, then automate. When you work alone, it's tempting to skip ahead to a tool. Do each step anyway.
Step 8: Check
Did the numbers move?
Why it matters
Most improvement efforts get measured exactly once, at the demo, by the person who built it. Then everyone moves on, and nobody ever finds out whether it worked. This step takes thirty minutes and puts you ahead of nearly everyone.
How to run it
Put at least three dates on the calendar the day the first change goes live. We recommend four: about two weeks, one month, two months, and three months out. The exact days don't matter. Even better is a short check-in every two weeks, with a bigger review at three months. Early on, the more often you check, the better. At each one, the leader and the owners answer five questions.
Did the numbers move? Compare each one with the baseline you wrote down, and pay special attention to the primary number you identified. Up, down, or flat, write it in. Then ask the follow-up questions below.
Did the tasks happen? Go down the roadmap. Done, late, or dropped, and why.
Where did the time go? If the change freed up hours, what are people doing with them? If the answer is "nothing in particular," the hours weren't really saved.
How are the people? Ask the three pulse questions again. Compare with the start.
What surprised us? Good or bad. Surprises are where the next improvement is hiding.
On the numbers, don't stop at whether they moved. Ask:
How much did they move?
Did that meet your expectations?
If a number didn't move, is there a clear reason why?
Did this uncover a constraint you weren't aware of?
Is it still the right number to be watching?
Worksheet
| Baseline | About 2 weeks | About 1 month | About 2 months | About 3 months | |
|---|---|---|---|---|---|
| The goal number | |||||
| The first change's number | |||||
| Pulse: I understand where we're going | |||||
| Pulse: I have a say | |||||
| Pulse: I'm willing to change how I work | |||||
| Hours freed, and what they went to |
Reading the result
Success can take a lot of forms. The most important thing is that you're doing the review at all. The job is to pause, look honestly at what happened, and then adjust.
The numbers moved. Good. Make sure it's because of the change and not the season. Then say so, out loud, to the people who did it.
The numbers didn't move, but the change is in place. You may have fixed something that wasn't holding you back as much as you thought. That's useful. Go back to Root causes with what you know now.
The change isn't really in place. That usually points back to the People step more than to the tool. One way to tell is to count how much of the work now runs the new way, and how much still runs the old way. Find out what's in the way, and put it on the wall.
It made things worse. Stop it. Nobody gets blamed for an honest test. People get blamed for defending a bad one for a year.
You picked the wrong number. It happens. Choose a better one, write down today's value, and keep going.
A different constraint showed up. Something you didn't know about has to be solved first. Solve it. Your target can stay the same.
Check yourself
Are we measuring the real numbers, or the flattering ones?
Can the team see what changed, in plain words, without asking anyone?
Have we told the people involved what we found, including the parts that didn't work?
Are the check-in dates on the calendar, starting about two weeks in?
Did we ask the follow-up questions, or only whether the numbers moved?
Step 9: Adapt
Adjust as you learn. Then go after the next problem.
Why it matters
When you fix the thing that was holding you back, something else becomes the thing holding you back. That means it worked. The constraint moved.
The risk is that nobody notices. The team keeps polishing the old problem out of habit, and the habit becomes the new constraint.
How to run it
Ask where the constraint is now. Where does work pile up today? If things are working, it's hopefully somewhere new.
Revisit the wall and the board. Which notes can come down? What new ones have appeared? Update your board to match, so it stays the record of where things stand. AI can help keep it current, with someone checking that it's right. Before you change anything, save a photo of the old wall, or a dated copy of the old board. It's good for morale.
Update the roadmap. Keep what's working, change what isn't, and cut what no longer matters. Cutting is encouraged.
Pick the next change. It can be one change or a small set, like a quarter's priorities. We recommend an absolute maximum of seven per quarter. Size it to your team. Each change needs its own owner, and all of them need real attention.
Set the rhythm. Owners look at their numbers weekly. The leader and owners check in every two weeks, or monthly at the least. The team reviews the roadmap at least every quarter. Once a year, block real time and run the whole of Aim again.
The rhythm
Treat these as minimums. Early on, more often is better.
| How often | Who | What |
|---|---|---|
| Weekly | Each owner | Is my number on track, behind, or stuck? What one decision do I need? |
| Every two weeks, or monthly at the least | Leader and owners | Check results. Clear what's stuck. |
| Quarterly | The team | Where's the constraint now? Update the roadmap. Pick the next change. |
| Yearly | Everyone | Step back and run Aim (steps 1 to 4) again from the top. |
Check yourself
Has the constraint moved? Where is it now?
Are we still working on the old problem out of habit?
Does the review rhythm exist on a calendar, or only in this document?
When's the next time we step back to look at the whole picture?
Is the board up to date, and is someone keeping it that way?
Does each change we picked next have its own owner, and can the team give all of them real attention?
When to bring someone in
You can run all of this yourself.
Here are some situations where it might make sense to bring in a third party. This is the work we do, and we'd be glad to help.
Candor when people outrank each other. When the leader both takes part and referees, people either defer or stay quiet. A neutral outsider usually gets franker answers than a boss does.
Your own blind spots. By definition, you can't see them. Someone from outside can.
Time. Doing this well takes hours nobody inside has, on top of the work they already do.
Asking the obvious question. It's easier when you won't be at the office party.
Keeping up with the tools. They change faster than a working business can follow. Someone whose job is to follow them can save you from buying the wrong thing, or from signing a contract you'll regret.
You need help building a specific workflow. You know what needs to change, and you want help designing it, building it, or making sure it's done right.
Signs it's time
People wrote different goals, and nobody will say whose is right.
The constraint might be the leader, and the leader is running the session.
You've tried this before and it faded by the following Monday.
You want it to be fun, and you'd rather take part than run it.
AI for admin and shared project management
This is where we often start with AI.
The problem. Admin has to get done to run any business and to do any job. Plenty of people who are good at their work struggle with it. Notes don't get taken, records don't get kept, and paperwork runs late. Across a team, tasks, projects, and deadlines fall between people.
What we set up.
For each person: AI that helps them keep their own notes, records, tasks, and deadlines in order.
For the team: one shared place for projects, where everyone can see what's owned, what's due, and what's done.
For the changes you've agreed on: tracking that keeps going after the meeting ends, so owners are reminded and slipping work gets noticed.
Training: we teach each person to use it on their own work, so it keeps getting used after we leave.
What it's built on. Wherever we can, the software you already pay for, plus a basic AI subscription. That combination can do far more than most people expect. Often it starts in a spreadsheet.
What changes. Your people keep doing the work. Less gets lost, and there's no longer a weak link when it comes to organization.
Email dylan@dunderblixem.com and we'll tell you whether it's a fit.
How we can help
Work through it with others. A small guided group, working through one part of this playbook together. You tell us which part you want help with, and we build the group around what people ask for.
A day with us. One day on how to think about AI in your business, and how to get your people on board. We run "the Wall" exercise with your team. It fits well inside a retreat or a leadership day you've already planned.
2-week implementation sprint. We run this playbook with your team. One prep week, then two weeks of sessions and work, with at least one day on site recommended. Most sprints end with a six-month roadmap, the first change underway, and a clear set of measurements to judge success by. The first change can be a specific workflow, or the admin and project management system described above.
Build a specific workflow. If you already know which process, workflow, or tool setup needs work, we design and build it with the people who use it.
AI for admin and shared project management. The system described above. We can set it up on its own or as part of a sprint.
Built around your retreat. Our favorite. Groundwork before, the real work at the retreat, and follow-through after. Reminder: we lean hard on the fun part, and every game ends with something your team will use.
We speak at events.
Whatever we do with you, training your people to use it is part of the work.
If what you need isn't on this list, tell us. We build the work around what you need.
Your time and money are valuable, and so are ours. We only take work where we think we can help. If you don't need us, we'll say so.
Recommended additional reading
Finding the constraint and mapping cause and effect
The Goal, Eliyahu Goldratt. The theory of constraints, told as a novel about a factory.
It's Not Luck, Eliyahu Goldratt. The sequel, where the cause-and-effect mapping tools are introduced.
The Logical Thinking Process, H. William Dettmer. The full manual for those mapping tools.
Thinking in Systems, Donella Meadows. A short introduction to loops and how systems behave.
The Fifth Discipline, Peter Senge. Systems thinking applied to organizations.
Goals and principles
Principles, Ray Dalio. Write down the principles you work by, and keep adding to the list.
The 7 Habits of Highly Effective People, Stephen Covey. Begin with the end in mind.
The Five Most Important Questions, Peter Drucker. Mission, customer, and what the customer values.
Measure What Matters, John Doerr. Objectives and key results (OKRs).
The 4 Disciplines of Execution, McChesney, Covey, and Huling. A few goals, a scoreboard, and weekly follow-through.
Traction, Gino Wickman. A two-page vision and quarterly priorities.
The Advantage, Patrick Lencioni. One rallying goal for the whole team.
Competing Against Luck, Clayton Christensen. What your customers really hire you for.
Getting organized
Getting Things Done, David Allen. Get everything out of your head and into one place before deciding what to do about any of it.
Teams and running a good session
The Five Dysfunctions of a Team, Patrick Lencioni. People won't name real problems until it's safe to.
Gamestorming, Dave Gray, Sunni Brown, and James Macanufo. Run a session as a game, with an opening, a middle, and a close.
The Surprising Power of Liberating Structures, Henri Lipmanowicz and Keith McCandless. Simple formats that get everyone in the room contributing.
Thank you
Thank you for reading this far, and for doing the work. We sincerely want this playbook to be helpful and useful, and we hope you get good results with it.
Tell us how it went
This is version 0.1, and we're actively developing it. If you run it, we'd love to hear what worked and what didn't, especially what didn't. Tell us if an instruction was confusing, if part of the design is ugly, or if handing it to your AI didn't work. Tell us whether your results were good or bad. It's how the next version gets better. Email dylan@dunderblixem.com.
An offer for playbook readers
If you've worked through any part of this, we'd like to hear about it and help you with the next step. Book a 20-minute call and bring what you have: your goals, a photo of your wall, your map, or your roadmap. We'll talk it through with you and tell you what we'd do next. We will never put you through a high-pressure sales pitch.
If you want to go further
Everything in "When to bring someone in" is work we do: a guided group, a day with your team, the 2-week implementation sprint, building a specific workflow, AI for admin and shared project management, and a retreat built around this process. Whatever we do with you, we train your people to use it.
Thank you again. We hope this serves you and your people well.
Dylan and Bert
Dunder Blixem