Ask for an AI strategy and you usually get one sentence back: “We want AI everywhere.” Six months later, nobody can say whether it worked. A real strategy names what changes, by how much, and who owns it. Here are six steps to build one, and a free workbook to do it in.
Nothing under the slogan has a name or a number, so in six months there’s nothing to check.
One change, today’s number and the target, an owner, and the fix that gets it there.
Six months on, revenue is up. Maybe the AI helped, maybe it didn’t, because revenue moves for a dozen reasons. Without a number tied to the work itself, nobody can say whether to keep going.
So what’s different since the spring?
Nobody can say whether to spend more or stop, so the slide goes back up.
The trouble started with the ask itself. It was the size of the whole company.
Not to a department either. Work is a chain of steps: read the request, price the job, approve the terms. AI helps with steps, so a strategy has to be written at that size.
AI goes on steps, so a strategy has to be written about steps too.
That starts with finding the ones that hurt.
Ask your leaders, and a few people who do the work, where things go wrong. What would break if we doubled? Where do we type things twice? Then score each answer 1 to 5 for how often it happens and how bad it is, and multiply.
Three pains, picked by their numbers rather than by whoever spoke loudest.
Next, look closer at the top one. Where does its time actually go?
Before picking a fix, ask the people doing the work how long it takes, roughly, and how often, and note who told you. Here, almost all of the three days is the request sitting in an inbox.
The work itself is quick. The time goes on waiting.
So that’s where the fix belongs. First, check what each pain depends on.
Three quick questions for each pain: is the process written down, do the systems talk, is the team ready? A “no” just tells you what comes first. Then pick the fix. Only one of the four kinds is AI.
Two of the three don’t need AI at all, and that’s normal.
It’s a stronger story for the board, too: money goes on whatever fixes each problem, and AI only where it genuinely helps.
Each row turns a pain into a change: today’s number and who said so, the target, the owner, the fix, and a rough sense of what it’s worth. Add one line for what you’re not doing yet. Here’s a finished one.
Not doing yetThe board pack. It hurts, but only once a quarter, so it waits.
Three rows, three owners and three numbers anyone can check in six months.
Writing it down is the easy part. Proving the first fix works comes next.
Take twenty past quotes where you know the right price. Agree the bar first, then run them. Next, run it alongside the old way for a couple of weeks. And write down what never happens without a person.
Matched the real quoteUnsure, so it askedWrong, and caught
Clears the barBoth run on the same requests. Where the two quotes differ, a person looks at why before anyone switches over.
AIWhen it isn’t sureIt stops and asks the sales manager, instead of guessing.
If it misses the bar, it isn’t ready yet, and the step stays with a person until it is.
Once it’s live, the question changes. It isn’t “does it work?” any more. It’s “did it pay off?”
Read each number 30 days after launch, then every quarter, against where you started. Expect it to improve gradually as people adjust. And plan for them: tell each team what changes, have their own leader say it, and expect a dip first.
Two rows can answer the board’s question. The third can’t, because nobody timed it at the start.
Which is why every number on the page needs a name and a date.
It points everyone at a tool and nobody at a problem. The people doing the work hear “use AI more” and have no idea what to do differently on Monday.
Often the best fix is a clearer process, or two systems connected so nobody retypes. Put AI where something genuinely needs reading, drafting or sorting, and use something simpler everywhere else.
Your own ChatGPT works because you feed it your files and check what it says. Roll the same tool out to a team and nobody is doing that checking, which is why the page has to say who confirms what. More on AI across a team.
You need an owner for each change, and that’s usually the person already running that process. A new title above them doesn’t own any of the rows.
Writing the page is the easy part. The hard part is the review a few months later, when a row hasn’t moved and nobody wants to say so. These four rules keep it honest.
Once it runs to a second page, people stop reading it. Saying what you’re not doing yet is a decision too, and it stops the same ideas coming back every month.
The owner is usually whoever already runs that part of the work. Today’s number comes with a name and a date: who told you, and when.
Prices, payments, anything sent to a customer: a person checks it before it counts. Write that list down before launch, not after the first mistake.
If a row hasn’t moved in two reviews, it’s either the wrong row or the wrong fix. Work out which, and swap in the next pain on your list if it’s the row.
You can do the first four steps yourself, with the people you already have, and the workbook walks you through them. Here’s the order that works for most teams.
Ask your leadership team and a few people who do the work. Score each pain by how often it happens and how bad it is, one to five each, and multiply.
Ask how long each one takes, roughly, and how often, and note who told you. Then: is the process written down, do the systems talk, is the team ready?
Twenty past cases, a bar agreed up front, and a person confirming anything that matters. If you’d like help with this part, it’s what an AI Jumpstart is for.
An AI Jumpstart runs these six steps with you for one part of the business and proves the first fix on your own records. It takes two to four weeks, and you finish with the page filled in.