AI Boss Tools | Canvas • Back to the Hub
Most AI projects fail for a plain reason: nobody named the one number they were trying to move. This canvas fixes that before you spend a cent. One page, filled in with your team, that names the number, the process behind it, the person who owns it, and the check that tells you in four weeks whether to keep going. It borrows the spine of the North Star metric Microsoft uses to point thousands of people at one direction, and shrinks it to fit a business that will never run an A/B test.
A strategy deck describes everything and commits to nothing. A canvas does the opposite. It fits on one page because it forces a choice: one number, one lever, one process, one owner. When the whole team can see the plan at a glance, the plan gets argued with, and a plan that survives an argument is worth running.
The order matters. You name the outcome first, the number that has to move this year, and only then the process and the tool. Most teams do it backwards. They buy the tool, then go looking for a problem it solves. That is how you end up with a licence nobody uses and a board that stops believing the next AI request.
Boxes 1, 4, 5 and 9 are the minimum: the number, the process, the owner, and the four-week check. Fill only those and you already carry more discipline than most AI programmes ever manage. Everything you type stays in your browser. Nothing is sent anywhere.
Open the Canvas →A North Star metric is the single measure that best captures the value you deliver to a customer. Microsoft's own product teams run on it, and their research lays out the structure plainly: a North Star sits between fast-moving input metrics that you can change this week, and slow-moving business results like revenue that arrive far too late to steer by. Around it sit guardrail metrics, the things that must not get worse while you chase the number, and a habit of testing whether a change actually caused the movement, rather than just happening beside it. That is the whole discipline. This canvas maps it onto nine boxes a small team can finish in a meeting.
One outcome that captures the value you deliver. Not revenue, which moves too slowly to steer by, and not a vanity count of logins, which moves too fast to mean anything. The honest number in between, with a target and a date.
The parts you can move this week: the customer you serve, the lever you pull, and the one process you change. Microsoft calls these leading metrics. They move first, and they are your early evidence that the North Star will follow.
Microsoft pairs every North Star with guardrail metrics: the things that must not get worse while you chase it. On the canvas that is your rules box. Who checks before anything leaves the business, and what data never goes into a free tool.
Microsoft does not trust a number that moved next to a change it made. It tests whether the change caused the move. You will not run an A/B test in a four-person firm, but you can still ask the honest question in week three, then keep it, fix it, or drop it.
Microsoft's product experimentation work defines the North Star as the outcome that best reflects the value delivered, sitting above fast input metrics and below slow business results, paired with guardrails and tested for real cause and effect. That structure is the backbone of this canvas.
MIT found about 95 percent of enterprise generative AI pilots produce no measurable profit impact. That is what happens without a named number and an owner, and it is why the canvas asks you to pick one process and check it in four weeks.
BCG puts roughly 70 percent of the value in people and process, and only the rest in the technology. The canvas spends most of its boxes there, on the owner, the process, the data, and the rules, because that is where the return actually comes from.
Gartner expects at least 30 percent of generative AI projects to be dropped after proof of concept, on poor data, weak controls, or unclear value. A written keep-fix-drop decision in week four is how you make that call on purpose instead of by drift.
Fill it in with your team
One page. One process. Fill it in before you buy any tool. Boxes 1, 4, 5 and 9 are the minimum. Your work saves in this browser as you type.
The one number this year must move. Pick the number that would tell you the AI worked, not a vanity count.
Who buys from you, and what do they want faster, cheaper, or with less hassle? Where do they lose patience with you today?
Which one moves your North Star number? Pick one to start. You can change it next round.
One process, named. Where it starts, where it ends, and how long it takes today.
One person who runs the experiment. Not the busiest, and not the youngest by default.
Sales, messages, receipts, stock. Where does it get recorded so AI can read it?
Which tool, what it costs per month, and who holds the login. Size it from the outcome →
Your guardrails. Who checks before anything leaves the business, and what data never goes into a free tool.
Did the number move? By how much? Then make the call, and name the next process.
A finished canvas is not the goal. The goal is the loop it starts. Name the number, give one person one hour a week, run the process for three weeks, then look at the number and decide. Keep what works, fix what nearly works, drop what does not, and move the same discipline to the next process.
Do that four times a year and you will have tested twelve weeks of real work against real numbers, while your competitors are still on slide fourteen of the strategy deck. That is the whole point of a North Star. It turns a vague ambition to use AI into a short list of things you can actually check.