It comes up in almost every training session I give. Usually during the coffee break, a little sheepishly: “OK, I get that AI is powerful. But in practice, here, what do we actually use it for?”
Not a naive question. The most honest question a manager can ask in 2026.
Because for most organizations, the problem is no longer access to the tool. The licenses are paid for, the accounts are open, and a few curious employees are working wonders on their own. The problem is that we’re walking around with a shiny new hammer, looking for nails. And when you go looking for nails, you end up hitting anything.
This article offers a simple tool to flip the approach: a nine-block canvas, inspired by Alexander Osterwalder and Yves Pigneur’s Business Model Canvas, that starts from the real work rather than from the technology. I call it the AI Canvas.
A tool in search of a use
Most of the AI adoption efforts I see start at the wrong end. It begins with a flashy demo, an article, a persuasive vendor. Then someone asks where on earth we could plug it in.
The outcome is predictable. Pilot projects that dazzle in meetings and quietly die three months later. Employees using AI to rewrite emails that were already fine, while the real paperwork, the kind that eats their days, stays untouched. And leadership concluding, a bit disappointed, that AI is overrated.
AI isn’t overrated. It’s poorly targeted.
Three things explain the gap.
Real work is invisible. A job description says what someone is supposed to do. It says nothing about the twenty minutes spent every morning copying data from one system to another, or the monthly report nobody reads but everybody produces.
Pain points become normal. After a few years, you stop seeing what slows you down. It’s part of the scenery. Ask someone what frustrates them at work and they’ll have to think hard. Shadow them for a day and you’ll find ten things.
Fear clouds the thinking. When people hear “AI,” some of them hear “automation” and translate it as “my job is on the chopping block.” Nobody eagerly volunteers the tasks that could disappear. Least of all their own.
So we need a tool that makes the work visible, brings the pain points to the surface and reassures people before transforming anything.
What we can borrow from Osterwalder
The Business Model Canvas, published in 2010 by Alexander Osterwalder and Yves Pigneur, changed how thousands of companies think about their business model. Its strength comes down to one idea: put a complex system on a single page, in nine blocks, so everyone around the table sees the same thing.
A few years later, the same authors zoomed in on two of those blocks with the Value Proposition Canvas. You describe the customer through their jobs, their pains and the gains they hope for, then check whether your offer truly relieves those pains and truly creates those gains.
That’s exactly the reflex missing from most AI initiatives. Start from the need. Not from the solution.
But these tools can’t be transplanted as is. They need three adjustments.
The customer becomes a role. You don’t analyze “the company,” which is too broad to produce anything but generalities. You analyze a specific role: the administrative assistant, the payroll team, the field advisors. One canvas per role or per process.
The offer becomes a mode of intervention. AI doesn’t do just one thing. It can automate (it runs on its own), augment (it prepares, the human decides) or assist (the human does the work, AI helps as needed). Choosing the mode is already half the job.
There has to be room for what must stay human. That’s the blind spot of the original canvas. And it’s precisely where team buy-in is won or lost.
The AI Canvas in nine blocks
The AI Canvas keeps the structure and layout of the BMC. Each block takes on a new meaning, but the logic stays the same: on the right, what is desirable; on the left, what is feasible; along the bottom, what is viable.
| # | BMC block | AI Canvas block | Question to ask |
|---|---|---|---|
| 1 | Customer segments | Beneficiaries | Who benefits from the improvement: the employee, the client, the manager? |
| 2 | Value proposition | AI value | Which pain are we relieving? Which gain are we creating? |
| 3 | Channels | Point of integration | Where does AI fit into the daily routine: email, business software, Teams, a form? |
| 4 | Customer relationships | Human-AI relationship | Automate, augment or assist? Who validates? What stays human? |
| 5 | Key activities | Target process | Which steps change, before and after? |
| 6 | Key resources | Data and tools | What data is needed, in what shape, accessible to whom? |
| 7 | Key partners | Allies and dependencies | Who needs to get on board: IT, the vendor, an internal champion, the union? |
| 8 | Cost structure | Costs and risks | Subscriptions, training, compliance, the cost of a mistake? |
| 9 | Revenue streams | Measurable gains | Hours recovered, errors avoided, turnaround times cut. And where does that time go? |
The right side: is it worth it?
The AI value block is the heart of the canvas. This is where you graft in the entire Value Proposition Canvas: tasks, pains, gains, and a fourth zone I always add, meaning. What the person loves doing. What makes them proud of their work.
You name it before talking about automation. Not to make it look nice. So people know what you’re protecting before you ask them what you’re changing.
The Point of integration block is the most underestimated. A solution that forces people to open a new tool, log in somewhere else or copy and paste between two windows will die of friction. The best AI is the one you run into where you already work.
The left side: can we actually do it?
Plenty of good ideas die here, and that’s a good thing. The Data and tools block asks the uncomfortable question: does the information really exist, or does it live in one person’s head and in three Excel files that don’t talk to each other?
This zone also forces you to face privacy law head on. In Quebec, this is where Law 25 comes in; elsewhere, GDPR or your local privacy legislation. As soon as a process touches personal information (employee files, client data), you need to know where the data goes, who can access it and whether a privacy impact assessment is required. Better to find out in a workshop than after signing a subscription.
The bottom: does it hold up?
The Measurable gains block replaces revenue. You write down what you can really measure: hours, turnaround times, errors. And a question almost nobody asks: that recovered time, where does it go? I’ll come back to it, because it’s probably the most important question of the whole exercise.
Before the canvas: pick the right processes
Filling out a canvas takes a solid hour of discussion. You don’t do it for every task in the organization. You have to sort first.
For that, I use a two-axis matrix. The first axis measures pain level: how repetitive, slow, error-prone or mentally draining the task is. The second measures human value: how much it calls for judgment, relationships, empathy, or gives meaning to the person doing it.
| Low human value | High human value | |
|---|---|---|
| High pain level | The chore: we automate | The noble burden: we augment |
| Low pain level | The noise: we wait | The sanctuary: we protect |
The chore is data entry, report formatting, sorting emails, booking appointments. Nobody will shed a tear if AI takes it over. It’s the ideal ground for early quick wins.
The noble burden is more subtle. Think of a counselor’s case notes, a technician’s inspection report, the analysis of a credit file. The task is heavy, but it requires professional judgment. Here, AI drafts a first version, summarizes, structures. The human keeps the pen and the responsibility. This is often where the biggest gains lie dormant, because these are tasks you can neither eliminate nor delegate.
The noise covers tasks that are neither very painful nor very meaningful. You can get to them later. Or never.
The sanctuary is the heart of the job: welcoming a young person who’s struggling, negotiating with a key client, the team meeting where a lingering tension finally gets resolved. AI has no business there, except to free up time so people can devote more of themselves to it.
This matrix does something the canvas alone doesn’t: it makes explicit what you refuse to automate. And that completely changes the conversation with teams.
How to run a first workshop
One week before: the friction log. Ask participants to note, over five working days, every moment they sigh. One line is enough: the task, how long it took, what was irritating. You’ll walk into the workshop with real material instead of impressions. It’s probably the most rewarding step of the whole process.
In the workshop: sort, then dig. Place the pain points you collected on the matrix. Choose two or three processes in the chore or the noble burden. One canvas per process, as a team, with the people who actually do the work. Not just their bosses.
The order matters. Start on the right: Beneficiaries, AI value (including the meaning zone), Human-AI relationship, Point of integration. Then move to the left to test feasibility, and finish with the bottom. Start with costs and imagination shuts down before it ever opens up.
Traps to avoid.
- Pain points that are too vague. “Communication” is not a pain point. “I retype the same information into three systems” is one.
- The manager who fills it out for the team. They know the official process. The team knows the real one.
- The vendor in the room. The moment you name specific tools, the discussion swings to the solution. Save product names for later.
- The perfect canvas. An empty block is useful information: it tells you what you need to go check.
After the workshop: a test, not a project. Pick a single process, test it for four to six weeks, and measure what you wrote in the Measurable gains block. Then decide. AI adoption is built through small, documented wins. Not through a grand transformation plan.
The recovered time: where does it go?
Back to the Measurable gains block and that question almost nobody asks.
If you don’t answer it, the answer shows up on its own. The freed-up time fills with emails, meetings, new requests. Six months later, the organization has invested in AI, employees are working as hard as ever, and nobody knows exactly what changed.
If you do answer it, everything shifts. Time recovered from the chore goes back to the sanctuary. More time with clients. More time to think. More time to train the next generation, to deal with the tension you’ve been putting off for months, to take care of the team.
That, in my view, is where the real promise of AI at work lies. Not replacing people. Not making them faster. Giving them back time for what only they can do.
The AI Canvas is just a sheet of paper. Nine blocks, a few questions. But it forces an organization to take a hard look at its work before looking at the tool.
Most have never done it. Maybe that’s the first gain.