The applied-AI grid

Ask ten chief executives where AI should start and you will hear ten answers, most of them "everywhere". Everywhere is how programmes stall. The companies that make progress pick a square, finish it, and use what they learned to finish the next one faster. The grid is the map for doing that on purpose.

Two directions

Down the side: the functions. Revenue, marketing, service, delivery, finance, everyday work, product, software quality, security and oversight. Each has routine work that agents can carry and decisions that must stay with people.

Across the top: three stages.

Land, the first 30 days: connect the data, name an owner, and get one team doing one job with agents. Expand, days 30 to 90: every team in that function, and the next job. Compound, months three to twelve: managers supervise agents day to day, and what one function learns starts to help the next.

The grid

Each row names the piMonk app for that function, its real status today, and what Land looks like. Expand and Compound follow the same pattern in every row.

Every status is the real one on 29 September 2026. Launch dates are targets, not promises.

Where to start

Start where three things meet: the work is repetitive, the data already exists, and someone is willing to own the result. For most growing companies that is revenue or service. The calls and tickets are already there, the jobs are easy to count, and a better week shows up quickly.

Do not start where the stakes are highest and the data is thinnest. Finance and security matter most, which is exactly why they benefit from rules and habits proven somewhere else first.

Pick a square, finish it, measure it. Then the next square costs less than the first.

How the second square goes faster

The grid is also a map of acceleration. Three things carry from one square to the next.

The approval rules. Once a company has written down what an agent may finish alone and who approves the rest, the second function borrows the pattern instead of inventing it.

The measurement. The same baseline-and-day-30 method works in every row, so the board sees results in one format.

The habits. Managers who have supervised agents in one team know what to expect in the next. Adoption is a skill, and it compounds.

A fourth will join them: memory shared across the apps, so the second app knows what the first learned. Connecting the apps to one memory is on our roadmap; today each app keeps its own.

The case against a map

Some argue that AI moves too fast for plans, and that companies should let a thousand experiments bloom. Experiments are useful for learning what is possible. They are poor at changing how a business runs, because nothing is ever finished. A map does not slow the experiments down. It decides which of them become the way work is done.

What this means for CEOs and investors

For a chief executive: choose the first square this quarter, name its owner, and agree the day-30 decision before it starts.

For an investor: ask each portfolio company to place itself on the grid. Where each one sits, and how fast it moves between squares, is a clearer signal of AI progress than any count of pilots.

The bridge from AI ambition to business results is built one square at a time. We argue why in The bridge to applied AI and how to count the result in Where the return comes from.