Stay in control
AI integration has become a program. Several initiatives, several vendors, real money, and no single place that says what each one costs and returns. Adoption is spreading faster than it is being organized. The finance function wants to see net value rather than net spend, and the legal function wants to know that the risks are identified, controlled and documented rather than discovered later.
Ninety seconds: how much governance your ambition actually needs, and the jump from a department working differently to a governed capability. Everything below is the same argument in detail, for whoever wants it.
AI integration run like any other investment program. Ideas go in one end as initiatives, each with a name next to it. Nothing gets money before it has a business case. Gates weed out the weak ideas fast. Progress is visible every week. Nothing is called done until the value shows up in the numbers, and leadership can see what each initiative cost and returned. The same gates give the legal function what it needs: every initiative carries a record of what data it touches, what it is allowed to decide on its own, who approved that, and what happens when it is wrong. Not a policy document in a drawer, a field on the card that has to be filled before the money moves.
Net value rather than net spend. Bad ideas die in week two instead of month six. The finance function sees the return. Senior time goes to the decisions that matter rather than to refereeing initiatives that compete for the same people. We measure impact, not activity.
Business units have to show up as owners, or the office has nothing to run. The office is only as good as its cadence, so the rhythm is set before the tracker. And our role is oversight and stewardship: we run the program and its governance; the results of each initiative stay with the unit that owns it.
A program is much easier to run when everyone agrees where the company actually is today. This is the ladder we use. It is ours rather than an industry standard, and so far it has described every organization we have walked into. It describes the company, and we name its rungs rather than numbering them, because the numbered levels belong to individual initiatives on the separate scale set out below.
What separates the rungs is not how much AI a company uses. It is how initiatives come into existence, and who is in the room when they do. In order: Unlicensed, nothing sanctioned. Licensed, tools bought and used ad hoc. Pockets, someone capable builds something good that depends on them. A department, one real win with a named owner and a measured before and after. Governed, a program mid-transformation with gates and a business case before money. Structural, the same thing in steady state.
The jump that decides everything is from A department to Governed. Below it you are collecting wins. Above it you are running a system that produces them, and that is a different company. If the intention is genuinely to change how the organization works at scale rather than to run a few good projects, Governed is the floor, not the stretch.
| Rung | How things get built | Who is at the table | Why you would leave |
|---|---|---|---|
| Unlicensed | They do not. Somebody may be using a personal account on their own work, which nobody has sanctioned and nobody has stopped. | Nobody. It has not come up. | Because the shadow usage is already happening and you are carrying the risk of it without any of the benefit. |
| Licensed | Ad hoc. Tools are bought, people use them for whatever occurs to them that day, and nothing is written down. | Procurement signed a contract. That is the whole of it. | Because you are paying for seats and cannot name a single process that works differently. The value is real and completely invisible. |
| Pockets | Whoever is capable and curious builds something. It works, and it works because of them. | The builder, and their manager, informally. | Because it does not survive the person leaving, it cannot be reported to anyone, and two departments are quietly solving the same problem twice. |
| A department | On purpose, with a named owner and a before and after measured on the department's own numbers. One real win, defensible in front of anyone. | The department, and finance for the first time, because somebody wants the number checked. | This is the rung where most companies decide to move, and they are right to. You have proof. The instinct that follows is to spread it, and spreading it without a process is how the next four attempts fail. |
| Governed | Through a program, and the company is mid-transformation. Ideas enter as initiatives, a business case comes before money, gates weed out the weak ones, one tracker holds all of it, the value is checked after. Volume is ramping and the shape keeps changing. | Finance has a seat and a veto. Legal has a seat and a standard. The business units show up as owners rather than as requesters. They are all in the room because every decision is still a new one. | Because a transformation is a phase, not a destination. The program is expensive by design, it is meant to end, and every quarter it runs past its usefulness is a quarter of overhead on work the organization should now be doing by itself. |
| Structural | To order, in steady state. The transformation is finished. Initiatives are produced through a process that runs at whatever rate the business needs, whether or not anyone is championing them, the way the company already produces anything else. The development of them has been industrialized. | Nobody in particular, and that is what finished looks like. It is its own function with its own profit and loss, which finance reads like any other. Legal is not reviewing initiatives one by one because the guidelines and the safeguards were defined once and are now in place. | Nowhere to go. You are here. |
The distinction people find hardest is the last one, so here it is in one line. Governed is a transformation. Structural is a steady state. At Governed you are ramping: the volume is climbing, the shape keeps changing, and every decision is still novel enough that the people who carry the risk need to be in the room for it. At Structural the ramp is over. The capability has its own profit and loss, which the finance function reads the way it reads any other function. Legal is not at the table because the guardrails were defined once and are now simply in force, which is the opposite of nobody having thought about them.
That is why the third column empties out at the top rather than filling up. At the bottom, finance and legal are on the side, blessing a pilot if anyone asks. At Governed they are in the room with an actual say, and that is not bureaucracy, it is what makes a number believable and what stops the first serious incident from being the moment anyone thinks about exposure. At Structural they step back out, because the questions they were answering have been settled and written down.
Very few organizations of any kind are at Structural for anything, which is worth knowing before somebody promises it in a board paper.
Why Governed is the floor at a large ambition. Not for the process itself, but for the arithmetic underneath it: at that ambition you will be spending real capital, touching a wide range of systems and a wider range of data, and changing how people in several functions do their jobs. Improvising that is how companies end up with money spent that nobody can account for, data somewhere it should not be, and a program that cannot be defended when somebody finally asks. The structure is not the ambition getting slower. It is the thing that lets the ambition be large without being reckless.
The ladder says where you are. The ambition says where you are going, and those two together are what decide how much structure you need. A department is a success story you can tell. Governed is a capability, and it is the first rung at which the finance function sees a return rather than a collection of good anecdotes.
The rungs above describe the company. These describe one initiative, and they are what the tracker actually runs on. Six levels, four of them before anything is built, because most of what kills a program happens in the planning half and nobody notices until the money is already spent.
In short: L0 scoped with a rough value, L1 an owner named and the monthly impact sized, L2 the business case approved, L3 a detailed plan that releases the people and the budget, L4 every milestone executed, L5 the benefit accruing at full run rate. Only L5 is finished.
Two gaps do most of the damage. A program with twenty initiatives in flight and none past L2 has a business case problem, not a delivery problem. And a program that declares victory at L4 has implemented something without checking that it paid.
| Level | What is true at this point | Who signs it off |
|---|---|---|
| L0 | Scoped, with a rough value estimate. Somebody has written down what this is and roughly what it might be worth. | None needed. Any idea is welcome at L0 |
| L1 | An owner is named, and the monthly impact is sized. The owner is the senior-most person who actually does the work, not a sponsor. | The integration office and the workstream sponsor |
| L2 | The business case is approved. This is the gate money passes through, and the one most programs do not really have. | The finance function and the sponsor |
| L3 | A detailed execution plan exists: milestones that are specific and actionable, each with a date and an owner. Clearing this gate is what releases the people and the budget to build. | The integration office and the workstream sponsor |
| L4 | Every milestone is executed. The initiative is implemented. Note what this does not say: it does not say the value has arrived. | The integration office and the workstream sponsor |
| L5 | The financial benefit is accruing at full run rate. Only here is an initiative finished. | The finance function |
Four of the six levels sit before anything is built, and that is not caution for its own sake. The money is spent on the build. Everything before it is cheap, and everything after it is committed, so the planning half is the only place where a bad idea can still be stopped for the price of the conversation.
Which is really a filter for one thing: whether an initiative is valuable or merely helpful. Spending a million dollars to remove a hundred thousand a year of work is helpful. Everyone involved will be happier and the work will genuinely be gone. It is not valuable, and a program that cannot tell those two apart will fill its tracker with things nobody can defend to a finance function. L2 is where that question gets asked properly, which is why it is the gate the money passes through.
The gate out of L3 is the one that matters operationally, because that is where an initiative stops consuming thinking and starts consuming people. Clearing it is what releases the builders and the budget. Before it, an initiative costs meetings. After it, it costs money, which is why the two approvals either side of it are the ones nobody should be able to skip when a sponsor is in a hurry.
On the second of those two gaps, the one between L4 and L5: most programs declare victory at L4 because that is when the work visibly stops, and it is exactly the point at which adoption, measurement and the change in actual cash flow are still ahead. Holding the distinction is what lets a finance function believe the number at the end.
Most governance arguments are an unspoken disagreement about the ambition, which is why they go in circles. Settle what you are actually trying to do and the structure stops being a matter of taste. This asks where your governance is today, where you want the business to get to, and what the gap between those two honestly requires.
How is AI governed in your company today?
Has any department in the company actually changed how it works, with a before and after somebody measured?
How far do you want to take integrating AI into the business?
By when?
How many initiatives would be moving at once at the peak?
Who owns this across the company today?
How much do you expect to invest in building initiatives over the next twenty-four months?
Could you say today what any of it has returned?
Answer each one to see the read.
Tell us who you are and it appears below, yours to use with or without us. If you do book a call afterwards, we open on your ambition rather than on discovery.
Everything else on this page depends on four answers that only your leadership team can give. How far do you want to take this. By when. How much are you prepared to invest. And how much governance do you actually want. Settle those together and the structure, the sequence and the money all fall out of them. Leave them unsettled and every one of those gets decided piecemeal by whoever is loudest that quarter.
In practice this usually runs as a session or two with the leadership team, because these are not questions anyone answers well by email. You come out with the ambition written down, the lightest structure that would carry it, and the order of what happens next.
One caveat that decides whether this page is even for you yet. The answer to how far you want to take it is not a fixed fact about your company. It moves with what you have proved, and it should. If no department has yet changed how it works with a before and after somebody measured, then the ambition is currently a judgement about what AI can do rather than about what it has done here, and organizations treat those two very differently when real money is involved. It will lose every budget argument it has with something that already works.
If that is you, the honest advice is to go and get one department win first. It takes a quarter rather than a year, it costs a fraction of what a program costs, and it changes the conversation more than any amount of planning will, because afterwards somebody can point at a result. That is the Keep up situation, or Start if nothing at all has begun. Come back to this page with the number and the same four questions get much easier to answer honestly. The ambition tool above asks this first and will tell you if it thinks you are early.
Assuming the ambition is decidable, there are two ways to go, and we will do either.
We stand up the office and operate it: the intake, the gates, the tracker, the cadence, the monthly close and the reporting that says what the program has returned. What we are responsible for is that the process runs cleanly and on time, and that nothing gets funded without a case or closed without a check. The initiatives themselves belong to the business units that own them, and so do their results.
Monthly, ongoing, cancellable. You build it and you run it. Philippe takes a place at your operating committee, a few named executives get direct access between meetings, and business cases and vendor proposals get reviewed as they arise rather than after they are signed. The decisions stay yours.
The mismatch runs both ways, and both are expensive. A company with six initiatives a year does not need an office, and a company attempting one hundred and fifty will not survive on a spreadsheet and good intentions. Getting that match right at the start is most of the value, and we are just as happy telling you the answer is lighter than you feared. The ambition tool above is the self-serve version of the first conversation in that session.
Not things to buy now. This is the shape of the work once the ambition exists, in the order it normally happens, so that you can see where the first step actually leads.
Go wide across the business and come back with a list of initiatives worth considering, each one at L0 with a rough value on it. You cannot sequence a pipeline you have not written down.
Not all of them. The ones where the sweep found enough to justify going deep. That is the Keep up diagnostic, and it is what turns a rough value into a number a finance function will accept.
How much build capacity the resulting queue needs to move at the rate you want, and how much of it comes from inside. That is the Go faster question, and you can put rough numbers through it now.
Value against effort gives the order, the gates give the discipline, and the roadmap gives everyone whose initiative is not first a reason they can accept.
Which is a year of work described in four lines, and none of it is worth starting before the first step, because each one of those four requires knowing what you are aiming at.
A well-run operating framework spread across five disconnected tools, with the same figure appearing in several of them under two different definitions. A clerk that reads the sources, keeps one record and writes nothing without approval.
Read the caseA high-growth manufacturer needed leadership measures that meant the same thing in every plant before it could manage anything centrally.
Read the caseAn Integration Office set up around your initiatives: a charter with decision rights, stage gates, a business-case template and intake, one tracker where every initiative lives, an operating cadence, and, as the program grows, a dashboard with a monthly close and an integration profit and loss view. Then a pipeline of departments run through the gates.
Thirty minutes is enough to see whether setting the ambition together or an advisory seat is the right way in.