10 Operating Model Signals That Your AI Is Producing Growth, Not Just Activity
Most firms can show AI activity. Far fewer can show what it returned. The operating model is where the difference becomes visible.
- Boards have stopped asking whether AI works. They are asking what it returned, and the answer is often thinner than the original case suggested.
- The ten signals below describe what AI looks like when the work has actually changed around it — and what it looks like when the tools went in and the work stayed the same.
- None of the ten are technical. All of them are about how the business runs. That is the point: technology speeds up whatever it finds, so if the work did not change, the spend bought speed rather than return.
The question every mid-market board is now asking — what did we get for the AI spend? — is uncomfortable because the honest answer is usually partial.
The pilots demonstrated capability. The investment was material. The board paper made a reasonable case. But the conversation about what actually came back, twelve to eighteen months on, is harder to defend than anyone expected.
The problem is not that AI failed to deliver. It is that most firms cannot cleanly separate AI that is producing commercial return from AI that is producing activity. Both look busy. Both have champions. Both generate dashboards and case studies and enthusiasm. Only one shows up in the numbers a finance director recognises.
This matters now because the next round of investment is being shaped. Boards are increasingly unwilling to fund more pilots without evidence that the last ones paid. The conversation has moved from proving the technology works to proving the money worked.
The ten signals below are diagnostic rather than theoretical. Each one is recognisable in your own business if you look for it.
Each use case has a named commercial number it moves — not a technical one
The portfolio review opens. Each use case is presented. The first slide shows model performance: accuracy, precision, recall, latency. The metrics are technical.
Where AI is producing return, the first slide shows the commercial number instead. The conversion lift. The retention lift. The reduction in fraud loss. The hours taken out of handling time. The technical metrics are still there, supporting the commercial conclusion rather than standing in for it.
The test is easy. Open your next review and count the slides leading with technical numbers against those leading with commercial ones. If technical dominates, you are looking at activity. The fix is to name the commercial number when the use case is designed, not to reach for one at review time.
Someone senior owns the whole portfolio, not just the delivery of it
The review has a presenter. Usually the head of data, the analytics director, or the programme manager. Their job is delivery, not deciding where the money goes.
Where AI is producing return, an executive owns the portfolio commercially. They report to the executive committee, they present to the board, and they are accountable for what the portfolio returned rather than whether it shipped.
This single appointment is the clearest structural difference between firms whose AI pays and firms whose AI is merely active. Without it, the portfolio is a collection of departmental projects. Most mid-market firms have not made the appointment, because departmental sponsorship has not yet produced a visible failure — only quiet underperformance that nobody owns.
The way the business runs has actually changed because of AI
The test is direct. In the last twelve months, what have you decided to do differently because of what AI can now do? Different decision rights. Different handoffs. Different jobs. Different measures on the Monday report.
If the honest answer is “training and some process updates”, the technology has been bolted onto unchanged work. The capability exists. The business runs exactly as it did. That is the definition of technology without transformation, and it is the most common outcome in the mid-market.
Where AI is producing return, the changes are visible and specific. The service team works differently because first-contact resolution changed. Pricing works differently because the decision moved. The fix is to design the change to the work as part of each use case, not to hope it follows the tool.
A board can no longer be impressed by AI activity. The question now is what it returned — and the way the work is done is where the answer shows up.
The uplift is visible in a cohort, not just at programme level
The review presents impact. Usually at programme level: AI is driving a twelve per cent improvement in something across the business. The number is credible and the board accepts it.
Where AI is producing return, the impact is shown at cohort level. Customers exposed to the capability show measurable movement against a comparable group who were not. The channel where AI is live behaves differently from the channel where it is not.
Programme-level claims can be defended for a quarter or two by other factors. Cohort evidence cannot be explained away as easily. The fix is to set the measurement up before the build, with a comparison group and an agreed method, rather than assembling it afterwards to answer a board question.
The money sits in the operating budget, not only in project capital
The spend is approved and in the budget. The question is which budget.
Where AI is activity, it sits in project or transformation capital. It is funded once, for a one-time outcome. Afterwards the capability either survives on almost nothing or gets quietly dropped when budgets tighten.
Where AI is producing return, the money is in the operating budget. The people who keep model performance honest. The function that watches what is running in production. The retraining. Working AI is infrastructure with continuing cost, not a project with a closure date. The fix is to budget it that way when the use case is designed, rather than hoping it finds a home in operating expense after delivery.
Recognising the difference is the first step. Knowing which one you have is the next.
A free Executive AI Opportunity Assessment. Forty-five minutes, online, with your leadership team. We arrive with a model of how firms in your sector make money and where AI has moved those numbers — and you tell us where it is wrong.
Book a free assessmentAnything a customer sees can be explained, governed and audited
Customer-facing AI is exposed to scrutiny. Customers see the output. A regulator may ask about the input. A partner may want documentation.
Where AI is activity, this was built to demo standard. It works when it works. When it does not, the technical team can describe the model but nobody can explain the decision to someone who is not technical.
Where AI is producing return, it was built to enterprise standard from the start. Explainability designed in, documentation current, decision trails kept, fairness assessed. It was built as though someone will ask about it next week, because increasingly they might. The fix is to set the governance requirement when the use case is scoped, not when a complaint arrives.
Pilots inside departments. No portfolio view. The work is unchanged. Progress reported through dashboards.
Pilots scaled and the technology proven. Some commercial measures in place. The work is still largely unchanged.
Use cases producing measurable movement in a cohort. The work has been redesigned in parts of the business.
AI is part of the operating model design. Returns compound. The portfolio stands up to investor scrutiny.
Most UK mid-market firms sit between stages one and two. The move from two to three is where AI starts to pay, and it is where most firms stall — because it is the point at which the work itself has to change.
AI sits in the customer’s experience, not only in the back office
The capability exists. The question is where it lives.
Where AI is activity, it lives in internal tools. The marketing team’s recommendation engine. The service team’s triage assistant. The finance team’s variance checker. Real and useful, but internal.
Where AI is producing return, the customer meets it. They see the recommendation. They experience the pricing. They get the faster answer. Customer-facing AI moves commercial behaviour directly. Back-office AI produces efficiency, which is worth having but is not the same thing. The fix is to design use cases against points in the customer journey rather than points in an internal process.
Three questions that separate return from activity
- For each use case, can you say in one sentence which commercial number it moved, by how much, and who is accountable? If not, it is probably activity.
- What have you decided to do differently in the last twelve months because of what AI can now do? If the answer is “training and process updates”, the work has not changed.
- If you stopped a third of the use cases tomorrow, would any commercial number notice within ninety days? If probably not, you have activity.
The business owner is accountable for the outcome, not the data team
The use case has an owner, usually inside the data function. The data scientist owns model performance. The engineer owns the pipeline. The product manager owns deployment.
What is rarely named is the business executive accountable for the commercial result the whole thing was built to produce. Nobody holds the sales director to the AI-driven sales lift, or the service director to the retention improvement.
Where AI is producing return, that accountability is explicit. The business executive owns the tool and the number. The data function is a service provider, not the owner of the outcome. The fix is to name the accountable executive when the use case is designed rather than leaving it implied by the org chart.
Money can be moved between use cases when the evidence changes
The portfolio has eight or twelve or fifteen use cases. Some are doing better than others. The question is whether the firm can act on that.
Where AI is activity, use cases are committed bets. Once funded they continue. Stopping one is structurally awkward: no process, nobody accountable for the decision, no route to move the money somewhere better.
Where AI is producing return, the portfolio is actively managed. Use cases stop when the economics do not appear. Money moves to the ones that are working. The composition changes quarter to quarter, not as a failure but as ordinary capital allocation — the same discipline a finance director would apply to any other portfolio.
AI is part of how the business is being redesigned, not a parallel project
The most structural signal of the ten. The firm has a change programme. It also has an AI programme. Sometimes they talk. The change programme designs how the business will run. The AI programme deploys technology. Different timelines, different governance, different measures.
Where AI is producing return, they are one thing. The design of how the business will run includes what AI will do inside it. The AI portfolio is shaped by those decisions rather than running alongside them.
This is the difference that matters most. AI is not something the operating model accommodates. It is part of how the operating model produces value. The fix is to design it in from the start, rather than run it adjacent and hope the two meet.
What this means
None of these ten signals are technical. All of them are about how the business runs. The difference between AI that pays and AI that is merely busy is not which models were deployed or how good the data science is. It is whether the work changed around the technology.
What ties them together is integration. Where AI produces return it is joined to the commercial numbers, to how decisions get made, to the customer’s experience, to who is accountable, to the operating budget, to governance, to how money gets allocated. Where it produces activity, it sits next to all of those — connected by reporting lines but not by design.
The economics favour doing this at the start. Building the commercial case and the change to the work into each use case costs far less than retrofitting both across a portfolio that has already been built the other way.
Which is the whole argument, in the end. Change the work before you change the technology. The technology will speed up whatever it finds — so it is worth being certain about what it finds.
Is your AI producing a return, or activity that looks like one?
Forty-five minutes with your leadership team, online, at no cost. We bring a working model of how firms in your sector make money and where AI has moved those numbers. You tell us where it is wrong. Then we tell you what we think — including “don’t invest”, if that is the honest answer.
Book a free assessment45 minutes · online · no cost · nothing to prepare
Continue reading
9 Signs Your AI Programme Is Creating Governance Risk Instead of Value
The risks building up inside AI programmes that still report green at board level.
For chief executives12 AI Readiness Gaps Boards Are Beginning to Question
The gaps boards now expect leadership teams to have an answer for.
For chief executives10 Reasons AI Pilots Stall Before They Scale
Why pilots succeed and then go nowhere in UK mid-market firms.