Accelerate & Transform
The AI spend has not reached the P&L
What does AI return once the full cost of the work it replaces is counted?
What does AI return once the full cost of the work it replaces is counted?
Nine in ten companies use AI in at least one function and fewer than half have scaled it across the business, by McKinsey's 2026 count; about a third report an effect on EBIT. In the company this reads as a budget line that has tripled in two years, a hundred pilots, a dozen in production, a CFO who asks at each review where the money shows up. The answer is a productivity figure measured on the people who use the tool well, which is not the whole workforce, set against a cost of licences, integration, data work and governance that has not been added up in one place.
Underneath, the legacy is doing what legacy does. The AI runs on systems that cannot log what it decides, cannot trace a product-level carbon number, cannot expose the data a regulator will ask for. Each new use case pays the technical debt twice: once to build around it, once to explain why the result cannot be audited.
Which use cases deserve to scale, which to stop, what has to change underneath for the first group to pay.
Whether to invest in AI was settled long ago; the decision is about the arithmetic: the full cost of the work a use case replaces, including the people who now operate the tool and the ones who cannot, set against the full cost of the use case, including the data, the governance and the legacy it drags. Done honestly, that arithmetic usually kills part of the portfolio, scales a handful of use cases and names the few systems whose retirement would change the return on the rest.
The work sits in what we call AI, Digital & Technology. It is an audit, and it ends in decisions where a strategy would end in a roadmap.
We map the deployment posture: every tool in use, the use cases behind it, what each actually returns measured on the whole population that touches it, not just the champions, then what each costs once data, integration, governance and licences are counted. We audit the technical debt with the same eye: what the legacy stops the company from doing, including the responsible performance capabilities it cannot support, from product-level carbon tracking to AI decision logging to reporting that does not need a quarter of manual work. For the CFO this produces the bridge between the AI budget and the accounts; for the CIO the build, buy or retire decision for each system with the number attached; for the COO the places where data, and not strategy, is what limits decision speed.
Then we decide with you: the use cases to scale, with the workforce transition designed into each, because a tool that part of the team cannot operate returns that much less than the business case says; the use cases to stop; the systems to retire and the sequence that funds the retirement from the gains. For the CHRO the transition is designed in full: roles redrawn, the capability programme sized, the people involved from the start.
What leaves the room is the portfolio decided, the legacy decisions with their cases and the transition plan for the people. The audit is short; the retirements take the time the systems take. We stay for the first ones.
In the case that follows we show that the failures share a route rather than a technology: a build where a buy would have done, a buy where no tool was needed, a pilot scaled before the data could carry it. The audit reads the company's portfolio against that path before the next budget is approved.
95% of enterprise AI initiatives fail to deliver measurable ROI. Only 31% of AI use cases have entered full production by 2026. The gap is not a technology problem.
ReadFabrice Macarty, Founder, CEO. Twenty-five years across Google, Microsoft and Deloitte, from technology to strategy, in financial services, energy, defence, retail, industry and the public sector. He works with CIOs and CFOs on the business models that technology is about to move. On this situation he holds all of it: the arithmetic of the portfolio, the legacy decisions and the value architecture that lets the company keep what the AI produces.
In the Harvard and BCG field experiment, the same AI raised the quality of work by 40% on tasks within its reach and cut performance by 19 points outside it. The return is decided by who can operate the tool; the licence count decides little.
Read the force in Our Approach: force 03The budget that tripled, the hundred pilots, the question the CFO asks at each review. Tell us where it binds and we will tell you where to start.