A redesign project starts in January. The scoping is approved in February, after three meetings. The mockups go to committee in March, come back with contradictory requests in April, go back through in May. The production itself will have taken eleven days.
Eleven days of making, five months of circuit. Nobody in the company finds that abnormal, because it has always worked that way. Except that one thing has just changed, and it changes the whole calculation.
What the study few executives have read actually says
In August 2025, the MIT Media Lab’s NANDA initiative published a report titled The GenAI Divide, State of AI in Business 2025. It draws on executive interviews, surveys of managers and employees, and the analysis of around three hundred public enterprise artificial intelligence deployments.
The figure that circulated everywhere: 95% of pilots produced no measurable effect on the profit and loss account.
The figure is spectacular, and it has been widely repeated, sometimes badly. So let us be clear about what it says and what it does not. It does not say that AI does not work. It says that 95% of the projects launched changed nothing about the company’s result. And above all, it names a cause, which is the most useful part of the report and the least repeated: it is not the quality of the tools that blocks things, it is the learning gap between the tool and the organisation receiving it. The friction, not the technology.
An honest caveat on that figure. We have not consulted the primary report, only its relays, and those relays differ on the exact size of the interview sample. The 95% rate and the analysis of three hundred deployments, on the other hand, are consistent from one source to the next.
The bottleneck was never the making
The same finding comes back, phrased differently, from those who produce.
David Heinemeier Hansson, creator of the Ruby on Rails framework and chief technology officer at 37signals, was asked the question in an interview in August 2026: why do the big software publishers show no visible acceleration when their production capacity has exploded? His answer comes in two parts.
The first: as soon as human teams work together, the bottleneck is rarely implementation, it is human bandwidth and communication. He describes the mechanism bluntly. A product manager, two designers, a director above them, a technical director above that, each wanting to take part in the scoping to justify their presence. That is where productivity dies.
The second: most organisations do not know what they want. They are not limited by their ability to produce, they are limited by ideas, vision and taste. His formula hits home: you can bring a great many bad ideas into existence, and then what?
That is a practitioner’s testimony, not a measurement. It is worth what an opinion argued from forty years of craft is worth. What makes it interesting is that it reaches the same conclusion as MIT by a completely different route, and on completely different ground.
The calculation that tipped without warning
Here is why this concerns you even if you do not produce software.
Five years ago, a heavy approval chain was expensive, but it was proportionately expensive. Making took weeks, approval took weeks, each paid its share. One month of circuit on three months of production was a third of overhead. Unpleasant, not decisive.
When the making drops to a few days, the ratio inverts brutally. Five months of circuit on eleven days of production is no longer an overhead, it is a ceiling. Your competitors with two approvals instead of four do not go 30% faster. They go several times faster, and the gap widens with every project.
That is the real change, and it went unnoticed because it shows up on no invoice line. The cost of silos is no longer an expense. It has become a maximum speed.
What it changes in practice on a communications project
Take an ordinary case. An SME entrusts its website to a web agency, its visual identity to a design studio, its video to a producer, and its SEO to a consultant. Four suppliers, four internal contacts, and a committee that arbitrates.
The classic argument against that arrangement is the hidden cost of coordination and the incoherence of the narrative. It remains true. But a third one is missing, and it is now the heaviest: every additional intermediary between intention and execution consumes the gain the period makes possible. You are no longer just paying an overhead. You are structurally forbidding yourself from moving fast.
The symptom is easy to spot. If, on your last project, the cumulative making time represented less than a fifth of the total timeline, it is not your supplier who is slow.
And if the argument applies to us too
It has to be said, because otherwise someone will say it for us, and rightly.
This argument turns against any agency that merely adds itself to the existing approval chain. If the number of intermediaries is the problem, one more agency is not a solution, it is an extra layer with an invoice.
The difference does not lie in the pitch, it lies in the nature of what is entrusted. A team that takes on a complete scope and returns a decision shortens the circuit, which is exactly what the big groups are trying to reconstitute by merging. A team that inserts itself between you and three other suppliers lengthens it. These are two opposite models bearing the same name.
The question to ask any supplier, ourselves included: how many people will stand between my decision and the work delivered? If the answer is more than two, ask why.
Three things to measure on your next project
None of this is fixed by a reorganisation. Three measurements are enough to see where you stand.
Count the number of formal approvals between launch and going live, then ask yourself, for each one, what it actually changed about the deliverable. Those that changed nothing are pure delay. The same question applies to the document that triggers the project, and a badly written brief costs even more than one approval too many.
Measure the ratio between effective making time and total elapsed time. It is the most honest indicator of your speed ceiling, and nobody ever calculates it.
Identify who actually decides, as opposed to who gives an opinion. A chain where four people give an opinion and nobody settles it produces exactly what MIT describes: a project delivered, no measurable effect.
Approval chains often have a good reason to exist. They protect against a mistake, they reassure a board, they keep a governance alive. That reason deserves to be weighed. What has changed is the price on the scales: what used to cost a third of the timeline now costs half the potential.
