Systems integration
The productivity gap is a systems problem, not a technology one
McKinsey's November 2025 survey found 78 percent of companies using AI regularly, but only 33 percent scaling it across the organisation. MIT NANDA's much-cited figure — around 5 percent of pilots reaching lasting financial gain — points the same way, though its method has been contested and it is treated here as directional, not measured.
Executive summary
78%
of companies use AI regularly, per McKinsey (Nov 2025)
33%
are scaling it across the organisation — the same survey
~5%
of pilots reach lasting financial gain, per MIT NANDA — a contested, non-peer-reviewed figure
Core conclusions
- Adoption is now near-universal; scaling is rare — and since the technology is identical in both groups, the gap points at what surrounds it, not the AI itself.
- What separates scaled deployments from stalled pilots: a use case chosen for real friction, data on production terms, a pre-go-live baseline, and named month-thirteen ownership.
- A stalled pilot leaves behind an unmaintained connector and a diverging data copy — residue that makes the second attempt more expensive than the first, not less.
Getting an organisation to try AI is no longer the hard part. The distance between trying it and running it across the business is where the value is either captured or quietly lost.
The McKinsey pair describes the gap from both ends: adoption is near-universal; scaling is rare. The MIT NANDA figure is often quoted alongside it and deserves a caveat in the same breath. It comes from a small sample — on the order of 150 survey responses and 50 interviews — was not peer reviewed, and counted a pilot as failed if it showed no measurable P&L impact within six months, a definition that has drawn substantive methodological criticism since publication. It is cited here as a directional indicator that agrees with the larger survey, not as a measurement in its own right. The argument below rests on the McKinsey gap, and holds without NANDA's number.
That pattern does not point at the technology, which is identical in both groups. It points at what surrounds it.
What separates the two groups
| Stuck at pilot | Scaled |
|---|---|
| Use case chosen for how impressive it demonstrates | Use case chosen for where the operational friction actually is |
| Data assembled specially for the pilot | Data available on the same terms the production system will have |
| Success measured against the demo | Success measured against a baseline established before go-live |
| Ownership ends with the project team | Ownership named for month thirteen, and priced |
Why waiting costs more
Every pilot that stalls leaves behind an integration that half-exists: a connector nobody maintains, a copy of the data that has begun to diverge, a process with a manual step reinserted to work around it.
That residue is not free. It is the estate the next project has to work through, and it is why the second attempt is usually more expensive than the first rather than less.
