AI Strategy
From prototype to production: why most AI pilots never ship
The demo works. The pilot impresses. Then nothing happens. The gap is not the model — it is everything around it.
Every organisation we meet has a folder of AI prototypes that worked and a production estate that contains none of them. The pattern is consistent enough to name.
The demo is the easy part
A prototype proves that a model can do a thing. Production requires that it does the thing reliably, at cost, under load, with the right data, inside existing systems, with someone accountable when it fails. None of that is model work.
Four gaps that kill pilots
- Data access. The prototype used an export. Production needs a live, governed connection.
- Evaluation. Nobody defined what "good enough" means, so nobody can sign off.
- Integration. The output has nowhere to go that changes a decision.
- Ownership. The team that built it is not the team that will run it.
What we do differently
We start every engagement by designing the production path — data contracts, evaluation harness, integration points and operating model — before the first prompt is written. The prototype is built to be promoted, not thrown away.
AI that works in a notebook is a research result. AI that works on a Tuesday afternoon with real users is infrastructure.
More insights
Designing agents with a boundary around autonomy
Autonomous agents are only useful if you can trust what they will not do. The boundary is the product.
Language-first design for citizen services
If a service only works in two languages, it only works for some citizens. Voice and translation change what is possible — if you design for them from the start.
Sovereign AI: what it actually requires
Sovereignty is not a slogan. It is a set of engineering constraints that most AI architectures silently violate.