AI Strategy
Moving beyond random AI pilots
Why so many organisations have a portfolio of pilots and nothing in production, and how to fix it.
By Whitesoft · · 7 min read
The pattern is common enough to have a shape. Enthusiasm from the top, a call for ideas, a dozen pilots across the business, a few impressive demos, and eighteen months later nothing in production and a general sense that AI was oversold. It was not oversold. It was unmanaged.
Why pilots stall
- No pass mark. The pilot was never given a definition of success, so it cannot succeed or fail. It simply continues.
- No owner of the outcome. Technology teams own the pilot; nobody owns the business result.
- No path to production. Security, data access, integration and support were never scoped, so the pilot cannot leave the sandbox.
- Chosen by enthusiasm rather than value. The loudest idea won, not the most valuable.
- Each pilot rebuilt the same foundations, so none of them could afford to finish.
Run a portfolio, not a collection
Treat AI opportunities as a portfolio with a shared scoring model. Every candidate is assessed on the same criteria: expected value, complexity, data readiness, feasibility, organisational readiness, risk and time to value. Plot them. Fund the ones in the high-value, manageable-complexity corner first. Deliberately park the rest, and say why.
Give every pilot a pass mark and an end date
A pilot exists to answer a question. Write down the question, the threshold that would justify production investment, and the date by which the answer is due. When the date arrives, decide: go, adjust or stop. Stopping is a legitimate and valuable outcome.
Build foundations once
Data access, identity, model gateway, evaluation and observability should be shared services, not pilot-by-pilot inventions. The first production system pays for them; every subsequent one gets cheaper and faster. This is where cloud and platform engineering earns its keep in an AI programme.
Make someone accountable for the outcome
A business owner, not a technology team, should own the metric the pilot is meant to move. If nobody in the business wants to own it, that tells you something about the opportunity.