AI pilots stall for predictable reasons. They are run by enthusiasts in ideal conditions, with no plan for what happens if they succeed, no budget for rolling out, no agreed decision point and no business owner ready to take them on. The result is "pilot purgatory": projects that worked well enough to continue but never well enough, or clearly enough, to scale.
Six reasons pilots get stuck
1. Designed to demonstrate, not to scale
The pilot used hand-picked users, clean data and extra support. Real rollout would face ordinary users, messy information and no extra help. Leaders sense the gap and hesitate.
2. No decision point
Nobody agreed when the pilot would end or what result would trigger rollout. So it simply continues.
3. No business owner
The pilot was run by an innovation team or IT. When it worked, no operational leader was ready to own it day to day.
4. No rollout budget
The pilot was funded, but scaling needs more licences, training and possibly integration. Nobody budgeted for success.
5. Unclear results
Without a baseline, the pilot's results are anecdotal. "People liked it" is not enough to justify wider investment.
6. Change was not planned
Scaling means changing how many people work. That needs training, communication and process updates, which were never part of the pilot plan.
Designing a pilot that can scale
Build these into every pilot from day one:
| Element | Question to answer before starting |
|---|---|
| Business owner | Who will run this as normal work if it succeeds? |
| Realistic conditions | Are we testing with typical users and real data? |
| Baseline | What are today's time, quality and cost figures? |
| Success criteria | What result triggers rollout? |
| End date | When will we decide? |
| Rollout plan | What would scaling need: licences, training, integration? |
| Rollout budget | Is it provisionally approved if criteria are met? |
What to do with pilots already stuck
Review each stalled pilot against three options:
- Scale it: assign an owner, set rollout criteria based on results so far, and fund the next phase.
- Re-run it properly: if results are unclear, run a short, structured second phase with a baseline and end date.
- Stop it: if it has not shown value, end it, record the lessons and free up the resources.
Stopping a pilot is not failure. Leaving it in limbo is.
The leadership factor
Moving from pilot to scale is mainly a leadership challenge: making decisions, securing budget, assigning owners and leading change. Our Leading AI Innovation & Organisational Transformation course covers turning successful experiments into standard ways of working and sharing them across teams. You may also find how to start an agentic AI pilot useful for pilot design principles.
Frequently asked questions
How long should an AI pilot run before deciding?
Usually 60 to 90 days. Long enough to see real patterns, short enough to keep pressure for a decision.
What percentage of AI pilots should scale?
There is no target. A healthy programme stops some pilots and scales others. What matters is that every pilot ends with a clear decision.
Who should decide whether a pilot scales?
The business owner and the executive sponsor, using the success criteria agreed at the start.