Insights

Why AI pilots stall after they work

A pilot proves a tool can do a task. It proves nothing about whether an organisation can absorb it.

  • Ravi R. Singh
  • 4 min read

The week after the demo

The most dangerous moment for an AI project is the week after the pilot succeeds. The demo went well. The team that ran it is proud of the result. Then the project meets the rest of the organisation, and it slows to a stop.

The evidence says this is the normal outcome. MIT's NANDA initiative published The GenAI Divide: State of AI in Business 2025 in August 2025. As Fortune reported it, about 5% of AI pilot programmes achieve rapid revenue acceleration. The rest stall and deliver little to no measurable impact on the profit and loss account.

Gartner made a similar call in July 2024. It predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. The reasons it gave were poor data quality, inadequate risk controls, escalating costs or unclear business value.

Read that list again. None of the four reasons is about the model. Each one describes the organisation around it.

A pilot answers a small question

A pilot answers one question: can this tool do this task. That is a useful question, and it is a small one. A pilot runs with volunteers, clean sample data, a patient sponsor and no pressure from the real business. It is a laboratory.

Production asks harder things. It asks who owns the output when it is wrong. It asks which step in the existing process disappears, and whose job changes when it does. It asks what the finance team will accept as a saving and who approves the data the tool reads.

I have spent more than 20 years taking markets and businesses from zero to revenue across India and Southeast Asia, and the pattern is familiar from earlier technology waves. The tool arrives ahead of the operating practice. The gap between the two is where projects go quiet.

What a stall looks like

A stalled pilot is rarely cancelled. It gets a second pilot. A new team tries a different tool on a different task, and the first one sits in a shared drive as a success story nobody uses. After a year the organisation has six pilots and no change in how work gets done.

The cost is larger than the licence fees. Each pilot that goes nowhere teaches staff that AI projects are theatre. The next request for their time gets a slower answer, and the people who volunteered first are the least likely to volunteer again.

Leadership often misreads the situation. Six pilots look like momentum in a board paper. In practice they are six unanswered questions about ownership, measurement and process.

Four things a pilot does not give you

Moving from pilot to practice needs four things that a pilot never produces by itself.

The first is a ranked list of where AI should help, with an owner against each item. Most organisations start with whichever idea had the loudest sponsor. The second is a business case in numbers that finance has agreed to, including what will be counted and when.

The third is a redesigned process, with the changed roles written down and the affected people involved early. The fourth is a regular check, months later, on whether the gain is still there.

These map to four stages I call Start, Think, Work and Grow. Start decides where AI should help first. Think tests whether it is worth doing, in numbers. Work makes the change stick. Grow checks that it is still paying off. An organisation can enter at whichever stage it has reached, and each stage reuses what the earlier ones found.

Why the gap is wider across markets

Regional businesses face a harder version of this problem. Many run lean head-office teams, with operations spread across several countries, languages and systems. A pilot built in one market on one team's data meets different invoices, different approval habits and different regulations the moment it crosses a border.

I have led multi-market teams, and the work of making one process behave the same way in three countries was always larger than the technology. Budget for that work when the pilot is approved. Leaving it until the pilot succeeds is how the stall begins.

What to do with a pilot that went quiet

If you have a pilot that worked and then went quiet, do not commission another one. Take the one you have and write three things on a single page. Name the person who owns the result in production. State the number that would prove it is working and the date you will check it. Describe the one process step that changes.

If you cannot fill in the page, you have found the reason the pilot stalled. That is useful. It tells you the next piece of work is organisational, and no new tool will do it for you.

The organisations that get past this point treat the pilot as the start of an operating change. They budget time for the dull parts, which are ownership, measurement and process. The model was never the hard part.

Sources

See the four advisory offersAll articles