The Pilot Trap: Why Some Indian Enterprises Move AI Into Production and Others Don't

Beyond the Hype: EY Survey Benchmarks

EY India's page on its generative AI survey, based on over 200 Indian enterprise leaders surveyed in 2025, reports that 76% believe generative AI will have a significant impact on their business. On maturity, 23% were at the early stage or running pilots, about 47% had multiple use cases live in production and 10% were scaling use cases across the business. Nearly half (47%) said more than 10% of their proof-of-concept projects had moved into production.

The Conversion Gap: Navigating Pilot Purgatory

The survey also asked about jobs. According to the page, 64% reported only selective displacement, concentrated in outsourced and standardised functions such as administrative operations, customer success, tele-calling and back-office work, and the findings suggested spending was being redirected toward automation and efficiency rather than shrinking internal teams.

On the choice between buying and building, 91% ranked speed of deployment as the top factor. The page we read did not report governance gaps or barriers to adoption; those may be covered in the full report, which we have not reviewed. (The page's metadata carries a different first-publication date from its displayed date of 16 November 2025, so cite it as a 2025 survey.)

“A pilot proves an idea can work. Production proves the business can.”

Reading the Numbers: What the Adoption Curve Shows

Two features stand out. The first is that moving from pilot to production is no longer rare. With about half the sample reporting multiple live uses, adoption has passed the experimental stage for many large organisations.

The second is that conversion remains modest: nearly half say more than 10% of proofs of concept reach production, which implies that a majority of experiments do not. That is not necessarily a failure, because testing ideas is meant to filter. But it suggests that a company's real advantage is in choosing and finishing the right projects.

What Separates Scaling Projects: Our Six-Point Framework

The survey does not tell us why some succeed, so what follows is our editorial analysis, drawn from the shape of the data and common practice.

1. A named business owner. Pilots led only by a technology team often lack a user who needs the result. A project with a head of sales or operations who wants it tends to survive.

2. A narrow, measurable first use. Reducing the time to answer a customer query is easier to measure than "improving service". A clear metric allows an honest decision to continue or stop.

3. Clean data. Many projects stall because the data is scattered or poorly labelled. Fixing that is dull but decisive.

4. A plan for the people affected. If staff expect their jobs to disappear, they will not help. The survey's finding of selective displacement suggests that most employers are redirecting work rather than removing it, and saying so clearly helps.

5. A budget for running costs. The pilot is cheap; the production system needs monitoring, support and updates.

6. Speed with guardrails. The 91% who prize speed are right that waiting is costly, but a model that gives wrong answers to customers damages trust.

The Buy vs Build Strategic Calculation

With speed as the leading factor, many firms will buy tools from vendors rather than build their own. That is reasonable for common tasks such as document summarising or support chat. Building makes more sense where the data or process is unique and gives a competitive edge. A mixed approach, buying the general layer and building the specific one, is common.

For mid-sized and smaller firms: The survey covers enterprise leaders, so its results do not apply directly to a 50-person company. Still, the lesson carries over: start small, choose a task with a clear measure, and finish it before starting another. Many small firms get value from simple uses such as drafting quotations, answering routine customer questions and organising documents.

What to Watch & Pre-Approval Decision Filters

What to watch: The next reports should show whether the share of firms scaling across the business rises above 10%, and whether governance catches up with use. Both will say more about the staying power of this wave than the headline adoption rate.

Questions to ask before approving the next AI project: What decision or task will change, and who will do it differently? How will we measure the result in three months? What data does it use, and who owns it? What happens when the system is wrong? What will it cost to run each year? If a proposal cannot answer these in plain language, it is probably not ready to leave the pilot stage.

Editorial analysis. Survey figures are EY's as presented on its web page; our reading is labelled as such.