A pilot is built to impress a meeting. Production has to survive bad data, angry customers, auditors and a monthly bill. Most AI budgets are spent on the first and surprised by the second.
RAND notes that, by some estimates, more than 80% of AI projects fail, twice the rate of IT projects without AI (RAND, 2024). Gartner predicted at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, weak risk controls, rising costs and unclear business value (Gartner, 2024).
India shows the same gap between trying and shipping (EY, 2025):
Indian enterprises and GenAI, January 2025
Where the demo hides its cost
RAND interviewed 65 experienced data scientists and engineers and found five leading root causes, including solving the wrong problem, missing data, chasing new technology over user needs, and weak infrastructure (RAND, 2024). In my experience, each one shows up as a cost line the pilot never had:
| Pilot shortcut | Production cost |
|---|---|
| Twenty hand-picked examples | Cleaning real data, often for months |
| No customer data involved | Consent, masking and a DPDP review |
| One happy path | Fallbacks, retries and human review for edge cases |
| Free credits | A per-call bill that grows with usage |
| A demo laptop | Logging, monitoring and an on-call owner |
Run the pilot like it's already live
Before you approve a pilot, check it scores 2 (yes) on each, not 0 or 1:
- Problem: a business owner signs off what "success" means, in numbers.
- Data: the pilot uses a messy, real sample, not curated examples.
- Security: the data path is reviewed before the first call, not after.
- Edge cases: failures have a fallback and a human to catch them.
- Cost: the monthly bill at full volume is estimated up front.
| Score | Next move |
|---|---|
| 9–10 | Fund the pilot, with a date to decide go or stop. |
| 6–8 | Fix the gaps first; they're cheaper now. |
| 0–5 | Don't start. You'd be funding a demo. |
Your next step
Take your current or next AI pilot and add one line to its plan: "the production version will cost ₹___ a month and needs ___ to approve its data use." If nobody can fill in those blanks, the pilot isn't ready.
Sources
- RAND (2024). The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed. 13 August 2024.
- Gartner (2024). Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025. 29 July 2024.
- EY (2025). The AIdea of India: 2025. EY India, 14 January 2025.