Share with your CIO
Most enterprise AI pilots die quietly, and Ashish Chandra, a global AI thought leader, argues the cause is almost never the technology. In a recent Analytics Insight podcast episode, Chandra puts the failure rate at roughly 80% and traces it to four compounding problems: fragmented data, siloed systems, weak governance, and AI initiatives that were never tied to a measurable business outcome in the first place. His prescription is structural, not technical, which puts the accountability squarely on enterprise leadership rather than the AI vendors.
What this means for your business
The 80% figure is doing real work here. Organizations that have greenlit multiple AI pilots this budget cycle are either on the right side of that number or they’re not, and the dividing line Chandra draws is whether the initiative was attached to a business outcome before the first model was trained. Companies still running disconnected proof-of-concepts, with data that doesn’t flow between systems and no governance layer to enforce quality, are spending money to learn what they already could have known.
Chandra’s framing of “AI Theatre” is the sharper insight. AI Theatre is the pattern where a company demonstrates AI capability internally, generates executive enthusiasm, and produces no change in how work actually gets done. It’s the enterprise equivalent of a product demo that never ships. The recurring failure mode looks like this: a pilot succeeds on a narrow dataset curated for the experiment, earns a slide in a board deck, and then stalls when it meets the messy reality of production data and cross-functional workflows. Chandra’s argument, and he’s right, is that governance and data quality aren’t prerequisites you clear before the interesting work begins. They are the work.
The budget decision this reframes isn’t whether to fund more pilots. It’s whether your current AI portfolio has any pilots that could survive contact with your actual data infrastructure. If the honest answer is no, then the next AI investment worth defending isn’t a new use case. It’s the data and governance foundation that makes any use case stick. I’d revise this assessment if Chandra offered evidence that companies with strong data foundations are actually converting pilots at materially higher rates, but absent that, the structural diagnosis holds on its own logic.
Concept deep-dive: AI governance
AI governance is the set of policies, controls, and accountability structures that determine how AI systems are built, monitored, and corrected inside an organization, roughly analogous to financial controls but applied to model behavior and data integrity. It exists because AI systems can drift, produce biased outputs, or simply be trained on data that doesn’t reflect current business reality. Without it, a pilot that works in month one may quietly degrade by month six, and no one owns the problem.
Based on reporting from Why 80% of AI Pilots Fail: Ashish Chandra on Moving From Experiments to Enterprise Impact, originally published 2026-07-22 12:28:00.
