Two Surveys, One Contradiction: What PE's AI Numbers Are Really Telling You

Two credible reports published in the first half of 2026 contain numbers that cannot both be describing the same reality.

MIT's NANDA initiative found that 95% of enterprise generative AI pilots deliver no measurable P&L impact, across $30 to $40 billion in enterprise spending. McKinsey's global 2025 survey found that although 88% of organizations now use AI somewhere in the business, only 6% qualify as "high performers" extracting 5% or more EBIT impact from it.

FTI Consulting's 2026 Private Equity AI Radar, surveying 200 fund and operating leaders, found that 95% of funds report their AI initiatives meeting or exceeding their original business case.

Read those two 95% figures side by side. One says AI is failing almost everywhere. The other says private equity is the rare exception, succeeding almost everywhere. Both cannot be fully true. The more useful question isn't which report is wrong. It's why a fund would report success on a scoped-down business case and call it a win.

FTI's own report answers this: those cases were often conservatively scoped from the start. A self-graded pilot with modest goals will succeed almost by definition. That is not the same as building a compounding advantage over the fund down the street, and LPs are starting to price the difference. Private Equity International's 2026 LP Perspectives survey found that 47% of LPs are already closely monitoring how their GPs adopt AI. A separate industry survey of limited partners found 99% want GPs using AI in dealmaking, with two-thirds specifically expecting it in due diligence.

So what actually predicts durable advantage instead of a flattering pilot report? Not deployment count. FTI's own data points to the real constraint: 35% of funds cite talent, not technology, as the primary barrier to scaling AI past the pilot stage. Allvue's 2026 GP Outlook adds the operational half of the story: most GPs still run core workflows on fragmented, manual, Excel-based systems, the exact condition MIT identified as the reason pilots don't compound into anything durable.

The firms worth watching in 2026 will not be the ones claiming a 95% success rate on a narrow pilot. They will be the ones who can show an LP, inside a diligence questionnaire, exactly what their AI has learned across a hundred deals that it did not know after the first ten. That is a different, harder, and far more defensible claim. It is also the only one that will still be true in three years.

Self-reported success is a marketing number. Compounding judgment is a moat.