Dun & Bradstreet's 10,000-Business Survey: AI ROI Is Real, but Data Readiness Isn't
Source: Dun & Bradstreet via PR Newswire — 2026-07-28
Summary
Dun & Bradstreet's latest AI Momentum survey of 10,000 businesses worldwide finds enterprise AI has crossed a real threshold on returns: more than three-quarters of companies now report measurable ROI, and the share actively "scaling AI" has climbed to 34%. But the bottleneck has shifted rather than disappeared — only 6% of respondents say their enterprise data is "fully ready" to support AI at scale, with the rest split between "mostly ready" (36%) and "partially ready" (47%). The picture is a maturing AI market whose returns are increasingly capped by data infrastructure rather than model capability or executive appetite.
Key Takeaways
- 76% of surveyed enterprises report some measurable AI ROI: 48% call it "pockets of ROI," 28% call it "broad" or "strong" ROI across multiple projects.
- The share of companies now "scaling AI" (beyond pilots) reached 34%.
- Only 6% describe their data as "fully ready" to support AI at scale; 47% are "partially ready" and 36% are "mostly ready" — leaving the large majority still building the data foundation underneath their AI investments.
- The survey spans 10,000 businesses across 32 countries, making the data-readiness gap a broad, cross-industry pattern rather than a niche complaint.
- The framing implies the next constraint on enterprise AI ROI isn't model quality or leadership buy-in — it's unglamorous data plumbing: governance, lineage, and access controls most organizations haven't finished building.
Reel Script
Hook Three out of four companies now say AI is making them money. So why are only 6% of them ready to actually scale it? The bottleneck isn't the model anymore — it's what's sitting underneath it.
Core Concept Dun & Bradstreet just surveyed ten thousand businesses across thirty-two countries, and the headline number sounds like a victory lap: seventy-six percent report some real return on their AI investment, and a third say they've moved past pilots into actual scaling. But sit that next to the second number and the story flips. Only six percent say their enterprise data is "fully ready" to support AI at scale. Data readiness here means the boring stuff that never makes a keynote slide — is your customer data clean and deduplicated, is it labeled consistently, do you actually know which system owns the authoritative version of a record, can the right people access it without a six-week ticket. Model capability stopped being the constraint a while ago. This is the new one.
Hands-On Break the data-readiness number down and it gets more interesting: 6% fully ready, 36% mostly ready, 47% partially ready. That's not a binary "haves and have-nots" split — it's a long tail of companies that built just enough data infrastructure to get a pilot working, and now that pilot wants to become a program, the cracks show up. A recommendation engine running on a clean sample dataset behaves very differently once it's pointed at the messy, duplicated, inconsistently-labeled production data most large companies actually have. That's the gap between "we saw ROI in a pocket of the business" — the 48% — and "we're seeing broad ROI everywhere" — the smaller 28%. The pocket succeeds because someone quietly cleaned up the data for that one use case. Scaling means doing that unglamorous work everywhere at once.
Takeaway If your AI initiative is stuck in pilot purgatory, the fix probably isn't a better model or a fancier agent — it's paying down the data debt nobody wants to prioritize. Companies that invest in data governance now are the ones who'll be in that 6% instead of stuck explaining why last year's pilot never scaled.