We tried to trace every statistic in this field back to a primary source. Most of the good ones hold up. Four of the most-quoted numbers in corporate innovation do not have a primary source at all.
Graded: A = study with a disclosed sample. B = vendor-published or sample undisclosed. Cite the A rows freely; cite the B rows with the caveat attached.
| Number | What it measures | Source | Grade |
|---|---|---|---|
| 81% | of corporates convert fewer than 1 in 4 pilots into commercial deals | 500 Startups, 2017 · n=100+ Fortune 1000 execs | A |
| 48% | convert fewer than 1 in 10 (read from a chart, not a printed figure) | 500 Startups, 2017 | A |
| 7.3 vs 8.7 | months from first contact to signed contract — corporate’s count vs startup’s count | Match-Maker Ventures × Arthur D. Little, 2016 · n≈400 | A |
| 59% | take 3+ months on pilot procurement alone; 21% do not streamline procurement for startups at all | 500 Startups, 2017 | A |
| ~6% | engagement to signed deal, across 1,500+ engagements at 120+ corporates | European Innovation Council, 2025 | A |
| 23.8% | of accelerator participants sign a contract; a further 54.4% run a PoC and sign nothing | Venture Capital journal, 2025 · n=101 | A |
| 60% / 14% | of corporate accelerators shut within two years; only 14% were more active after two years | CB Insights, 2019 · no disclosed sample | B |
| 60–80% | drop-off at the scale decision, vs 10–20% during pilot execution | Synthesis across sources — direction solid, precision soft | B |
If you have been handed these in a board deck, they came from somewhere other than where the deck says they came from.
“Structured pilots with measurable KPIs are 3× more likely to scale”
Attributed to BCG
Misattributed We pulled the BCG report. It contains no “3×”, no “three times”, and the phrase “proof of concept” never appears in it. The real finding is Sapphire Ventures’: POCs lasting under three months are 3× more likely to reach implementation. The variable was silently swapped from duration to “structured with measurable KPIs” — which is a different, and much more flattering, claim.
“More than half of proofs of concept fail”
Attributed to BCG
Misattributed Also absent from the BCG report. It originates as a Sapphire Ventures blog headline, which was itself restating a different measurement: that 78% of surveyed IT executives said fewer than half their POCs reach production. Sample undisclosed.
“70% of transformations fail”
Attributed to McKinsey
Zombie Traces to Hammer & Champy in 1993, who described it themselves as an unscientific estimate about business reengineering. A 2011 review in the Journal of Change Management examined five published instances and found no valid empirical evidence for any of them. McKinsey never measured it. The companion “14%” is real, but it describes an AI-adoption stage distribution — not pilot conversion.
“43% of enterprise deals stall in security review”
Attributed to two security vendors
Unverifiable Traces to a security-questionnaire vendor’s blog post citing two sources with no link, page, method or sample — neither locatable. The post then uses the figure to compute the ROI case for its own $149/month product. Security review is a genuine bottleneck; this particular number is not evidence of it.
How this propagates is the part worth knowing. The page laundering the BCG misattributions carries source links tagged utm_source=chatgpt.com — unchecked machine-generated citations, quoted onward in good faith by people who reasonably assumed somebody had already verified them. Also excluded from our corpus for the same reason: “90% of innovation labs fail” (agency blogs only) and BCG’s much-quoted “only 5% of collaborations work well” — which is a founder’s opinion quoted inside the report, not a BCG finding.
Whether you’re a corporate with a pilot that needs a path to a P&L, or a startup trying to get one over the line — this is the problem I work on. I’ve led three corporate spin-outs through exactly this transition, and I sit on both sides of the table across eleven APAC markets.
If you’re building partnerships between corporates and startups, reach out. Happy to compare notes even if there’s nothing to sell.
Related: Conversion rates → · The full research →