Original Research · Arc Shift Ventures
We went looking for one number from every major corporate startup program — the rate at which startups become commercial customers of the corporate running it — and exactly one program model publishes it.
Four numbers that frame it
of US accelerator programs leave startups worse off than never joining, in the largest study yet run — ~750,000 startups, 329 programs. Selection, not acceleration, explains most of the rest.
Baek & Hegde, NBER Working Paper 35063, April 2026 · Grade A · working paper, not yet peer-reviewed; disclosed sample. Conflict disclosed: one author directs an accelerator the paper places in its top-performing tier
startups received credits from the largest cloud provider’s startup program — US$6 billion over ten years. Published figures on how many became paying customers of that cloud: zero.
The provider’s own 10-year program retrospective, 2023 · Grade B · vendor-published
raised by the alumni of another tech giant’s accelerator; 109,000 jobs across 1,700+ alumni. Published figures on alumni who became cloud or enterprise customers: zero.
The company’s 2025 accelerator impact report · Grade B · vendor-published
of startups that pilot with one carmaker’s venture-client unit convert to paid follow-up projects — 280+ pilots, 6,000+ startups assessed. The one program type that publishes conversion is the one whose unit of activity is a purchase order.
The unit’s own figures, via an industry interview · Grade B · self-reported
The finding
The counter-model’s own survey: venture-client units adopt 25% of piloted solutions vs 10% without one — vendor numbers (Grade B), but the only conversion rates published anywhere in this field.
Where it breaks
of 1,500+ corporate–startup engagements reached a signed deal when someone actually tracked them end-to-end — the field’s low anchor.
European Innovation Council headline counts, 2025 · EST · a ratio we computed from 1,500+ engagements and 100+ reported deals, not a rate the EIC publishes
of corporates convert fewer than 1 in 4 pilots — the same unowned handover to a revenue owner that kills the program funnel at stage three.
500 Startups, 2017 · Grade A · via Beyond the Pilot
What works
Numbers that don’t survive a source check
“60% of corporate accelerators fail within two years”
Attributed to a widely-cited industry research firm
Unverifiable as quotedThe number exists only in the headline of a 2019 research page — no sample, method or definition of “fail” is publicly visible, and no underlying dataset has ever surfaced.
“Corporate accelerators cut a startup’s success rate to 8%”
A venture builder’s internal analysis, via startup press
Quoted far beyond its weight classOne internal analysis of a single 2013 founding cohort (baseline 11%, accelerator 12%, corporate accelerator 8%), sample size and “success” undisclosed, never peer-reviewed.
“96% of our accelerator alumni survive, while over 80% of startups fail”
A global tech company’s accelerator impact report
Vendor stat on a zombie baselineA heavily selected numerator against an unsourced denominator — the “80% fail” baseline traces to no disclosed study, and selection alone predicts high alumni survival.
“80% of the world’s unicorns run on [our cloud]”
A major cloud provider, citing a market-data firm’s unicorn list
True-ish, but a category error where it’s deployedA cloud market-share observation deployed inside the provider’s credits-program retrospective as program evidence — it says nothing about credits-to-customer conversion, or even whether those unicorns took credits.
I connect startups to corporate innovation teams across eleven APAC markets, so I watch daily where program engagement does and doesn’t become revenue. If you want to know what your funnel’s bottom looks like — or whether a program is worth three months of founder time — reach out.
Book a diagnostic Connect on LinkedIn →