Bought almost everywhere, activated almost nowhere
Paid AI now reaches 66% of large businesses on Ramp, against a broad-market floor of 37%. Buying is close to universal among early adopters, yet neither series shows the AI reaching production.
The fund has the receipt. The asset does not.
95% of private equity funds say AI met the business case, yet only 7% have it fully integrated across the portfolio and 11% of investors link AI progress to the exit.
The external record
15 findings from named investors, advisors and usage data show the same shape without any Crossover data: the fund reports success, the portfolio reports pilots.
Private equity and advisory surveys9
- Reported, primary
95%of funds say AI met or beat the business case
Inside the portfolio, only 7% of companies have AI fully integrated and 36% use it daily.
ConfirmsThe fund sees success the portfolio has not reached. Enterprises are three times as likely to be in production (37% against 12%).
- Reported, primary
66%expect AI benefits within 12 months, up from 34%
Only 31% call implementation efficient, and 19% of high performers beat the AI business case against 5% of everyone else.
ConfirmsSpend runs ahead of proof.
- and StepStoneReported, primary
39%of general partners expect no material financial impact in 2026
The best returns so far sit in the fund's own diligence and sourcing, not in portfolio operations.
ConfirmsDeployment comes first; proof has to follow.
- Reported, primary
76%of investors cite an unclear return on AI
82% track the return, 40% have taken a valuation haircut of 5% or more, and only 11% link digital progress to the exit story.
ConfirmsAn unclear return is the top barrier in every CFO size band.
- Reported, primary
25%have moved 40% or more of their pilots into production
Sanctioned access to AI rose from under 40% to about 60% of workers.
ConfirmsMost pilots stall before production.
- Reported, primary; correlation
~130%higher median revenue multiple for broad AI adopters
Across 471 private-equity-backed companies ($1M to $250M revenue, 2023 to 2025), multiples run 31x, 20x and 13x by AI maturity. This is a correlation, not a cause.
ConfirmsThe premium is paid after proof.
- Reported, secondary
88%use AI in at least one business function
Only 39% attribute any earnings impact to it, and for most that impact is under 5%.
ConfirmsAdoption is not the measure; attributable return is.
- and Long AngleReported, secondary
$285Berased from software stocks in one session, Feb 3, 2026
Forward multiples fell from about 39x to 21x, and the Software Equity Group SaaS index was down 25.7% for the year at March 31.
ConfirmsPublic markets apply the same proof gate to software.
- SaaS pricingReported, secondary; Intercom primary
15%of SaaS companies still price purely per seat, down from 21%
Outcome pricing is fully in place at 9% of companies and being explored by 47%. Intercom charges $0.99 per resolved conversation.
ConfirmsPer-seat pricing loses: 47% of compliance software customers reject it.
Usage and spend data6
- Ramp AI IndexObserved from the series
65.7%of large businesses on Ramp pay for AI
Medium businesses are at 62.5% and small at 50%, up from 51.7%, 47.7% and 38.6% a year earlier. This is an early-adopter ceiling; the Census survey is the floor.
ConfirmsThe midmarket buys packaged AI; it does not pilot.
- Ramp AI IndexReported, primary
1%of firms account for 80% of spend on the two largest labs
Adoption is at the top of its S-curve, so growth has to come from intensity. Token spend held flat for two months while volume rose.
ConfirmsThe purchase line is crossed; the production line is not.
- Vercel AI GatewayReported, primary
58.9%of tokens are now agentic, up from 31.6% in six months
Spend follows the cost of being wrong: business-to-business work pays about twice as much per token as consumer work, and 35 models now serve 10M+ requests.
ConfirmsAccuracy ranks first, the human stays on the exception path, and no single model is the moat.
- OpenRouterReported, primary
128.4Ttokens a week, from 0.5T in January 2025
Agents passed humans as the main source of traffic on Feb 6, 2026. Open-weight models carry 79% of tokens while closed models take 73% of spend.
ConfirmsCapability is not what holds buyers back.
- Menlo VenturesReported, primary
$37Benterprise generative AI spend in 2025, 6% of SaaS
76% of it is bought rather than built, and AI buyers convert at 47% against 25% for other software.
Confirms95% of finance leaders would buy rather than build; the gate is weakest where proof is instant.
- Anthropic Economic IndexReported, primary summary
75%of enterprise API use is automation, led by back-office tasks
Consumer use leans the other way: 52% augmentation against 45% automation.
ConfirmsEnterprises automate the back office and keep a person where judgment sits.