95% of PE funds report AI initiatives meeting business case targets

Operators and investors,

95% of private equity funds report their AI initiatives meeting or exceeding the original business case, according to FTI's 2026 Private Equity AI Radar, a December 2025 survey of 200 senior deal and operating leaders at firms with at least $1B in AUM. The same survey puts revenue acceleration as the top AI priority at 41% of initiatives, against 24% aimed purely at cost reduction, and names talent as the main scaling constraint at 35%. The number is being repeated across sponsor decks and conference panels as evidence that AI in PE has moved past the proof stage.

The data outside the survey tells a different story. S&P Global's Voice of the Enterprise survey of 1,006 companies tracked the share abandoning most of their AI initiatives rising from 17% to 42% in a year, with the average organization scrapping 46% of its proofs of concept before production, and MIT's NANDA study from August 2025 found 95% of generative AI pilots delivering no measurable P&L impact. The contradiction even lives inside FTI's own pages: only 17% of funds significantly exceeded their business case, 43% of portfolio companies are described as not materially deploying AI at all, and just 7% reach enterprise-scale deployment.

My read from portfolio work is that both datasets are accurate and they are measuring different things. I have not sat in a single operating review where an AI initiative cleanly closed its business case with a P&L number attached. What I see instead is teams still grinding through data cleanup 9 months in, pilots renewed for another 2 quarters because nobody wants to write them off, and business cases written softly enough that "meeting targets" is close to guaranteed. The 95% says less about AI performance and more about how PE funds write AI business cases. This issue covers where the claim holds, where it breaks, and what to do with your own portfolio's AI reporting.

Read the two panels together and the survey answers itself. The same 200 leaders reporting 95% success also report that fewer than half their portfolio companies materially deploy AI at all. In our portfolio work the left panel is what reaches the board deck and the right panel is what we find in the first 2 weeks of an audit: pilots with soft baselines, data dependencies nobody priced, and a CRM that blocks half the use cases before they start.

Where the claim holds

The claim is real in a specific segment: fund-level deployments with narrow scope and a clear baseline. Deal-sourcing screens, diligence document review, portfolio-monitoring dashboards, and drafting workflows inside the fund itself are the deployments that come in on target, because the work is repetitive, the data is controlled, and the user base is a few dozen professionals rather than a portfolio company's entire commercial team.

FTI's Alpha Tier confirms the pattern from the other direction. The funds with consistent outsized returns run dedicated AI teams, disciplined use-case selection, and staged funding. Sumeet Gupta, who leads AI and digital transformation at FTI, frames it as "disciplined investment, focused execution and optimized operating model choices, which is separating top-performing PE funds in delivering outsized returns," and the operative words there are discipline and operating model rather than the technology. The funds that built institutional AI capability, the way EQT did with Motherbrain or Vista and Hg did with their portfolio-wide programs, earn their success rates through selection: a fund that only greenlights use cases with controlled data and measurable baselines will report high numbers, and it will also leave most of the harder, higher-value portfolio work untouched.

Where the claim breaks

The claim breaks at the portfolio company level, where the revenue acceleration the survey ranks first actually has to happen. A fund's diligence copilot has 30 users and a controlled corpus. A portco's AI-assisted sales motion has hundreds of users, a CRM with 5 years of inconsistent hygiene, and a change-management problem the business case never priced. This is where the S&P Global abandonment numbers and MIT's P&L findings live, and it is where I see engagements stall.

The gap starts in how the business case gets written, and FTI concedes the point in its own analysis, noting that the outperforming initiatives ran against cases that "were often conservatively scoped." The report quantifies the scoping: roughly two thirds of AI initiatives target improvements of just 5% to 10%, below the 15% impact benchmark FTI itself cites from Gartner, and expectations above 20% are rare. When the survey asks whether the initiative met its business case, a pilot that cleared a deliberately low bar counts the same as one that moved revenue. Self-reported success against self-written targets is the softest measure in operating diligence, and it is the measure the 95% rests on.

The talent constraint at 35% is the tell inside the survey. If 95% of initiatives were delivering against real commercial targets, funds would be scaling winners, and scaling a working playbook is a capital problem rather than a talent problem. A talent bottleneck at this scale means the work is still heavy, manual, and dependent on scarce people who can bridge the model and the operating reality, which is what an honest read of the portfolio layer shows. It is also why demand for external AI delivery capability keeps growing; we maintain a directory of the consulting market at allconsultingfirms.com, and the implementation side of AI is where portfolio companies go looking first.

This is the curve that never makes it into a sponsor survey, because abandoned initiatives stop having owners who answer surveys. Between the 2024 and 2025 survey cycles, a quarter of the market moved from committed to abandoned, and the PoC scrap rate at 46% means the median company kills nearly half its AI work before production. The delta between this chart and FTI's 95% is where I'd focus any operating review.

What to do in the gap

The gap between reported success and realized P&L impact is an operating discipline problem, and it is fixable at the business case stage:

• Write the business case in P&L terms: a revenue line, a cost line, or a working-capital effect, with the baseline measured before the pilot starts. Hours saved only count when the hours are redeployed or removed.

• Set a kill date at 90 days: the pilot either graduates to a funded rollout with an owner, or it stops. Renewal by inertia is where abandoned initiatives hide between survey cycles.

• Name the operating owner before funding: an AI initiative owned by IT or by the fund's digital team reliably stalls at the workflow boundary. The function whose number moves owns the deployment.

• Audit the data dependency first: most stalled deployments I see were blocked by the CRM, the product data, or the reporting layer underneath, and that work was never in the business case. Price it in or expect the timeline to double.

• Report realized impact quarterly against the original case: the same discipline the better funds already apply to synergy tracking in M&A, applied to AI spend.

The critical point for operators: when a portfolio review shows AI initiatives at a 95% success rate, that is a reporting artifact to investigate rather than a result to celebrate. The funds actually compounding AI into returns are the ones whose reviews show killed pilots, reallocated budgets, and 2 or 3 initiatives with hard P&L attribution. A clean scorecard means the targets were soft.

Mario

📊 Most portcos track EBITDA only at the company level. Run four or five GTM motions and nobody can say which one earns margin and which books revenue at a loss, a gap that surfaces 18 months into the hold when "improve sales efficiency" has no owner. Track margin at the motion level.

⚡ Slow pages leak revenue. A 4.2-second mobile load and a 61% bounce rate get blamed on bad leads, when a 2-second delay alone cuts mid-market B2B conversion by 7-12%. Treat page speed as a commercial number, not a UX detail.

💸 ARR up 22%, cash in the bank flat. Both numbers were real: new logos and expansion looked healthy while dollar churn ate almost all of it and a few large accounts contracted, all averaged into one comfortable line. Read the blend behind the headline metric.

🛒 Agentic commerce is not here yet, but it is getting close. Google Payments and Link are still out of reach for agents through MCPs or OAuth SDKs, while consumer shopping already scales and DTC will push these tiers into B2B and the mid-market. Watch the agent-commerce stack.

Enterprises are moving from renting AI to owning it as usage bills scale. Hugging Face CEO Clem Delangue says roughly half the Fortune 500 now deploys private and open models through the platform, with the recurring pattern being companies that start on frontier APIs and migrate once scaling costs hit the P&L.

Production AI workloads are shifting to open models while frontier systems settle into a premium layer. Chinese open-weight models took 41% of Hugging Face downloads this spring, surpassing US models, the top 6 models on OpenRouter are all Chinese open releases, and open models handled nearly a third of AI requests on Vercel in June.

Take-private capital is now reaching the largest fintech names. Stripe and Advent International reportedly bid more than $53bn for PayPal at $60.50 per share, backed by roughly $50bn in committed bank financing and a 28% premium, after the stock lost over 40% of its value in the past 12 months.

Distribution concentration is surfacing as the structural risk in semi-liquid private credit. Investors pulled 15.4% of assets from Blue Owl's $3bn technology income fund in Q4 2025, with redemption requests reportedly passing 40% of NAV in Q1 2026, after UBS, whose wealth clients supplied about 60% of the fund's capital, advised trimming private credit exposure.

Multi-category KDP: 2-year-old Amazon KDP consisting of a multilingual catalog across evergreen self-help, health, children's books, test prep, and oriental discipline niches. Business almost fully automated via specialized software for ads and profit monitoring. $85,000

Logistics SaaS: 16-year-old SaaS platform powering logistics operations for courier, freight, and virtual locker companies worldwide. Generate revenue via custom configurations, feature requests, and specialized software integration. $750,000

Niche Travel Business Coaching Site: 3-year-old digital education business focused on helping women build profitable experiential travel businesses. Generates revenue via a multi-level monetization model. $900,000

Furniture Shopify Brand: 6-year-old Shopify brand specializing in Italian-made upholstered sofas, beds, and armchairs. Operated by a lean team with streamlined manufacturing and fulfillment partners. $5,801,904

For PE partners and operators seeking alpha

🌐 Scaling $50M - $500M+ mid-market companies with value creation through RevOps, data engineering, and WordPress. DevriX provides full RevOps consulting + delivery with GTM enablement for PE-backed portfolio companies, traditional tech, healthcare, finance, and professional service businesses pacing toward revenue growth initiatives. Our standard retainers between $10K and $60K include revenue lifecycle services for marketing and sales leaders, FP&A for financial teams, pipeline enrichment through websites and dozens of lead sources, automations and delivery integrations, CRO and ongoing testing, product delivery and platform integration solutions, and more through our consulting solutions.

🚀 1:1 Advisory retainers. Supporting operating partners, private equity funds, family offices, and mid-market executives in different capacities, from value creation through due diligence to portfolio digital GTM management in my async advisory programs via Growth Shuttle.

📈 GTM while scaling. European and international businesses can opt in for doola LLC and their “Business in a Box” model. Scaling founders can find smaller digital opportunities on Flippa. And additional opportunities across my investments can be found here.