Quick insights
The whole report in five points.
- 01PE conviction is near-universal at 84%, but only 7% of portfolio companies run AI at enterprise scale. The binding constraint is operating model, not belief.
- 02AI value splits into two engines: efficiency gains that protect margin today, and opportunity gains that change the growth story and the exit multiple.
- 03Both adoption modes are required. System-led AI is measurable but hard to implement; user-led AI is easy to start but hard to attribute.
- 04What stops AI from scaling is talent, data readiness, and time-to-value. Budget ranks near the bottom of the barrier list at 20%.
- 05The entry point is deliberately small: one portfolio company, a 30 to 60 day diagnostic, ending in a fundable 90-day roadmap.
In one paragraph
Private equity does not have an AI awareness problem. It has a translation problem: 84% of funds expect AI to transform portfolio value, while only 7% of portfolio companies have reached enterprise-scale deployment. This brief positions AI as a third value-creation lever alongside financial engineering and operational excellence, and argues that the value splits into two engines. Efficiency gains protect and expand margin by doing today's work faster and cheaper. Opportunity gains open work that was previously uneconomic for an SMB or midmarket company, which is what moves the growth story and the exit multiple. Capturing either one requires a single governed loop running prioritization, baselining, system access, build, adoption, and measurement together, rather than a strategy deck without a build or a build without adoption.
What the report argues
The load-bearing claims, in the order the argument builds.
- AI is becoming a third value-creation lever in private equity, alongside financial engineering and operational excellence.
The old model ran on leverage, multiple expansion, and operational improvement. Bain summarizes the new deal math as '12 is the new 5,' which forces faster EBITDA growth from the operating side. EY frames AI as the third lever. The brief's positioning follows from that: turn AI from a belief, a scattered pilot, or a tool expense into a measured EBITDA lever across the portfolio.
- Conviction is high and execution is rare, and the gap is an operating-model problem rather than a belief problem.
84% of funds expect significant impact (EY) while only 7% of portcos run AI at enterprise scale and 43% are experimenting, limited-use, or not using it materially (FTI). 95% of funds say their AI initiatives meet or exceed expectations, but only 17% say they significantly exceed them, so the returns concentrate among disciplined executors rather than the biggest spenders.
- Value splits into two engines: efficiency gains that protect margin, and opportunity gains that change the growth and exit story.
Efficiency gains answer how to do today's work faster, cheaper, and better, and show up as margin expansion, cost reduction, and cash conversion through support automation, finance close, AP and AR, sales follow-up, procurement, and IT helpdesk. Opportunity gains answer what the company can now do that was previously too expensive, slow, or complex, and show up as revenue growth, pricing, retention, and exit narrative. Everything else in the brief is a proof point for one side of that split.
- Opportunity gains matter because AI gives a portfolio company enterprise-grade leverage before it has enterprise-grade scale.
A 200-person company can run near-enterprise customer analytics. A regional services business can run account-based marketing without a large marketing ops team. A lean finance team can produce board-level variance analysis and scenario planning without a full FP&A function. A multi-location operator can find performance patterns across branches, staff, inventory, and service quality.
- Adoption requires both system-led and user-led AI, because each covers the other's weakness.
System-led AI behaves like Salesforce or NetSuite: embedded in defined workflows, so it is easy to measure against baselines such as cycle time, cost per ticket, close days, and renewal rate, but hard to implement because it needs integration, process design, governance, and change management. User-led AI behaves like Excel: powerful because the people closest to the work find the local gains, but hard to track because the value is spread across hundreds of small moments. The service opportunity is to build the measurable workflow layer while instrumenting the distributed usage.
- Efficiency improves EBITDA, but the multiple moves with maturity level, and that evidence is associative rather than causal.
Across 471 PE-backed companies with $1M to $250M in revenue, McKinsey reports operating-model enhancement (Level 2) at a 14x median revenue multiple and roughly 20% higher revenue per employee than Level 1; product transformation (Level 3) at 20x, 43% above Level 2; and business building (Level 4) at 31x with revenue per employee of $180K against $118K at Level 3. The sample design cannot establish that AI drove the increase, so the honest read is association.
- Scaling stalls on people and process, not on tooling or budget.
The barriers portcos report are talent and skills (35%), data readiness and accessibility (33%), time-to-value and deployment speed (29%), integration with legacy systems (28%), organizational change (25%), siloed ownership (25%), and executive sponsorship (24%). Budget ranks near the bottom at 20%. That is an execution-capacity problem, which is exactly what a diagnosis-only or build-only engagement cannot fix on its own.
- The two providers PE defaults to each solve half the problem, so the engagement fails wherever the handoffs sit.
Large consulting firms are strong on framing, diligence, and board narrative, but the engagement is priced and timed for strategy work, and the team writing the AI strategy sits far from any team that would build and operate the tooling. Tech and dev shops ship integrations and agents quickly, but build to a spec rather than an EBITDA outcome and carry no change-management plan, so the tool works in the demo and gets ignored in production. Architecture without adoption is shelfware; a strategy deck without a build is a hypothesis.
- The modeling ranges are modest and defensible, which is what makes them usable in a value-creation plan.
63% of cost initiatives target 5% to 10% savings and 65% of revenue initiatives target 5% to 10% uplift, with 54% to 67% of use cases clearing a 10% ROI hurdle. McKinsey puts day-to-day genAI operating expense at roughly 1.0% to 1.5% of current IT budget, excluding cloud and personnel, and estimates that technology is only about 20% of the work.
- Speed without validation is its own risk in a diligence context.
McKinsey found that in seven of ten industries analyzed, genAI deep-research reports were more optimistic than reports built on expert interviews, and roughly 40% of the important data points surfaced in expert interviews were absent from the corresponding LLM answers even after further prompting. The missing items were the ones that decide a deal: contract structures, unit economics, channel breakdowns, regulatory hurdles, gross margin benchmarks, and inventory turns.
The numbers
Figures taken directly from the source. Each chart names where it came from.
Section by section
A walk through the source in its own structure. Collapse anything you do not need.
Executive thesis: translation, not awareness
sec.1
PE firms already believe AI matters. The unsolved problem is converting that belief into measurable EBITDA expansion, revenue growth, operating leverage, and exit value. The brief defines the work as AI-to-EBITDA translation and scopes it end to end rather than as advice alone.
- Identify where AI creates efficiency gains and opportunity gains.
- Prioritize the portfolio companies and workflows with the highest value potential.
- Build tailored AI workflow systems on best-in-class tools and models.
- Enable user-led adoption from the people closest to the work.
- Track usage, productivity, revenue lift, cost savings, and EBITDA impact.
- Produce board-ready reporting that links AI initiatives back to value creation.
The two engines of AI value
sec.2
The spine of the argument. Efficiency gains improve EBITDA by doing existing work faster, cheaper, and with fewer errors, which protects margin without relying on headcount cuts or price increases. Opportunity gains expand what the company is capable of doing at all, reaching capabilities and markets that were previously uneconomic at SMB and midmarket scale.
- Efficiency examples: support automation and ticket triage, finance across AP, AR, collections and close, CRM automation and proposal generation, HR and procurement, IT helpdesk and internal knowledge retrieval, and reduction of low-value agency and software spend.
- Opportunity examples: near-enterprise customer analytics at 200 people, account-based marketing without a marketing ops team, mining calls and tickets and contracts for churn and pricing signal, board-level FP&A from a lean finance team, faster product and QA cycles with a smaller team.
- The framing line: AI gives portfolio companies enterprise-grade leverage before they have enterprise-grade scale.
Adoption model: system-led and user-led AI
sec.3
Use cases alone do not create value; value appears when AI is embedded in how work actually gets done. The brief argues both adoption modes are required because they fail in opposite directions. System-led AI is measurable but hard to implement. User-led AI is easy to start but hard to attribute.
- System-led AI runs through defined workflows with existing baselines: cycle time, cost per ticket, close days, conversion rate, renewal rate, headcount leverage, error rate, revenue lift.
- System-led AI is gated on workflow mapping, systems integration, data access, permissions, change management, and executive buy-in.
- User-led AI produces distributed gains that arrive twenty minutes at a time, which is real value that standard reporting never sees.
- The differentiated offer is to build the measurable workflow layer and instrument the distributed usage at the same time.
Strategic frameworks, folded into one lens
sec.4
Three external frameworks are used as supporting structure rather than taught separately. BCG's Deploy, Reshape, and Invent maps onto user-led adoption, tailored workflow systems, and opportunity gains. McKinsey's four-level value creation ladder runs from opportunistic adoption to operating-model enhancement to product transformation to business building. McKinsey's two-by-two approach pairs quick wins with deep redesign.
- BCG's caution: buying tools and pushing usage does not create P&L uplift without active change management, workflow redesign, value tracking, and portfolio-level accountability.
- The brief's own distinction: do not build the foundation model or the generic horizontal tool. Build the workflow layer that makes AI useful inside the portco's actual operating model.
- McKinsey's practical sequencing: start two small use cases for momentum, and in parallel redesign two strategic workflows that represent the essence of the company.
- The implication: productivity is table stakes; AI-enabled operating models, products, and new businesses are where valuation upside sits.
Evidence bank
sec.5
The supporting research, organized by what each source is being asked to prove. Bain and EY establish the pressure on the old value-creation model. FTI supplies the maturity and target-range numbers from 200 PE decision makers surveyed in December 2025. McKinsey supplies the maturity-to-multiple association and the cautions on AI-generated research. BCG and McKinsey's State of AI establish how thin enterprise-level financial impact still is.
- Goldman Sachs estimates roughly $7.6T of global AI infrastructure investment from 2026 to 2031, which pulls industrials, logistics, energy, and services portcos into scope rather than software alone.
- McKinsey's State of AI: 88% of organizations use AI in at least one function, about a third have begun scaling enterprise-wide, and 39% report enterprise-level EBIT impact, most of it under 5%.
- BCG: 5% of companies are generating substantial AI value, 35% are scaling, and 60% report minimal revenue and cost gains, which is performance dispersion rather than a rising tide.
- Gartner, via ITPro: agentic AI breaks seat-based licensing, with up to $234B of application spending exposed to agentic arbitrage by 2030, which makes SaaS rationalization part of the value case.
- AlixPartners: across 500 software companies in 12 PE portfolios, only 14% had strong protection on both proprietary data and vertical specialization, while roughly 25% had weak defenses on both.
Competitive positioning: why the obvious answers fall short
sec.6, sec.7
Firms facing the execution gap default to a large consulting firm or a tech and dev shop. The brief argues both fail at the same seam. Consulting produces a strong diagnosis with weak, outsourced, or absent implementation. Dev shops produce a strong build with weak adoption and no translation back to EBITDA.
- Three capabilities have to live in the same engagement: technical fluency (architecture, model and tool selection, integration), business fluency (workflow economics, EBITDA framing, board reporting), and adoption discipline.
- Adoption discipline is the piece almost no provider carries systematically, and it is the one that decides whether a working tool changes anyone's day.
- The stated proof of the positioning is first-client evidence: the systems were built and run inside the author's own business before being sold, and an agency whose job is evaluating this class of work paid for it.
Service packaging: five ways to engage
sec.8
The offer is structured so each phase earns the next on its own measured economics rather than on a promise. The entry point is a single-portfolio-company diagnostic, deliberately small.
- AI-to-EBITDA Diagnostic: readiness assessment, workflow friction map, efficiency and opportunity inventory, use-case prioritization, data and systems readiness, risk and governance, a 90-day roadmap, and an EBITDA impact hypothesis.
- Workflow Reshape Sprints: two quick-win workflows for adoption momentum plus two strategic redesigns, delivered with tool selection, integration plan, training, adoption dashboard, and a KPI baseline.
- User-Led AI Enablement and Measurement: usage policy and guardrails, job-family training, prompt and workflow libraries, intake system, telemetry, and a value-capture dashboard.
- Portfolio AI Operating System: governance model, fund and portco roles, KPI standards, vendor strategy, shared playbooks, and board reporting templates.
- AI Product and Business-Building Strategy: for data-rich or AI-disrupted portcos ready to move past productivity into new revenue, including moat assessment, pricing, build-buy-partner, and exit narrative support.
From brief to deck
companion deck
The 18-slide deck built from this brief tightens the argument to one spine and adds material the brief does not contain. It carries the same efficiency-versus-opportunity split, then adds an at-scale case study, an operating loop, and a concrete first engagement.
- Replit's self-reported internal deployment is the at-scale proof: 2.9x code output per engineer and 5.8x total code contributed, 30% of human PR review time saved, and 60% faster resolution of the hardest support tickets, with review latency, defect trends, and incident trends flat. Company-reported, not third-party audited.
- The transferable claim is the operating pattern rather than the multiple: a governed agent loop with trusted data, permissions, playbooks, verification, and human escalation, reaching a non-code function through a verifiable work product.
- One governed loop converts strategy into measured EBITDA: prioritize, baseline, connect, build, adopt, then measure and scale, on a governance rail of permissions, audit logs, data controls, approval thresholds, quality evaluation, and cost monitoring.
- The ask is one portfolio company for a 30 to 60 day diagnostic shipping eleven deliverables and ending in a fundable 90-day roadmap, with each later phase gated on the prior phase's measured economics.
Open questions the brief leaves unresolved
sec.11
The brief is explicit about what it has not settled, which is useful context for anyone reading it as a strategy document rather than a finished offer.
- Whether the initial offer should be a portfolio-wide diagnostic or a single-portco sprint. The deck currently defaults to single-portco.
- Which verticals to prioritize first.
- Whether the primary audience is operating partners, deal partners, or portfolio company CEOs.
- Whether the emphasis belongs on portco operations, fund-level AI, or both.
- Pricing and deal structure, including retainer, success fee, or equity and warrants, which is not yet addressed anywhere.
Why it matters
How CS Ventures reads this for client work. Interpretation, not the source's claim.
- Conviction is not a differentiator
With 84% of funds already expecting transformative impact, believing in AI buys a portfolio company nothing. The scarce asset is a governed operating loop connecting a prioritized workflow to a measured baseline, which is where the 7% sit.
- Pick the engine before the use case
Efficiency and opportunity gains carry different evidence requirements, different sponsors, and different reporting. Deciding which engine an initiative serves, before selecting tools, is what keeps a portfolio AI program from becoming a list of pilots.
- Underwrite the barriers, not the tools
Talent, data readiness, and time-to-value are the top three constraints, while budget ranks near the bottom. A value-creation plan that funds licenses without funding integration, change management, and measurement is funding the wrong line.
- Keep the multiple claim honest
The 14x to 31x maturity ladder is an association across 471 companies, not a causal estimate. As a directional argument for moving past productivity it holds. As an underwriting assumption it does not.