PE: The $10–100M AI Gap
Why Lower-Middle-Market PE Firms Should Run AI Across the Portfolio — With a Partner
Private equity's value-creation math has changed, and the change is unkind to anyone relying on the old playbook. Cheap debt and reliable multiple expansion are gone. Bain & Company's 2026 Global Private Equity Report puts it bluntly — "12 is the new 5," meaning the EBITDA growth a deal used to need over five years is now expected in far less time, and the firms that win will, in Bain's phrase, "build systems, not slogans." Bain's 2025 report makes the same point about margins: after a decade in which margin expansion was a bonus, it has become a necessity, and AI is emerging as the most direct lever to pull on it across a portfolio.
For a lower-middle-market PE firm — operationally hands-on, investing its own capital, holding established manufacturers, distributors, logistics and industrial-services businesses in the $10–100M revenue range for the long term — this should be good news. These are exactly the kinds of unglamorous, process-heavy businesses where disciplined automation drops straight to the bottom line. The problem is that the companies most able to benefit from AI are also the ones least equipped to deploy it. That gap is the opportunity, and closing it company-by-company is the wrong way to do it.
The leaders are already proving the returns
This is no longer speculative. When Bain surveyed private investors representing $3.2 trillion in assets under management in late 2024, a majority reported that their portfolio companies were already in some phase of generative-AI testing, and nearly one in five had operationalized use cases producing concrete results — a striking figure for a technology this young.
The published results from the firms ahead of the curve are specific and operational, not hype:
At Apollo's Cengage, a set of AI projects cut costs by roughly 40% in selected content-production processes, 15–20% through automated lead generation, about 15% in customer care, and 10–15% in software development.
Across Vista's software portfolio, AI code-generation tools have driven coding-productivity gains of up to 30% for scaled adopters.
The pattern underneath these numbers matters more than the numbers themselves. Bain's clearest finding is that the winners treat AI as "a tool in service of strategy, not a strategy on its own" — they force portfolio companies to pick a short list of real priorities and apply AI there, rather than dabbling. That discipline is the difference between a result and a demo.
The lower-middle-market gap
Now look at where most lower-middle-market portfolio companies actually sit. Adoption is not the issue — capability is. RSM's 2025 middle-market survey found generative-AI adoption had surged to 91% of middle-market companies, up from 77% a year earlier. But beneath that enthusiasm, the same firms reported that their biggest obstacles were a lack of in-house expertise (cited by 39%), the absence of a clear AI strategy (34%), and data-quality problems — and only about one in four had AI genuinely integrated into core operations. The World Economic Forum frames the structural reason: the mid-market is the "missing middle," responsible for roughly a third of private-sector GDP and employment in developed economies, yet typically carrying weaker IT infrastructure than larger competitors — and as many as 95% of AI pilots fail to reach production.
The reason is not that these management teams are unsophisticated. It is that a $30M precision-machining business cannot justify — or win — a competitive hire for a data scientist or an AI engineer, and has no CIO, no platform team, and no slack capacity to redesign its own workflows. Academic work on mid-size adoption reaches the same conclusion: these companies face smaller IT departments, fewer specialized staff, and simply cannot compete with larger firms for scarce AI talent. So they buy a few tools, run a pilot, and stall in what one analysis aptly calls "pilot purgatory." The capability that would turn a pilot into a durable, margin-moving system never arrives, because no single portfolio company is large enough to build it.
The portfolio is the unit of leverage — not the company
This is the insight the most advanced sponsors have already acted on, and it is the heart of the argument. The right unit of investment is the portfolio, not the individual company.
Apollo's approach is the clearest template. Rather than asking each company to fend for itself, Apollo built a central AI "center of excellence" — staffed with external AI experts and an advisory board, wired into an ecosystem of specialists, technology partners, and service providers — that appraises vendors, evaluates use-case ROI, and matches each management team to the right implementation partner for its level of tech maturity. Hg, focused on midsize software, gets similar leverage a different way: because its companies share the same operating problems, a solution built once tends to work across many of them.
A megafund can staff that center of excellence internally — Vista has arrayed an internal army across 85-plus companies. A lower-middle-market firm cannot, and shouldn't try. But it doesn't need to own the capability to get the benefit of it. The efficient move for a firm holding eight or ten industrial businesses is to partner with a managed-services firm that functions as the fund's outsourced AI operations team — one accountable group whose cost is amortized across the whole portfolio, and whose playbooks compound from one company to the next.
That logic is especially strong for a homogeneous portfolio. A firm whose holdings cluster in metal stamping, machining, fabrication, distribution, and field equipment is, in Hg's sense, a collection of businesses with the same problems: the same month-end close, the same quote-to-cash friction, the same paper-heavy quality and compliance documentation, the same scheduling and reconciliation drudgery. Solve the finance-automation pattern once and it transfers to the next acquisition almost wholesale. The marginal cost of the second deployment is a fraction of the first — an economy of scale a single company can never capture on its own, and that a portfolio-level partner is built to harvest.
Where the value actually is in industrial portfolios
For manufacturing, distribution, logistics, and industrial-services businesses, the high-return AI work is not a customer-facing chatbot. It is in the back office and the operational core, and most of it should be built as reliable, deterministic systems rather than as a generative model sitting in front of every user — using AI heavily to build the system fast, then letting it run without paying a per-interaction token cost forever.
In practice that means the office of the CFO and the office of operations: automating invoice processing, month-end close, and reconciliation; compressing quote-to-cash and order entry; extracting terms and data from contracts, POs, and quality documents; cleaning and connecting the ERP and shop-floor data that every later AI use case depends on; and routing approvals and exceptions through rules rather than re-keying. These are unglamorous, measurable, and exactly the levers an operator-led firm already cares about — they remove manual hours and let a lean team support more volume, which is the entire point of margin expansion. They also build the structured, trusted data foundation that Bain identifies as the prerequisite for everything more ambitious that comes later.
The back office, function by function
The reason the back office is the right place to start isn't only that the work is repetitive — it's that the published evidence points there directly, and these are precisely the functions a portfolio-level partner can standardize once and redeploy across every company in the book.
The office of the CFO. This is the highest-conviction starting point — and, notably, an underexploited one. Finance has actually lagged HR, legal, and IT in AI adoption: Gartner finds about 85% of finance leaders optimistic about AI in the function, yet surveys by McKinsey and Deloitte have shown many CFOs still earmarking less than 1% of budget for it. That gap is the opportunity, because the playbook is well understood while most mid-market finance teams haven't run it. The durable wins are unglamorous and measurable — automating accounts payable and receivable, invoice and PO matching, reconciliations, the month-end close, and the broader procure-to-pay and order-to-cash cycles. McKinsey's own guidance to CFOs is to pick a small number of high-impact use cases rather than deploying AI "everyone, everywhere, all at once" — exactly the discipline a focused partner is there to enforce. The most recent CFO commentary points to agentic systems embedded in the ERP that accelerate the close, sharpen cash visibility, and automate controls while improving auditability — and that last point is worth real money to a PE owner, because cleaner, faster, more auditable numbers pay off at every board meeting and especially in diligence. McKinsey's State of AI research is also blunt that the measurable EBIT impact accrues to organizations that redesign the workflow around the tool, not those that bolt a chatbot onto a broken process. That is why this is build-and-operate work, not a software purchase.
The office of the CHRO. HR is further along, which makes its use cases lower-risk to deploy. By early 2025 a majority of HR leaders reported being in advanced stages of generative-AI implementation — up from under a fifth two years earlier — and SHRM's 2026 work has the large majority of CHROs expecting still-deeper adoption. The high-return targets are the high-volume, low-ambiguity workflows practitioners consistently recommend starting with: onboarding and document collection, PTO and leave requests, payroll exception handling and data validation, benefits and policy Q&A, and routine employee-service tickets. There's a structural reason this matters in the mid-market specifically — HR teams at these companies are lean and have often absorbed headcount cuts, so they have no slack to build anything themselves. The capacity gap is the problem. Done right, the same automation that removes manual HR hours also produces the audit trail a compliance-sensitive, multi-jurisdiction industrial employer needs, with a human kept in the loop on anything consequential.
Legal and contracts. The legal function has moved fastest of all. The FTI Consulting/Relativity General Counsel Report found corporate-legal generative-AI use roughly doubled in a single year, with the large majority of general counsel now reporting it, and the Association of Corporate Counsel's in-house survey names contract drafting and legal research as the top efficiency gains. For an industrial portfolio the value concentrates in high-volume, low-complexity contract work — NDAs, vendor and supply agreements, DPAs, simple MSAs — where first-pass review, clause identification, redlining against a standard playbook, obligation tracking, and renewal management compress cycle time and, in vendor ROI studies, measurably cut outside-counsel spend by shifting routine work back in-house. For a mid-market PE firm — with a legally fluent founding team that frequently provides portfolio-level general-counsel coverage — encoding the house contract playbook once and applying it across every portfolio company is a clean, repeatable win.
Procurement and IT. The same logic extends to procurement (spend analysis, supplier and PO triage, contract-compliance checks) and to internal IT and operations (ticket triage, access requests routed through identity and approval workflows). In every one of these functions the pattern is identical: high volume, rules-heavy, document-dense — and almost always better served by a reliable, integrated, deterministic system, with AI used to build it and to handle the genuinely unstructured edges, rather than by a generative model placed in front of every employee.
The through-line for a PE owner is that these are not five unrelated projects. Finance close, HR onboarding, contract review, and procurement triage share the same shape across every company in a portfolio of similar industrial businesses. Solve each pattern once — with governance and auditability built in — and it transfers to the next acquisition at a fraction of the cost, turning back-office automation from a company-by-company expense into a portfolio-level asset that compounds. It is also the most defensible kind of value: cost taken out of G&A and the finance function flows directly to EBITDA and margin, the two levers the new value-creation math now demands.
The long hold turns a constraint into an advantage
There's a timing trap in AI value creation that catches most sponsors. Industry practitioners consistently find that AI programs take roughly 6–12 months to show initial results and 18–36 months to reach full operational impact. Against an average hold that Bain pegs near 5.8 years — much of which is consumed by ramp-up and exit prep — that window is tight, and many firms run out of runway before the compounding kicks in.
A firm that explicitly tells the market it has no time pressure to sell its portfolio companies is in a structurally better position than almost anyone in the asset class to capture the full 18-to-36-month payoff — provided it starts early and sequences the work. The long hold, usually described as a patience advantage, becomes an AI advantage: it is precisely the horizon over which systematic automation compounds.
Why managed services specifically — not a hire, not a big-firm engagement
Three alternatives exist for closing the gap, and the trade-offs favor a managed-services partner for this market segment.
Hiring internally fails on math: no $10–100M company can justify a full-time AI team, and a fund of this size can't carry a large central one either. A large consultancy will produce an excellent strategy deck and a pilot, then leave — handing the portfolio company a system it has no capacity to operate, which is how pilots die. A managed-services partner is different in one decisive way: it doesn't just advise or build, it operates. It owns the system across the hold, keeps it running and improving, absorbs the maintenance and monitoring burden that a lean portco can't, and stays accountable for outcomes rather than slideware. The cost lands as a predictable service line rather than as headcount on a portfolio company's P&L, and — critically — the knowledge accumulates inside one partner who can carry it from the first portfolio company to the next.
This is the model Lunada is built for. We work as a fund's outsourced AI operations team across three stages — Audit the portfolio company to find where manual work and cost actually live; Build durable, integrated systems against those specific priorities; and Operate them through the hold so the value persists and compounds. Run once and reused across similar businesses, that capability gives a lower-middle-market sponsor the same center-of-excellence leverage the megafunds built internally, without the fixed cost.
It also builds a better asset to sell
The final argument is about exit, not just EBITDA. A portfolio company that has been systematically modernized — clean connected data, automated core processes, a documented operating system rather than tribal knowledge in a handful of long-tenured employees — is simply a more valuable and more defensible asset. Its margins are higher and more durable, its diligence is cleaner, and a sophisticated buyer will pay for the difference. In a market where, per Bain, the gap between top- and bottom-performing firms is widening and a data-backed edge increasingly separates them, an AI-modernized portfolio is becoming part of what a quality business looks like at sale.
The bottom line
For an operationally driven, long-hold, manufacturing-savvy lower-middle-market firm, AI is no longer optional to value creation — but doing it company-by-company is a losing structure. The companies that need it most can't staff it, the pilots that result mostly die, and the returns the leaders are already booking go uncaptured. Running AI as a portfolio-wide program through a managed-services partner inverts that: one capability, amortized across the portfolio, built as durable systems, operated through the hold, and compounding from each deal to the next. It is the highest-leverage, lowest-regret way for a firm of this size to turn the new value-creation math from a threat into an edge.
If you'd like to see where the opportunity sits in a specific portfolio, a LunadaLabs Audit is a fixed-scope first step — a clear-eyed map of where manual work and cost are concentrated across one or more portfolio companies, and which workflows are genuinely worth automating first.
Sources
Bain & Company, Global Private Equity Report 2026 — "12 is the new 5"; "build systems, not slogans." https://www.bain.com/insights/topics/global-private-equity-report/
Bain & Company, "Field Notes from the Generative AI Insurgency in Private Equity," Global Private Equity Report 2025 — $3.2T AUM survey; ~20% operationalized; Apollo/Cengage and Vista results; AI as "a tool in service of strategy." https://www.bain.com/insights/field-notes-from-generative-ai-insurgency-global-private-equity-report-2025/
Bain & Company, 2025 Private Equity Report (summary via Chronograph) — margin expansion as necessity; AI as efficiency lever. https://www.chronograph.pe/top-takeaways-from-bains-2025-private-equity-report/
RSM US, "Middle Market Firms Rapidly Embracing Generative AI, But Expertise Gaps Pose Risks: RSM 2025 AI Survey" — 91% adoption; 39% lack in-house expertise; ~1 in 4 fully integrated. https://rsmus.com/newsroom/2025/middle-market-firms-rapidly-embracing-generative-ai-but-expertise-gaps-pose-risks-rsm-2025-ai-survey.html
World Economic Forum, "It's time for AI's mid-market business moment" — the "missing middle"; weak IT infrastructure; up to 95% of pilots failing. https://www.weforum.org/stories/2026/01/ai-mid-market-business-growth/
"A Framework for the Adoption and Integration of Generative AI in Midsize Organizations and Enterprises (FAIGMOE)," arXiv — mid-size talent and resourcing constraints. https://arxiv.org/pdf/2510.19997
"What Is an AI Value Creation Plan for PE Portfolio Companies?" — hold-period sequencing; 6–12 month / 18–36 month impact timelines (citing Bain for the 5.8-year average hold). https://aiassemblylines.com/post/ai-value-creation-plan-pe-portfolio-companies
