I have become less interested in whether a portfolio company is “using AI.” Almost every management team can say yes now. Someone is using a copilot. A sales team has tried an AI prospecting tool. Finance is experimenting with document extraction. There are a few internal demos and probably a slide in the board deck.
That tells me almost nothing about value creation.
After moderating an operator-led investing panel at Max Pog’s VC conference in March, I kept coming back to the same issue. Capital is not the hard part of many businesses. Execution is. The same is true with AI. The model is rarely the bottleneck by itself. The harder question is whether a company can translate a technical capability into a better operating system for the business.
For private equity, especially in the lower middle market, that distinction matters. A portfolio company does not get credit at exit because it ran twelve pilots. It gets credit because revenue grew, margins improved, customer retention strengthened, working capital got better, or the business became more scalable and defensible.
Start with the P&L, not the model
I think the right starting point is almost boring: where does the business actually make or lose money?
If a sponsor owns a specialty distributor, the highest-value AI use case may have nothing to do with building a chatbot. It may be quote turnaround time, pricing leakage, inventory planning, inside-sales productivity, or customer-service workload. In a services business it may be utilization, proposal generation, staffing, collections, or knowledge transfer. In a healthcare company it may be prior authorization, scheduling, documentation, revenue-cycle work, or patient intake.
The use case should map to an operating metric that already matters. Otherwise it is too easy to declare victory because the technology worked.
This is also why I am skeptical of broad “AI transformation” programs that begin with a large technology roadmap. In many mid-market companies, the better move is to find two or three workflows where there is enough volume, enough pain, and enough measurable economics to justify changing how the work gets done.
The first 90 days should look more like operating diligence
If I were underwriting an AI value-creation plan during diligence or immediately after close, I would want a simple 30/60/90-day view.
In the first 30 days, map the workflows. Where are people spending time? Where are decisions slow? Where does rework happen? Which processes depend on a few experienced employees? What work is rules-based, what work requires judgment, and what data is actually available?
By 60 days, there should be a small number of use cases with owners, baseline metrics, technical constraints, and a clear definition of success. Not “improve productivity.” Something closer to reducing quote preparation from four hours to one, increasing collections throughput, improving call resolution, or increasing sales capacity without adding headcount.
By 90 days, at least one use case should be operating in a real workflow with real users. The question then becomes less about model quality and more about adoption, exception handling, economics, security, and whether the process can be repeated across the company or portfolio.
The portfolio advantage is repeatability
Private equity has one advantage that most standalone companies do not: pattern recognition across multiple businesses.
If a sponsor learns how to improve pricing, customer service, software development, recruiting, finance operations, or sales enablement in one company, there is an opportunity to turn that into a reusable playbook. The value is not the prompt library. It is the combination of workflow design, integration, governance, change management, and measurement.
That is where I think the real operating leverage sits. A PE firm does not need fifty unrelated AI experiments across ten companies. It needs a small number of repeatable capabilities that can be adapted to different operating environments.
The same logic applies to diligence. AI can compress document review, market mapping, customer-reference preparation, risk identification, and the first pass through a data room. But it should get the team to better questions faster. It should not create false confidence or replace sector judgment.
The harder part is organizational
The companies that struggle are often not short of tools. They are short of ownership.
No one is responsible for redesigning the workflow. IT owns the technology but not the business process. The business team owns the process but does not understand what the technology can do. The CFO wants ROI but the baseline was never measured. Employees are told to use AI but are not given clear guardrails or a reason to change how they work.
This is why I think AI value creation is closer to an operating discipline than a software deployment.
McKinsey’s 2026 private-markets work makes a similar point from the LP side: value-creation capability has moved up the list of what LPs care about when selecting managers. That makes sense. In a market where cheap leverage and easy multiple expansion are less reliable, the sponsor has to create more of the outcome itself.
What I would want to see
When a portfolio team tells me an AI initiative is working, I would want five things: a baseline, an owner, a changed workflow, an adoption measure, and a financial or operating result.
I do not need every project to create an immediate EBITDA line. Some will reduce cycle time, improve service, or create better data before the financial impact is visible. But the chain from technology to operating result should be clear enough that a board can decide whether to scale it.
The opportunity in lower-middle-market private equity is real because many companies still have fragmented systems, manual processes, and constrained management bandwidth. That also makes implementation harder than a polished demo suggests.
The firms that get this right will probably look less like AI tourists and more like good operators. They will pick fewer use cases, tie them to the economics of the business, put somebody in charge, and build a playbook they can reuse.
That is a much less exciting story than “AI transforms everything.” I think it is also much closer to how value will actually be created.
Sources and notes
McKinsey, Global Private Markets Report 2026: https://www.mckinsey.com/industries/private-capital/our-insights/global-private-markets-report – Private-market conditions, operational value creation and LP manager-selection priorities.
Bain & Company, Global Private Equity Report 2026: https://www.bain.com/insights/topics/global-private-equity-report/ – Value creation, fundraising and the post-cheap-capital operating environment.
Deloitte, State of AI in the Enterprise 2026: https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html – Enterprise scaling, production adoption and AI readiness.
Related reading: Why So Many Enterprise AI Pilots Stall Before Production | AI Readiness Starts With the Operating Model