When Does AI Change the Exit Story?

McKinsey published the analysis on June 23. It covers 471 privately held companies that took PE equity or debt from 2023 onward, with revenue between $1 million and $250 million, across 30 countries and 31 industries. Companies were sorted into four AI maturity levels using public evidence: websites, regulatory filings, product documentation, and press releases.

The four levels, and where the companies sit:

In software specifically, level three trades at 24x and level four at 33x. McKinsey's own reading is that "markets do not materially differentiate between companies that use AI for productivity and those that integrate it into their operating models," and that valuations rise significantly only when AI is embedded in the offering.

Two cautions before this reaches a board. The study shows association rather than causation, and companies at the higher levels may be stronger businesses for reasons the classification never measured. The absolute multiples are also not market benchmarks, because McKinsey limited the sample to companies with revenue between $1 million and $250 million and valuation-to-revenue multiples between 5x and 300x. What is useful is the pattern across the four levels. On the first two, McKinsey is blunt: "the difference between the median revenue multiple for level one and level two companies is negligible."

For portfolio planning, AI investment usually serves one of three economic objectives.

Operating-model AI improves how the company works. Engineering copilots, automated support, sales research, document processing, internal workflow automation. The financial case is productivity, capacity, or margin, and where the savings are realized rather than reinvested, some of that benefit reaches EBITDA. Median revenue multiple: 14x.

Product AI becomes part of the product or service the customer receives. McKinsey links this level to pricing optimization, personalization, improved win rates, and lower churn. The case runs through differentiation, retention, and growth. Median revenue multiple: 20x.

New business building creates a revenue stream the company did not have. It was the smallest group in the sample, 21 of 471 companies, and the one with the widest gap to the level below it. Median revenue multiple: 31x.

The distinction is whether AI improves how the company operates and delivers the offering it already has, or whether AI becomes part of the product or service the customer receives. An AI sales agent re-engaging dormant accounts is operating AI in McKinsey's classification, even though a customer is on the other end of it. The owner and the budget often make the objective clear.

The allocation should follow the value creation plan and the remaining hold period. A company with a margin problem and an eighteen-month hold may reasonably put most of its AI spending into operating efficiency. Another may need differentiation or a new revenue line more urgently.

The board should see that allocation. These investments can sit in different budgets and different parts of the organization. Product AI will often sit in the product roadmap, while operating AI may be funded through IT, individual functions, or a transformation program. Deloitte's 2026 technology leadership study, covering 662 senior technology leaders, found that 71% of them work in organizations with five or more C-suite technology leaders, with responsibilities split across data, cyber, operations, product development and engineering. Deloitte's own conclusion is that "reporting relationships alone will not determine whether AI scales successfully."

Some companies handle this with a steering committee, a transformation office or a CFO-level portfolio view. Without a consolidated view of that kind, the board may not see the full allocation in one place.

213 of the 471 companies sat at the operating-model level and another 137 below it, so nearly three-quarters had not reached the point where AI was part of what the customer buys. Operating-model AI was the largest category in the sample, and the valuation data shows little separation between levels one and two.

PwC sees a related shift on the deal side. Its midyear technology outlook says AI has moved "from investment thesis to revenue test," with diligence paying more attention to monetization and defensible workflows.

For companies in FY27 planning, the allocation should be made explicit before the budget is approved.

Infrastructure, vendor relationships, deployment experience, and engineering capability built for internal AI can carry into customer-facing AI, and McKinsey's own view is that companies advance through the levels gradually.

Customer-facing AI adds work around product economics, quality evidence, customer data, and commercial commitments. That work belongs in the product roadmap rather than being assumed to have been covered by prior internal AI investment.

Management should be able to show the owner and the material investment or capacity committed to each major initiative.

Product AI may sit in the product roadmap while operating AI is funded through IT, individual functions or transformation programs. Bring the major initiatives together for the operating review, grouped by what each is meant to achieve.

For each line, show the major initiatives, the accountable owner, the material spend or engineering capacity committed, and the expected value. Exact dollar attribution may not always be practical, particularly where AI work is embedded in an existing product roadmap. The board needs enough visibility to understand the allocation and make a decision about it. There is no portfolio-wide target ratio either. What matters is that the allocation reflects the value creation plan, the financial position and the remaining hold period, and that it was decided rather than inherited.

Each one should say what management expects it to change. For operating-model AI, that may be productivity, capacity, or margin. For product transformation, it may be adoption, retention, pricing, or revenue. For new business building it should be a revenue stream with economics of its own. An initiative with no stated business objective is still an experiment, and it should be presented to the board as one.

Deloitte's view is that "traditional project-based funding is often too rigid for AI because that approach is often designed for predictable technology investments," pointing to variable consumption costs, rapid experimentation cycles, changing vendor economics, and uncertain timelines for value. Some internal initiatives will return less than the plan assumed. Some product work will find adoption faster. Review the portfolio during the year and move capacity toward what is producing evidence, rather than holding the allocation fixed until the next planning cycle.

213.

That is how many of the 471 companies in McKinsey's sample were using AI primarily to improve their operating model. It was the largest group in the study.

For each portfolio company, look at its own allocation. How is the FY27 AI investment distributed across improving the operating model, changing the product, and building something new? There is no universal right mix. A company approaching exit with a margin gap should look different from one whose thesis needs product differentiation or a new growth line. The operating review should show the actual allocation and why management chose it.

Portco Brief is a weekly briefing for PE operating partners and portfolio company executives focused on technology, AI, and value creation. If this was forwarded to you, subscribe at portcobrief.com.

Sources

McKinsey & Company, "Beyond productivity: How AI creates value in private equity", June 23, 2026. Analysis of 471 PE-backed companies across 30 countries and 31 industries, deals completed from 2023 onward.

PwC, "Technology: US Deals 2026 midyear outlook", June 17, 2026.

Deloitte, "Rewiring the enterprise operating model for AI scale", June 29, 2026. Based on the 2026 Global Technology Leadership Study, 662 senior technology leaders surveyed December 2025 to February 2026.