Why are advisory firms asking managers about AI exposure now?
The first half of 2026 gave allocators a real-time stress test of software valuations. After Anthropic released plug-ins for its Claude Cowork agent on January 30, 2026, the S&P 500 software and services index lost about $830 billion in market value the following week, according to Reuters, with legal, data, and information-services companies such as Thomson Reuters and RELX among the steepest decliners. The market was not marking down software as a category so much as asking, company by company, which product an AI agent could replace.
Part of that decline has since been recovered. The iShares Expanded Tech-Software Sector ETF (IGV) posted a year-to-date NAV total return of 2.3% as of October 1, 2026, according to iShares, and a trailing one-year NAV total return of negative 17.3%. Aventis Advisors’ tracking of public SaaS companies shows the median EV/revenue multiple bottomed at 3.2x in June 2026 and recovered to 4.6x by August 2026, still well below recent levels.
Private markets have followed a different pattern. PitchBook’s Q4 2025 First Look reported that AI companies received more than half of the $512.6 billion in global venture deal value in 2025, and Carta’s 2025 review found median AI valuations 38% above non-AI valuations at Series A and 193% above at Series E and later. Meanwhile, Carta reported a down-round rate of 11.4% in Q1 2026, comparable to 2019 and 2020, even as it noted that early-stage SaaS valuations had softened. Public software multiples reset sharply in 2026, while private round pricing largely did not, and that gap is where manager judgment and disclosure matter.
Corporate operating data has not yet confirmed the scale of the market move. McKinsey’s August 2026 survey found that 37% of respondents attributed any EBIT impact to AI, essentially flat year over year, and that the share of organizations expecting AI-related workforce declines in the prior survey was roughly double the share that later reported them. Markets repriced AI risk faster than company results have confirmed it. That makes AI exposure a diligence question rather than a settled verdict.
What should you ask a venture or growth equity manager regarding AI exposure?
Allocators do not yet have a standard instrument for this. The ILPA Due Diligence Questionnaire was last updated in 2021. AIMA reported in September 2025 that 29% of surveyed institutional investors already included generative AI questions in their manager questionnaires. However, those questions focus on a manager’s own use of AI, covering oversight, data privacy, and intellectual property, rather than on how exposed the underlying portfolio companies are. The six questions below are designed to address that gap.
How should a manager evaluate where a portfolio company’s value comes from?
The useful question is not whether AI can perform the task a company performs. In most software categories, it increasingly can. The useful question is whether the source of the company’s value survives once it does. Four sources tend to hold up: proprietary data generated through the company’s own operations, trusted relationships with end customers, network effects among users, and high-stakes judgment where the cost of an error is high enough to warrant accountability. The exposed profile is the inverse: narrow workflow automation, value that consists mainly of speed or functional output, and no data the company owns.
Proprietary data is better treated as a claim to be tested than as a moat to be assumed. Andrei Hagiu and Julian Wright argued in the January–February 2020 issue of Harvard Business Review that data creates a durable advantage only under specific conditions: its value must be high relative to the product’s stand-alone value; its marginal value must decay slowly; it must not become obsolete quickly; and the improvements it enables must be hard for competitors to imitate. Martin Casado and Peter Lauten of Andreessen Horowitz made a related point in 2019, arguing that most claimed data network effects are really scale effects, with costs rising as each additional piece of data becomes harder to acquire.
A well-supported answer is specific. The manager can name the portfolio companies that meet these conditions, explain why, and identify those that do not. A response that describes the entire portfolio as protected by proprietary data, without company-level distinctions, warrants follow-up.
How exposed is the portfolio to per-seat pricing?
Per-seat pricing is a direct transmission channel from AI adoption to software revenue. If an AI agent enables a customer to run a function with fewer people, a vendor billing per user can lose revenue even when the customer keeps the product and values it more. A Bain & Company analysis of more than 30 incumbent SaaS vendors, published in October 2025, found that roughly 35% had simply raised per-seat prices to bundle AI features, about 65% had layered an AI usage meter on top of seat pricing, and none had fully shifted to usage- or outcome-based pricing.
The shift is real but gradual. Kyle Poyar’s 2026 State of B2B Monetization survey, conducted in April and May 2026 and covering more than 230 software companies, found that the share of respondents using hybrid pricing rose from 25% a year earlier to 37%, while 29% of companies with more than $150 million in ARR still relied on per-seat pricing. The survey draws on a self-selected newsletter readership and should be read as directional. On the demand side, McKinsey’s August 2026 survey found that 32% of respondents had decided against buying at least one software product because they could build it internally with AI coding tools.
A credible answer quantifies the exposure: the share of portfolio ARR priced per seat, the trend in seat counts at renewal, and each company’s pricing plan that tracks the value it delivers. A manager who cannot produce these figures may lack the product telemetry needed to manage the transition, which Bain identifies as a common gap among incumbent vendors.
Does customer mix change the timing of AI risk?
Customer mix shapes how quickly pressure appears, though the gap between segments is narrower than commonly assumed. SaaS Capital’s 2025 survey of more than 1,000 private B2B SaaS companies found a median gross revenue retention of 95% among companies with annual contract values above $250,000, compared with roughly 91% below that level. The same survey also found that in every contract-size band except the largest, at least a quarter of companies were experiencing net revenue contraction.
Larger enterprise customers are not necessarily slower to adopt AI, however. McKinsey’s 2026 survey found that large organizations are scaling AI agents faster than smaller ones, at 40% versus 22%. For enterprise-focused software, the more likely form of AI pressure is not outright churn but smaller renewals: fewer seats, a narrower scope, and tougher pricing negotiations. A strong response segments retention by customer segment and cohort and identifies which material contracts renew in the next 12 to 24 months.
Are portfolio companies using AI to grow or only to cut costs?
When AI frees capacity within a company, management faces a choice. The capacity can be redeployed into product development, customer coverage, and new services, or it can be converted into margin through headcount reduction. Both are legitimate, but they create different kinds of value. Cost reduction improves near-term margins; expanding what the company delivers to customers is more likely to build durable enterprise value because it strengthens the reason customers pay.
The evidence on which path companies are taking is incomplete. McKinsey’s 2026 survey identifies roughly 6% of respondents as AI high performers and reports that they pursue growth or innovation alongside efficiency and redesign workflows rather than automating existing ones; the finding is self-reported and correlational, so it does not establish that a growth orientation causes better results. Challenger, Gray & Christmas reported that through September 2026, employers cited AI in 120,136 announced US job cuts, approximately 21% of all announced cuts, though those figures reflect stated reasons in announcements rather than measured effects.
A credible answer includes specific examples of capacity redeployed to serve customers, with operating metrics to demonstrate it. A manager whose AI narrative consists only of margin expansion is describing a cost program, which may be valuable but is easier for competitors to replicate.
How are software positions marked against the 2026 public reset?
Private valuations adjust to public markets with a lag, and the 2026 reset was significant. Software Equity Group noted in its 2Q26 report that, despite the broad decline, the market is reserving premium valuations for companies that combine strong operating performance with defensible competitive advantages. That dispersion reflects operating quality rather than AI exposure directly, but it suggests that a single sector-wide markdown would misstate both the stronger and the weaker positions.
The relevant diligence points include which public comparables the manager uses, when the marks were last updated relative to them, and whether positions the manager considers AI-exposed are valued differently from those it considers resilient. A well-supported answer presents the comparable set and its date. A portfolio whose software marks have not moved since 2025 warrants a clear explanation of why.
Can the manager substantiate its own AI claims?
Managers who describe portfolios as AI-native, AI-enabled, or AI-resilient are making representations that regulators now scrutinize. The SEC Division of Examinations stated in its FY2026 priorities, released November 17, 2025, that it will review registrant representations regarding AI capabilities for accuracy and assess whether operations and controls are consistent with disclosures. In March 2024, the SEC settled charges against two advisers, Delphia (USA) Inc. and Global Predictions Inc., for $400,000 in combined penalties for false statements about their use of AI, including charges under the Marketing Rule. A well-supported answer is documentation: the analysis behind each AI-related statement in the manager’s materials, available on request.
What does a strong set of answers look like overall?
No manager will have complete answers to all six questions because the evidence is still forming. What distinguishes a credible response is specificity at the company level, a willingness to name the positions that are exposed, and consistency among the manager’s AI narrative, its pricing data, and its marks. A manager who describes every holding as an AI beneficiary is offering a narrative rather than an analysis.
Public markets marked down software stocks on an AI headline in 2026, and the underlying question remains unsettled. For allocators, the distinction that matters is between managers who have examined AI exposure company by company and those whose software marks have simply not yet been tested.