The Exit Backlog is a Product Problem
Private equity is sitting on roughly 33,000 unsold portfolio companies, according to Bain's 2026 midyear report. Median holding periods are running near seven years. Distributions as a share of NAV are at a four-year low, and about one in five LPs say they are pulling back on buyout allocations because of it.
The standard explanation is macro: rates, financing costs, a bid-ask spread between sellers' marks and buyers' prices. All true. For B2B software, it is also incomplete.
Buyers are pricing in a new risk: whether the product will still matter when agents do the work.
Look at what happened in the first half of this year. The S&P 500 software index fell 27% between January and late March. Buyout firms marked down US software holdings by 8.9% in a single quarter. PE capital deployed into software fell by roughly two thirds year over year in Q2. Public software has since rebounded, but the private exit window has not reopened at the same speed. More than a quarter of PE-owned software companies were acquired five or more years ago.
Those assets were underwritten on a thesis that looked durable in 2019 or 2020: seat-based pricing, a workflow UI users live in all day, and switching costs built on configuration and integration. Agentic AI puts pressure on all three at once. Fewer humans in the workflow means fewer seats. An agent that completes the task means less time in the UI. And an AI-native entrant with a semantic data layer can cut implementation time enough to erode the switching cost.
So part of the gap between the seller's mark and the buyer's price is not macro. It is a durability discount, and it is specific to the product.
What buyers are actually diligencing now
In the diligence conversations I see, four questions come up fast:
Is AI in the core workflow or bolted on? A chat panel over an unchanged process is defensive, not strategic. EVERY single software business has done this — can you image what customers are thinking?
Is there a context moat? Leveragable proprietary data, customer history, and domain specific context the product can reason over are harder to replicate than a model integration.
Does the pricing model survive fewer seats? If the answer is "we haven't modeled it…" or “we’re just focused on adoption right now…” the PE buyer will model it for you, conservatively.
Is your ICP able to leverage frontier models plus their own data to accomplish the same outcome? This typically shows up in slower bookings growth, a more expensive new logo pipeline, longer sales cycles and GRR slippage.
Carta's Q3 analysis put it plainly: investors are no longer paying premiums just because revenue is growing. They want a clear AI strategy tied to revenue, strong retention, efficient growth, and a durable advantage.
The hold period is the window
Bain estimates firms now need roughly 10% to 12% annual EBITDA growth to generate returns that 5% used to deliver. Multiple expansion is not coming to the rescue. That leaves operational value creation, and for a software asset, product is the biggest lever.
The good news: an extended hold is time. Twelve to eighteen months is enough to move a product from "AI feature on the roadmap" to "AI in the core workflow with measurable customer outcomes," if the work starts with product strategy rather than a model vendor selection.
The bad news: the companies that treat the backlog as a market-timing problem will use that time waiting.
Key takeaways
The exit backlog is partly an AI-durability discount, not only a macro problem.
Buyers are testing four things: AI in the core workflow, a context moat, pricing that survives fewer seats and customer DIY AI.
An extended hold is a product re-underwriting window. Use it or the buyer will price the gap.
Provex Advisory provides CPO-equivalent product leadership to PE-backed B2B SaaS companies at an inflection point. If a portfolio company's AI story needs to hold up in diligence, let's talk - book a discovery call today.
