The AI exit test in software is now evidentiary: product and adoption, or a client-book bid

"Real AI position" is doing heavy lifting in the Hg partner's formulation, so it is worth separating the two halves of the test. Benedikt Joeris says buyers are scrutinising whether a business has one, and that the companies able to show actual products and adoption are being rewarded ^[2]. Product can be demonstrated in a room. Adoption has to be evidenced by someone who is not the seller.

GTreasury is the version of that where the evidence held. Hg bought it for a reported USD 400m and sold it to Ripple for USD 1bn two years later, with the outcome largely attributed to an agentic AI buildout in the interim ^[4]. On the reported entry price that is 2.5 times money and USD 600m of headline uplift ^[5]. The buildout was not the thing that got paid for; a buyer's willingness to underwrite the buildout was.

Visma shows what the same market looks like when nothing has printed. Hg took control in a USD 5.3bn buyout in 2017, touted as the largest ever for a European software company ^[6]. It began exploring a listing as early as August 2025 and had more than ten banks engaged by that December, according to Mergermarket ^[7]; Bloomberg reported in March that an IPO was unlikely within the year, citing market volatility ^[8]. That is roughly seven months from process launch to a public report of no listing ^[9], with a bank syndicate already assembled. Process length, not deck quality, is the variable a limited partner can see.

The harder mechanism sits in Elizabeth Todd's description of the rollup channel. At-risk software companies are being acquired to absorb proprietary data, regulatory moats and client networks ^[13], and the Ropes & Gray partner's warning is about the denominator: for distressed SaaS, the valuation achieved is likely to value client books and data rather than actual revenue, which may not be a valuation that works for the current owners ^[15]. Holding marks and leverage packages built on a revenue multiple do not translate into a data bid. Todd also notes that acquisitions are easy relative to integrating them, and that integration is the real challenge ^[16]. Clio's purchase of vLex at a USD 1bn valuation gave Oakley Capital its exit ^[14]; the integration bill arrives later and lands on the buyer.

Then there is the liquidity assumption underneath all of it. MSCI's Uday Karri puts the listed share of the venture-backed companies MSCI tracks globally at 6.7% at the start of 2026, the lowest in decades, and expects a correction toward the historical 18% in less than a year on the strength of regulatory filings ^[10]. That is close to a 2.7-fold increase in the listed share inside twelve months ^[11]. Karri also concedes that current venture performance is mostly paper marks pending an IPO floodgate ^[12]. The forecast and the Visma process are describing the same window from opposite sides.

For a software leader, the practical reading is that the standard is set by whoever is in the room, because Joeris's phrasing stops at products and adoption without defining the metric ^[2]. Todd's read is that buyers will still pay reasonable multiples for the right asset, selectively, and that nobody is buying software broadly on the dip ^[17]. Between a 2.5x and a client-book bid, the difference is not the story about AI. It is whether a third party can verify use.