The end of the license-based model
For more than twenty years, the valuation of software companies has been based on established metrics such as annual recurring revenue, customer retention and EBITDA multiples. According to Agenda Digitale, this framework now appears outdated: the spread of generative AI and autonomous agents has triggered what operators call SaaSpocalypse, a reshuffling of valuations and business models that hits traditional software first, along with system integrators and outsourcing service providers. Value no longer grows with the number of licenses or active users, but with the ability to automate work and produce concrete results.
The price is tied to the outcome, but it must be demonstrated
The push comes from customers who, according to the source, no longer want to pay simply for the use of an application or for the hours of developers and consultants. The logic shifts from Copilot, in which technology supports the person, to Autopilot, in which the system completes the task within set and verifiable limits. This gives rise to outcome-based pricing models: Eudia in legal services works at a fixed price per engagement, while Crescendo in customer support charges for each request resolved.
The condition, often implicit, is that the outcome be documentable: the client must know what the system did, on which information, and with what authorization. Without this proof, outcome-based pricing remains a commercial promise; with it, it becomes a contract.
New metrics and the limits of automation
Economic and financial analysis is changing too. In mid-market due diligence, Revenue Per Employee is becoming central: the best traditional SaaS companies generate between $200,000 and $600,000 per employee per year, while ElevenLabs stands between $1 million and $1.5 million and Cursor, with more than $4 billion in annualized revenue and a headcount not made public, would be around $10 million according to estimates. The figure, however, measures how much is collected, not how much is retained: Gartner forecasts that spending on inference will continue to grow, because an agent consumes 5 to 30 times more tokens than a chatbot.
Gartner itself estimates that more than 40% of projects based on AI agents will be canceled by the end of 2027 due to costs, unclear value or inadequate controls. In regulated sectors, the third cause weighs most heavily: a probabilistic model does not guarantee zero errors, and downstream controls catch the damage once it has already occurred. Hence the push toward vertical models, expected to account for more than half of enterprise models by 2028, and deterministic components that establish what can be executed while leaving a verifiable trace.
Investors and the Italian landscape
Governance thus becomes a market-access requirement, even after the postponements of the AI Act introduced by the Digital Omnibus in force since July 27, 2026. Private equity funds are focusing on AI Rollups: acquiring traditional companies at low multiples and repositioning them as AI-driven businesses. Software Equity Group's 2026 survey finds that 83% of buyers recognize a premium for AI-integrated targets, but only 26% of the companies examined in 2025 truly are. A risk remains: if everyone automates, savings become discounts at renewals. In Italy, the 2026 Report of the AI4Innovation Observatory of the Politecnico di Milano indicates that only 24% of companies have a formalized AI roadmap.
