Enterprise AI revenue is proving less secure than expected, with companies reviewing suppliers frequently and fewer than half of pilot projects progressing into full production, new research suggests.
A survey of 150 enterprise IT professionals by venture capital firm Madrona found that 74% expect to increase their AI budgets over the next 12 months. The remainder said they planned to keep spending at its current level.
Despite that appetite for investment, respondents said fewer than half of their AI trials ultimately become fully operational systems. The figure represents an improvement on a widely cited MIT finding last year that 95% of enterprise AI projects failed to deliver a return on investment.
The more significant challenge for AI start-ups may come after a product has been adopted. Madrona found that 77% of businesses re-evaluate their AI suppliers every six months or continuously, creating a far less predictable revenue base than the multi-year contracts associated with traditional enterprise software.
Enterprise AI contracts face frequent reviews
“This creates a ‘fast in, fast out’ dynamic that is fundamentally different from traditional enterprise SaaS, where multi-year contracts provided a moat of inertia,” Madrona said in its report.
It added that switching costs were lower in enterprise AI and that supplier reviews were taking place with “relentless” frequency.
The findings raise questions over the durability of the annual recurring revenue figures reported by some of the sector’s fastest-growing companies. Trial budgets helped drive the initial corporate AI surge, while large contracts enabled some start-ups to report exceptionally rapid growth after winning enterprise customers.
Those agreements are now less likely to guarantee long-term income, even when a product has moved beyond testing and is being used within a business.
Pricing is another source of uncertainty. Research by Andreessen Horowitz, based on discussions with 50 technical AI buyers, found that more than half preferred fees linked to work completed or business outcomes rather than measures such as the number of tokens processed.
That approach would move AI pricing away from the usage-based model common in software-as-a-service. Instead of charging for consumption, suppliers could bill for tasks such as reports completed, customer-support tickets resolved or leads generated.
Andreessen Horowitz partners Tugce Erten and Sarah Wang said pricing based on “recognizable work” could make the value of an AI service clearer to both the customer and the supplier.
The shift has created more opportunities for start-ups to persuade companies to experiment with new technology, but it has also weakened the assumption that winning an enterprise contract secures dependable recurring revenue. Whether corporate buyers return to longer-term commitments will depend on how consistently AI products demonstrate measurable value.
