Enterprise AI contracts offer startups less durable recurring revenue

Enterprise AI spending is rising, but adoption is not creating the long-term contractual certainty that many software startups have historically expected. Venture capital firm Madrona found that 77% of enterprises reassess their AI vendors every six months or on a rolling basis, even after an AI product has moved beyond a pilot and into deployment.

The finding comes from a survey of 150 enterprise IT professionals. Madrona reported that 74% plan to expand their AI budgets during the next 12 months, while the remaining respondents expect to keep spending flat. At the same time, fewer than half of AI pilots reach full production, showing that budget growth does not automatically translate into durable supplier relationships.

Vendor reviews reshape the meaning of AI ARR

Madrona describes enterprise AI procurement as a “fast in, fast out” market. This differs from traditional enterprise SaaS, where multi-year contracts and operational dependence often created significant inertia for customers. In AI, the report says, switching costs are lower and suppliers face an unrelenting re-evaluation cycle.

That dynamic complicates the annual recurring revenue figures reported by fast-growing AI companies. Enterprise trials helped fuel the first wave of AI demand, and contracts can still enable a startup to move rapidly from no revenue to millions of dollars in annualised revenue. However, a production deployment is no longer equivalent to a durable commitment from the customer.

The backdrop remains substantial. IDC predicts companies will spend $4.25 trillion on technology in 2026, with AI driving much of that increase. Madrona’s data also indicates progress relative to the weak returns recorded in an MIT assessment last year, which found that 95% of enterprise AI projects had failed in terms of return on investment. Even so, a production rate below half leaves many experiments without a lasting operational role.

Buyers seek prices tied to completed work

Pricing is another source of uncertainty. Andreessen Horowitz surveyed 50 technical AI buyers and found that more than half wanted pricing connected to work performed or outcomes achieved, rather than usage measures such as tokens consumed.

For established SaaS categories, usage-based pricing can map relatively clearly to employee numbers, data volumes or capacity. AI buyers, by contrast, may want fees linked to recognisable work: reports processed, support tickets closed or leads generated. Andreessen Horowitz partners Tugce Erten and Sarah Wang argue that pricing around such work can make the product economically valuable to both customer and supplier.

What enterprise teams and startups should take from the data

Enterprises appear more willing to test AI tools, creating opportunities for suppliers that can prove a concrete operational benefit. But the evidence suggests that a pilot conversion should be treated as the start of a continuing value test, not the end of procurement risk.

For business planning, AI vendors need to demonstrate measurable work outcomes and expect regular competitive reviews, while enterprise buyers should define the outcomes and renewal criteria that determine whether a deployment remains in place.