Enterprise AI spending is expanding, but the revenue it creates for startups may be far less durable than traditional SaaS metrics suggest.
A [Madrona survey](https://www.madrona.com/wp-content/uploads/2026/08/Madrona-Research-Harnessing-Enterprise-Value-ROI-of-AI.pdf) of 150 enterprise IT professionals found that 74% expect to increase AI budgets over the next 12 months, with the remainder planning to hold spending steady. At the same time, fewer than half of AI pilots move into full production.
The more consequential finding for founders and investors comes after a deployment succeeds: 77% of respondents said they reassess AI vendors every six months or continuously. In other words, landing an enterprise account—and even converting a pilot—is no longer equivalent to securing a long-lived revenue stream.
The enterprise contract has changed
Traditional enterprise software benefited from inertia. Once a company standardized on a core system, migration costs, implementation work and internal training made replacement difficult. Multi-year contracts reinforced that stability.
AI products do not always carry the same switching costs. Models, applications and workflow tools are evolving rapidly, while buyers have a growing field of vendors to compare. Enterprises can test a product, use it in production, and still reopen the vendor decision shortly afterward if quality, integration, economics or governance requirements change.
That creates a “fast in, fast out” market dynamic, as Madrona describes it. It is good news for startups trying to get an initial meeting or pilot. Buyers appear more willing to experiment than they were in earlier software cycles. But it also raises the bar for keeping the account once the novelty has worn off.
For companies reporting rapid annual recurring revenue growth, that distinction matters. A surge in ARR can reflect real demand and deployed usage, yet still provide limited evidence that a customer will renew, expand or remain loyal over multiple years. The familiar shorthand of treating ARR as a proxy for predictable future revenue deserves more scrutiny in AI.
Pricing may be part of the retention problem
How startups charge could determine whether their products become embedded or remain replaceable. Research from [Andreessen Horowitz](https://a16z.com/you-are-not-a-model-dont-price-per-token/), based on a survey of 50 technical AI buyers, found that more than half prefer fees linked to completed work or outcomes rather than raw usage measures such as tokens.
That preference reflects a practical purchasing problem. Token consumption is an understandable cost input for an AI provider, but it is not necessarily a useful value metric for a business buyer. A customer can more readily assess the economics of reports processed, support tickets resolved or qualified leads generated.
Outcome- or workflow-based pricing is not automatically simple. Providers must define what counts as a completed unit of work, handle exceptions and avoid taking on risks they cannot control. Still, the research suggests that aligning pricing with recognizable business output could make renewal conversations less about model costs and more about measurable operational value.
What operators should watch
For AI startup leaders, the priority is shifting from pilot conversion alone to repeatable proof of durable value. That means tracking renewal rates, account expansion, time to value, active production usage and the operational metrics a customer uses to justify the budget. Gross ARR without those retention signals may be misleading.
Enterprise buyers, meanwhile, should treat frequent vendor reviews as an opportunity to establish clearer benchmarks: quality thresholds, human-review requirements, security controls, integration effort and unit economics. Those criteria can make experimentation more disciplined without eliminating it.
AI spending is still moving upward; IDC projects global technology spending will reach $4.25 trillion in 2026, with AI a major driver. The unresolved question is whether today’s experimentation matures into the long-term vendor relationships that powered enterprise SaaS—or whether constant reevaluation becomes a permanent feature of the AI market.
