Enterprise companies now reevaluate their AI vendors every six months or even more frequently, according to new research from venture capital firm Madrona published this week. The finding reveals a sharp break from traditional enterprise software buying, where multi-year commitments have long been the norm. The shift creates what the report calls a "fast in, fast out" dynamic that leaves startup revenue far less predictable than in the SaaS era.
The Madrona survey of 150 enterprise IT professionals found that 74% intend to increase their AI spending over the next 12 months, while the remainder plan to keep budgets flat. Yet fewer than half of AI pilot projects at these companies ever reach full production, though that marks an improvement from last year when MIT reported a 95% failure rate for enterprise AI initiatives measured by return on investment. The report also highlights that 77% of enterprises now reassess their AI vendors either every six months or on a continuous rolling basis. Separate research from Andreessen Horowitz that surveyed 50 technical AI buyers revealed that more than half prefer AI pricing tied to outcomes or work completed, rather than usage metrics like token consumption.
According to the Madrona report, the frequent vendor reassessment cycle is "fundamentally different from traditional enterprise SaaS, where multi-year contracts provided a moat of inertia." The authors write that "switching costs are lower and the re-evaluation cadence is relentless" in enterprise AI. Andreessen Horowitz partners Tugce Erten and Sarah Wang argue that pricing "around the recognizable work" makes AI products "economically valuable to both sides" by tying fees to concrete outputs like reports processed, tickets closed, or leads generated rather than abstract usage measures.
The underlying issue stems from AI startups struggling to settle on pricing models that prove their value to corporate buyers, the report explains. Token-based usage fees mirror SaaS-era business models, where enterprises already know they need a particular tool and simply pay based on scale. AI products lack that established necessity, so outcome-based pricing helps demonstrate worth by connecting fees directly to measurable business results. This pricing uncertainty combines with low pilot success rates to fuel constant vendor switching. Even when an AI product graduates from pilot to full deployment, companies no longer treat that adoption as a long-term commitment the way they did with traditional enterprise software.
The research suggests AI has introduced an era of perpetual enterprise experimentation, which lowers barriers for startups trying to break into corporate accounts but simultaneously means landing a contract no longer guarantees stable recurring revenue. Market researcher IDC forecasts that companies will spend $4.25 trillion on technology in 2026, driven almost entirely by AI investment, yet the report indicates those dollars remain up for grabs quarter after quarter. Whether enterprises will eventually return to their historical pattern of long-term vendor relationships remains an open question. For now, the annual recurring revenue figures that AI startups use to demonstrate growth carry less security than at any previous moment in enterprise software history. Investors and founders alike will need to adjust expectations around customer lifetime value and churn in a market where even successful deployments face routine reconsideration.