Until 2024, when startup founders put the line "Hit $10 million ARR in three months" on an investor deck, that number read like a promise. The term "annual recurring revenue" itself carried a built-in assumption: this contract will still be alive next year. A study Madrona Venture Capital released in 2026 contains a figure that unsettles that assumption. Seventy-seven percent of enterprise buyers said they re-evaluate AI vendors every six months or more often.
The review cycle has shrunk from annual to semiannual. What the ARR metric actually measures has changed.
Why Traditional SaaS ARR Held Up
Behind ARR's rise as a predictable metric during the SaaS growth era of the 2010s was switching cost. Once a sales team adopted Salesforce, six months of data piled up inside it, pipeline habits hardened around it, and reps finished training on it. By the time renewal came around a year later, even if a competitor showed up with better features, the cost and disruption of switching absorbed most of that advantage.
"Lock-in" can sound like a dirty word, but for vendors it was the foundation of predictable revenue. Customers had to put in real effort to leave, and that friction was what held ARR together on an annual basis.
What Changes With AI Tools
AI tools have little or none of that friction. Learning Salesforce takes months; switching away from an AI tool that drafts your sales emails takes days. Data doesn't accumulate deeply, and workflows don't lock in tightly. This is the backdrop for why buyers say they re-evaluate every six months.
According to a 2025 MIT report, 95% of enterprise AI projects failed to hit their ROI targets. Failed projects are unlikely to renew. Madrona's research shows that fewer than half of AI pilots advance to full production. That means more than half of pilot contracts never convert into ARR at all.
The ARR number itself can still climb fast, because companies are ramping up AI budgets quickly. In Madrona's survey, 74% of 150 IT professionals said they plan to increase AI spending within the next 12 months. The catch is that more of that budget flows into new pilot experiments than into contract renewals.
The difference between the two kinds of ARR doesn't start with how the contract is written — it starts with what actually gets left behind inside the customer's organization.
Where AI Pricing Models Are Splitting Three Ways
When a16z surveyed 50 AI buyers, more than half said they prefer outcome-based pricing over token-based usage billing. In effect, they'd rather pay based on how many deals the tool actually closed.
Once that preference starts showing up in real contracts, the ARR metric itself ends up measuring something different. Traditional SaaS ARR measured the fact that a contract exists. Outcome-based AI ARR measures the fact that the tool actually delivered again this month. The former can be proven with a signed contract; the latter has to be proven all over again every month.
The fact that pricing models haven't settled yet adds to the instability. For the same feature, one vendor bills per seat, another per API call, another only when a performance target is hit. When buyers say they re-evaluate every six months, the comparison itself keeps shifting more often because the yardstick hasn't been fixed yet.
What This Means for Korean Founders
IDC projects global enterprise tech spending will hit $4.25 trillion in 2026, with a large share of that going to AI. Korean companies are moving in the same direction. As AI adoption speeds up, the decision to swap out a tool that isn't working has gotten just as fast.
When a solo PM or a one-person founder signs a monthly subscription deal with a single enterprise customer, the standard for judging whether that ARR will still be there next year has shifted. In the past, you could count on inertia — "they already adopted it, so they'll probably keep using it for a while." That inertia is much thinner now.
Before looking at the numbers, there's something else worth checking when judging how solid your revenue really is: which task would stop dead tomorrow if the customer didn't use this tool, and whether that disruption outweighs the hassle of switching to a competitor. If that question is hard to answer, that ARR is likely to land back on the re-evaluation list in six months.
There's a principle business schools have taught for a long time: a good business makes it hard for customers to leave. That principle is being tested all over again in the AI tools market. In an environment where switching costs have collapsed, what actually keeps a customer isn't a clause in the contract — it's the real, felt inconvenience of working without that tool.
