a16z VC Jennifer Li warns founders against chasing vanity ARR metrics in the AI space, emphasizing product quality and sustainability instead.
Why ARR became the wrong scoreboard for AI startups
Annual recurring revenue is easy to say and hard to interpret. In AI, a company can post a large ARR figure quickly because early adopters sign up to experiment, budgets for anything labeled "AI" are unusually loose, and a handful of big contracts can inflate the headline. Jennifer Li's point is that this number, on its own, tells you almost nothing about whether the business will still be here in a few years. It measures money that arrived, not money that will keep arriving.
The distortion matters because founders start optimizing for the metric they broadcast. If the goal is a bigger ARR announcement, the fastest path is discounting, aggressive contracts, and features that demo well but don't get used. None of that builds a company; it builds a slide.
What a durable AI product actually looks like
The alternative Li points to is boring by comparison: build something people keep using because it does the job better than the alternatives. Durable revenue shows up when customers renew without being chased, expand their usage on their own, and would genuinely miss the product if it disappeared. Those signals are slower to accumulate than a signup spike, but they're the ones that survive a tighter budget cycle.
For AI products specifically, quality also means the output holds up in real workflows, not just in a controlled demo. A tool that's impressive once but unreliable on the tenth try trains users to stop trusting it, and that erosion happens quietly beneath a still-growing ARR line.
Metrics that tell you more than a headline number
If ARR alone is misleading, founders need a fuller picture of health. The useful signals are the ones that reveal whether revenue is sticky and whether growth is paying for itself rather than being bought.
- Retention and churn: are customers staying past their first contract, or cycling out after the novelty fades?
- Actual usage: are the people who signed up using the product regularly, or is it shelfware behind a paid seat?
- Expansion: does spend grow inside existing accounts without a new sales push?
- Unit economics: does each customer eventually cost less to serve and support than they pay, including inference and compute?
- Payback period: how long before a customer's revenue covers what it took to acquire and onboard them?
How founders can act on this now
The practical shift is to stop treating a large ARR announcement as the milestone and start treating retention and real usage as the thing worth reporting internally. That changes what teams build: fewer features aimed at closing a flashy contract, more work on reliability, onboarding, and the parts of the product customers touch every day. It's less quotable, but it's what keeps renewals coming.
Sustainability also means matching spend to genuine demand rather than to a growth target. Chasing an insane ARR figure often requires burning capital to manufacture numbers that don't reflect a healthy business. Li's advice is to relax that pressure — build a product good enough that the revenue becomes a byproduct instead of the point, and let the durable metrics, not the headline, tell you how you're doing.