two ai companies walk into a fundraise. both report $5m arr. one trades at 12x, one trades at 4x. the slides look identical. the cap tables don't.
the difference is what the arr number actually contains. one is contracted revenue with 130% nrr and 18-month commitments. the other is the trailing 30 days of usage annualized, with no contracts, no minimums, and a 6-month logo retention curve that bends the wrong way. they are not the same business. investors stopped treating them as the same business around q3 of 2024.
if you're an ai founder and your arr number is the trailing month times twelve, you have a revenue quality problem that the headline number is hiding. the meeting where you find this out is the meeting you wanted to win.
the question that changed in 2024
through 2023, ai companies got the saas treatment. arr was arr. a $5m number was worth roughly the same as a $5m saas number, sometimes more because growth was faster. then a few high-flying ai companies hit the renewal wall — customers who signed up to try the product, used it for three months, and didn't renew because the next quarter's product was a different thing.
by mid-2024, the question every series A and B partner started asking was "what's the contract length and what's the logo retention at six months?" not arr. not growth rate. retention quality.
that question changes the conversation because it forces the founder to separate three things — usage that has been contracted, usage that has been billed but not contracted, and usage that's been counted in arr but is just last month annualized. those are three different revenue qualities. they trade at three different multiples.
the chart above is two ai companies at the start of year one. one is at $5m of headline usage-annualized arr with a churn problem. the other is at $2m of contracted arr with a meaningful expansion engine. by year three, the smaller-looking company is bigger. the bigger-looking company spent three years explaining itself.
the four ways ai arr gets inflated
each of these is technically defensible. each one inflates the headline by 1.5-3x compared to the saas convention.
usage-month annualization. the customer used the product for $40k in the last 30 days. you book $480k of arr. saas convention requires a signed commitment for the year. the customer can drop to zero next month and the arr line stays healthy until the dashboard catches up six weeks later.
peak-month annualization. the customer hit $40k in one month and averaged $25k for the prior quarter. you annualize the peak. the customer would have to grow into it to justify the number. they usually don't.
credits booked as revenue. the customer prepaid $100k in credits in january. the credits burn over the next year, but the full $100k lands on the arr line on day one. it's revenue. it isn't recurring.
no contract length disclosed. the customer pays $5k/month with no notice period. you book $60k of arr. saas arr is contracted revenue. ai arr in 2025 became a month-to-month rolling number that founders learned to call arr because the word sounded the same.
the test investors actually run
three numbers, in order. partners learned this in 2024 and most founders haven't caught up.
logo retention at six months by cohort. of customers who signed up in january, how many are still paying in july? for ai, 50-70% is common, 80% is excellent, under 50% is a churn-and-burn product investors discount aggressively.
gross revenue retention by cohort. of january cohort dollars, how many are still being collected in july? gross retention isolates churn. for ai, 85% is good, 70% is suspect, under 60% means the product found product-market interest, not fit.
net revenue retention by cohort. gross retention plus expansion. 130% nrr means existing customers are compounding revenue you already paid to acquire. 80% nrr means every new logo replaces a chunk of last year before growth shows up.
the math, on two companies
company a — $5m headline arr, no contracts, last month annualized. logo retention at six months is 55%. gross revenue retention is 68%. nrr is 75%. growing fast on new logos. losing customers nearly as fast at the back end. at 2024 ai multiples this traded at 8x. at 2026 multiples with retention transparency, it trades at 3-4x.
company b — $2m contracted arr, 12-month commitments, logo retention 88%, gross revenue retention 92%, nrr 125%. fewer customers, each one expanding. at 2026 multiples, this trades at 10-14x.
both report $5m and $2m. one is worth more in absolute dollars at exit, not just on multiple. the smaller company crosses $7m of arr by year two from expansion alone, while the bigger one spends eighteen months replacing churn.
the question isn't how much arr you have — it's whether the customers in it will still be customers six months from now, and whether they'll be paying you more.
what to report instead of arr
the founders raising in 2026 stopped leading with arr and started leading with the three retention numbers and the contract mix. logo retention at six months, gross revenue retention, nrr. then the contracted vs uncontracted split — "$3m contracted, $2m usage-annualized, here's why we count them separately."
it's a longer slide. it's also a slide that doesn't get questioned for forty-five minutes. the investor sees an honest number. the founder skips the cross-examination.
how zift handles this
zift computes logo retention, gross revenue retention, and nrr by cohort from your stripe + product usage data, every fifteen minutes. it splits contracted revenue from usage-annualized revenue automatically. on the first of the month you see all three numbers, the delta from last quarter, and the cohort that moved them.
if you're a finance lead at an ai-first company tracking this across product lines or contract types, zift handles that too.
the founders whose ai companies trade at 12x aren't smarter. they just stopped letting their arr number tell a story their retention can't back up.