ChartMogul finds standard SaaS LTV forecasts miss actual cohort revenue by more than 50% nearly three times in ten

ChartMogul’s newly published 2026 SaaS LTV report tests a foundational SaaS metric against what customers actually paid. The company analyzed 35,512 quarterly customer cohorts from 3,331 ChartMogul accounts spanning Q1 2020 through Q4 2024. At each cohort’s formation it calculated LTV from the account’s blended ARPA and trailing six-month logo churn rate, then compared the implied revenue with actual MRR generated over 12, 24 and 36 months. The median 12-month error is only -2.4%, but 28.3% of cohorts miss actual revenue by more than 50% in either direction. ChartMogul finds those misses are structured rather than random: blended ARPA mismatch explains 84.9% of the variation in prediction accuracy, and expansion causes the median bias to flip from overprediction at 12 months to underprediction at longer horizons.

LTV influences how much SaaS companies are willing to spend on customer acquisition, what payback period they tolerate, how they value segments and how they discuss future economics with investors. A formula that is directionally useful but frequently wrong by more than 50% can create expensive false precision. The report does not mean builders should abandon LTV. It shows why cohort revenue curves, payback period and net-MRR movements need to sit beside it, especially for smaller SaaS companies where signup ARPA, churn and expansion can differ sharply from company-wide averages.

The typical 12-month cohort generated 2.4% less revenue than the standard LTV calculation predicted, which makes the formula look accurate in aggregate. But 28.3% of cohorts were wrong by more than 50% in absolute terms. Large positive and negative errors offset one another, leaving a median that looks much more precise than the underlying distribution.

The standard formula uses company-wide ARPA, which includes long-tenured customers that may have added seats, upgraded or expanded spend. A newly acquired cohort has not had time to accumulate that expansion. ChartMogul reports that ARPA mismatch explains 84.9% of the variation in cohort prediction accuracy, making it the strongest measured reason one cohort’s LTV estimate is more reliable than another’s.

The model also applies the account’s trailing six-month logo churn rate to the new cohort. In reality, enterprise buyers, SMBs, trial converters and customers acquired through different channels can have different survival curves. The median ARPA and churn errors partly cancel at 12 months, which is another reason the headline median looks deceptively good.

At 12 months the median prediction error is -2.4%, meaning LTV slightly overpredicts revenue. At 24 months the median becomes +6.9%, and at 36 months +14.4%, meaning the original LTV increasingly understates realized revenue. ChartMogul attributes the reversal to expansion among surviving customers, which gradually overwhelms the initial ARPA mismatch.

Within ChartMogul’s SMB/B2C-heavy dataset, the cheapest signup-MRR quartile retained 74% of customers after 12 months versus 45% for the most expensive quartile. Across companies, 37.2% showed a negative relationship where bigger customers churned sooner, 48.9% showed no meaningful relationship and only 13.9% showed the expected pattern of bigger customers staying longer. ChartMogul cautions that its dataset mix matters and that this pattern is not universal.

ChartMogul recommends complementing LTV with metrics grounded in actual cohort behavior. Payback period can be more useful for bounded near-term acquisition economics; cohort revenue curves show what customers actually generated; and net MRR movements expose expansion, contraction, churn and reactivation that one LTV number compresses together.