Increase Subscription LTV

Lifetime value is the only headline SaaS metric that is a forecast instead of a measurement. Every other number on a dashboard reports something that has happened. LTV reports something that hasn't happened yet, and it does so from two company-wide averages applied to a group of customers who arrived last quarter.

It has now been tested against what actually happened. Across 35,512 cohorts from 3,331 companies, the typical cohort generated 2.4% less revenue over twelve months than the formula predicted (ChartMogul SaaS LTV Report, 2026). That sounds like a metric working well. It isn't: 28.3% of cohorts were wrong by more than half, and the median looks reassuring only because errors in opposite directions cancel each other out (ChartMogul SaaS LTV Report, 2026).

The more useful finding is that the error isn't random. It has a direction, and the direction reverses. At twelve months the formula overestimates. By twenty-four it underestimates by 6.9%, and by thirty-six by 14.4% (ChartMogul SaaS LTV Report, 2026). The same metric is too optimistic about a young cohort and too pessimistic about an old one, for reasons that are mechanical and predictable.

Adapty's panel shows the same shape at the level of a single pricing decision. The plan worth $7.40 per subscriber on day zero is worth $54.50 by day 380, while the plan worth $42.08 on day zero finishes at $49.92 (Adapty State of In-App Subscriptions, 2026). Judged on the day of purchase, the winner and the loser are the wrong way round.

So the number is usable, though not on its own and not at a single horizon. Work out which direction yours errs in before you spend against it.

The standard formula is ARPA divided by customer churn rate. Both inputs are company-wide averages, and both are applied to a cohort that isn't average.

Failure mode one: new customers are not average customers

ARPA is blended across every subscriber you currently have, including customers who joined three years ago and have expanded their spend several times since. Using it to predict a new cohort assumes the people who signed up last month already look like that.

At the median this mismatch costs 4.9% of accuracy (ChartMogul SaaS LTV Report, 2026), which is small, and yet it explains 84.9% of the variation in how accurate the prediction is from one cohort to the next (ChartMogul SaaS LTV Report, 2026). The typical effect is modest, then, while the differences between cohorts are almost entirely this. If your LTV is unusually wrong, that's the first place to look.

Failure mode two: cohorts don't churn at the average rate

The formula projects survival using a trailing six-month company-wide churn rate. Enterprise buyers, self-serve converters and trial-to-paid users rarely share a survival curve, even when they arrived in the same quarter. The company average says little about the specific group in front of you.

Why the two together produce a small median

At twelve months these errors partially offset: the formula overstates what new customers are worth, and the higher-paying customers it overstates most also leave soonest. That is the whole explanation for the reassuring -2.4% figure (ChartMogul SaaS LTV Report, 2026), and it is why the median is the least informative number in the study.

Knowing the error has a direction is worth more than knowing its size, because the direction changes what you should do with the number.

The reversal, and what causes it

Early on, blended ARPA makes the projection too optimistic. As a cohort matures its survivors start expanding, and expansion eventually overtakes the initial bias. By twenty-four months the median error is +6.9% and by thirty-six it is +14.4% (ChartMogul SaaS LTV Report, 2026). The mechanism is visible directly: 45.6% of the revenue still being paid by survivors of the 2020 to 2021 cohorts comes from expansion after signup, not from what they originally bought (ChartMogul SaaS LTV Report, 2026).

The spread widens as the bias grows

The metric becomes less reliable with age as well as more biased: the spread of outcomes nearly doubles by thirty-six months. A three-year LTV is therefore both systematically low and individually less trustworthy than a one-year one, which is the opposite of how longer horizons usually behave.

What to take from it

Treat a twelve-month LTV as a ceiling and a thirty-six-month LTV as a floor. If you are using LTV to decide acquisition spend on a payback horizon under a year, the formula is flattering you. If you are using it to value a mature book, it is underselling you. The customer LTV calculator will give you the steady-state figure the formula produces; the point of this section is that you shouldn't stop there.

The cohort study measures companies. Adapty's panel measures pricing decisions, across roughly 16,000 apps and 500 million transaction events, and finds the same trap at a different scale.

The lowest day-zero value wins the year

Weekly plans convert at $7.40 of cumulative value per subscriber on day zero and reach $54.50 by day 380, growth of 636% (Adapty State of In-App Subscriptions, 2026). Annual plans start at $42.08, nearly six times higher, and finish at $49.92 (Adapty State of In-App Subscriptions, 2026). Any decision made on day-zero value picks the wrong plan, and it isn't a close call.

The mechanism is renewal frequency, and there's nothing mysterious about it: a weekly plan has more than fifty chances to renew in a year and an annual plan has one. Weekly subscribers on trials renew at 59.2% after the first billing cycle against 37.0% for those who bought directly, and that gap widens by the fifth renewal (Adapty State of In-App Subscriptions, 2026).

Trials raise lifetime value, except where they lower it

The same panel puts the trial premium at +85.1% in Utilities, +63.6% in Health and Fitness and +50.4% in Education (Adapty State of In-App Subscriptions, 2026). In Productivity it is -13.7% and in Lifestyle -21.2% (Adapty State of In-App Subscriptions, 2026). Two of five categories measured show trial users worth materially less than direct buyers.

The general claim, that trials increase lifetime value, is usually made without a category attached. It is well supported in some categories and reversed in others, and measuring your own categories is usually the only way to know which you are in.

The single most consequential assumption in most LTV work is that higher-paying customers are worth more because they also stay longer. In the cohort data, that's a minority case.

Retention falls as relative price rises

Splitting customers into quartiles by what they paid at signup, twelve-month retention runs 74% for the cheapest quartile and 45% for the most expensive (ChartMogul SaaS LTV Report, 2026). The median signup MRR of still-active customers falls over time, which is the same finding from the other direction: within a cohort, the higher payers leave first.

The enterprise pattern holds for one company in seven

Measured per company, the correlation between signup MRR and twelve-month survival is negative at 37.2% of companies, flat at 48.9%, and positive at only 13.9% (ChartMogul SaaS LTV Report, 2026). The pattern most people assume, bigger customers staying longer, is the least common of the three.

It isn't evenly distributed, and the distribution is the useful part. The positive group is 93% B2B and skews to larger companies (ChartMogul SaaS LTV Report, 2026), which is the classic enterprise dynamic of integration depth and switching cost. The negative group is 77% under $500k ARR and contains most of the B2C businesses in the sample (ChartMogul SaaS LTV Report, 2026). A cheap subscription sits below the threshold at which anyone bothers to make a cancellation decision; an expensive one gets reviewed.

Why this matters for the formula

LTV is most accurate where its two errors offset, which is the negative-correlation case, and least accurate where they compound. Companies where bigger customers stay longer see a median error of +1.3% but 15% of cohorts unreliable; companies where bigger customers leave sooner see -5.7% and 45% unreliable (ChartMogul SaaS LTV Report, 2026). Knowing which of the three you are is a prerequisite for reading your own LTV at all.

Because expansion drives so much of the mature-cohort value, net revenue retention is the companion metric worth tracking beside LTV; the benchmarks for it are on the NRR and GRR benchmarks page.

Lifetime value is normally treated as ending at cancellation. For a meaningful share of customers it doesn't, and the returning customer is the cheapest revenue in the business: no acquisition cost, and a product they have already used.

Returns are front-loaded

45% of returns happen within 30 days of churn and 66% within 90, with a median gap of 38 days (ChartMogul SaaS Winbacks Report, 2026). Fewer than one in ten come back after a year. A win-back program that starts at ninety days has missed most of what it was built to catch.

They do not come back on a discount

This is the finding that contradicts how win-back is usually run. Only 25% of returning customers come back on a lower-priced plan (ChartMogul SaaS Winbacks Report, 2026). 42% return on the same plan and 33% return on a higher one, and the median change in ARR between the plan they left and the plan they returned on is exactly $0 (ChartMogul SaaS Winbacks Report, 2026).

The rate itself scales with company size, running 7% under $500k ARR and 13% above $30M, with 7% to 13% given as the typical range (ChartMogul SaaS Winbacks Report, 2026). Delay costs value as well as volume. Customers returning within 8 to 30 days come back on a lower plan 23% of the time, rising to 29% for those who return after more than a year (ChartMogul SaaS Winbacks Report, 2026).

Where this belongs in the metric

A returning customer is a second lifetime on an acquisition already paid for, which is why win-back sits in a guide about lifetime value instead of one about reducing churn. Churn reduction is covered separately in how to reduce subscriber churn. If your billing system counts a returning customer as a new acquisition, which most do by default, your LTV and your CAC are both wrong in the same direction.

One number appears in more retention writing than any other: that a 5% increase in retention increases profit by 25% to 95%, a claim almost always credited to the Bain paper (Bain Prescription for Cutting Costs, 2001). We hold that document, and it doesn't say so.

What it says, in a single sentence, is that in financial services a 5% increase in customer retention produces more than a 25% increase in profit (Bain Prescription for Cutting Costs, 2001). One sector, and a floor, not a range. The figure 95 doesn't appear anywhere in the document. The range enters the literature through a 2014 Harvard Business Review article, and the underlying work is Reichheld and Sasser's 1990 paper on defections rather than the Bain piece it is usually attached to.

The correction isn't that retention doesn't pay. It is that the most-quoted evidence for it is narrower than its reputation: a business outside financial services has no published basis for expecting that specific result, and anyone quoting the range is repeating a restatement, not the study.

Five things follow from the evidence above, in the order they are worth doing.

Compute LTV at more than one horizon

A single number hides the reversal. Twelve, twenty-four and thirty-six month figures for the same cohort tell you which direction your own error runs, and that is more actionable than any benchmark. That's cohort work in your own data, not something a calculator can do for you: the standard formula, and our own calculator, return one steady-state number by construction. If you can only afford one horizon, the twelve-month figure is the conservative one for acquisition decisions.

Use cohort ARPA, not blended ARPA

This is the single highest-value change available, since the blended-ARPA mismatch explains most of the variation in accuracy between cohorts. Use the ARPA of the cohort you're predicting, not of the base you already have.

Check which correlation group you are in

Measure whether your higher-paying customers actually stay longer. One company in seven finds they do. If yours is one of the 37.2% where they leave sooner (ChartMogul SaaS LTV Report, 2026), your LTV is more accurate than average but your expansion assumptions are not.

Count returning customers as returning customers

A win-back is a second lifetime, not a new acquisition. Track it separately or both your LTV and your CAC will mislead you in the same direction, and re-engage inside the first month, not the first quarter.

Treat LTV:CAC as a sanity check, not a target

The traditional cross-industry benchmark is 3:1 (First Page Sage LTV:CAC Benchmark, 2025), but the source we hold for the segment-level figures is one agency's client roster, disclosed by the source itself as skewing organic-channel and B2B. Use the ratio to notice when something's moved a long way, not to defend a decision to a board.

What good looks like

An LTV you can state at three horizons, computed from cohort ARPA and cohort churn, with returning customers counted separately and a stated direction of error. With more than a quarter of cohorts wrong by more than half, that position will hold up better than any single number, however carefully benchmarked.