How to Forecast LTV and Payback for Apps and Games

When I started buying users for mobile games, I made the mistake every beginner makes. I saw a cohort spend less in its first week than I paid to acquire it, and I wanted to shut the campaign off. That instinct is wrong, and understanding why is the whole reason forecasting exists.

Almost every install you buy loses money on day one. The revenue that makes UA profitable arrives slowly over months, and a lot of it arrives with only a probability of showing up at all. So you cannot wait for the truth. You predict it early, act on the prediction, and correct as real data lands. This piece is about how that prediction actually works.

Why you predict LTV instead of waiting

LTV is the total revenue a player brings across their whole lifecycle, from their first launch to their last. It is the single most important number for deciding whether your advertising is profitable. The problem is that a lifecycle can run for a year or more, so the true number only exists once the cohort is basically dead. By then the decision to scale or kill that campaign is long gone.There is a second complication. Mobile monetization is brutally skewed. In free-to-play games it is common to see roughly 0.2% of users drive around 86% of revenue. A handful of paying players, the whales, carry the entire cohort. Whether your forecast is right often comes down to whether those few high spenders show up, which is exactly the part that arrives with probability rather than certainty.So forecasting is the only way to make a budget decision on time. You accept the early loss, predict where the cohort is heading, and commit money against that prediction.

The early signals that drive a forecast

A forecast is only as good as the signals feeding it. The ones I lean on most are retention and revenue at fixed checkpoints, because they show up fast and they correlate strongly with where a cohort ends up.

• Day 1, 7 and 30 retention: the share of installs still opening the app on those days. Retention is the backbone of the curve, because a player who churns cannot spend.
• Early ARPU: average revenue per user in the first few days, which tells you how quickly monetization ramps.
• Payer conversion and early payer behavior: what fraction start paying at all, and how big those first purchases are. This is your read on whether whales are hiding in the cohort.
• Retention decay shape: how steeply the curve falls, since Day 90 retention in many games averages only around 20%.

These work because player behavior early in the lifecycle rhymes with behavior later. A cohort with strong Day 7 retention and healthy early ARPU almost always outperforms a flat-looking one, long before either reaches its true LTV.

Cohort-based forecasting and the ROAS curve

The cleanest way to think about a forecast is per cohort, meaning all the users you acquired in a given window, usually a day or a week. You track each cohort's cumulative revenue as it ages, and you fit that growth to a curve so you can extend it into the future.The simplest starting estimate is LTV = ARPU x Lifetime, where lifetime is how long an average player stays active. That is a fine mental model, but real cohorts do not spend at a flat rate. Revenue piles up quickly at first and then flattens as retention decays, which is why practitioners fit the cumulative curve and read a projected LTV off it.The same data drives your ROAS curve, cumulative revenue divided by acquisition spend, plotted over cohort age. It starts far below 100% and climbs as revenue accrues. Your payback window is the cohort age where the ROAS curve crosses 100%, the point where a cohort has finally repaid what you spent to buy it. Teams like Panoramik built their planning around exactly this kind of LTV forecasting, projecting the curve early so they can commit UA budget before cohorts fully mature.

A simple worked example

Let me make this concrete with round numbers. Say you buy a cohort of 1,000 installs at a CPI of $2, so your acquisition spend is $2,000. In the first 7 days that cohort generates $600 of revenue. Waiting-mode thinking says you lost $1,400 and should stop. Forecasting mode asks a better question: where is this heading?

Suppose your historical curves show that, for this game and geo, Day 7 revenue is typically about 30% of Day 180 revenue. Your projected 180-day revenue is $600 / 0.30 = $2,000. That gives a forecast LTV of $2,000 / 1,000 = $2.00 per user.

Now compare that to your CPI of $2.00. Projected LTV equals CAC, so this cohort is forecast to roughly break even by day 180. ROAS at Day 7 is $600 / $2,000 = 30%, and it is projected to reach 100% around the six-month mark. That is your payback window.

If your target is a 120% ROAS inside 180 days, this campaign is slightly under the bar and you would trim the bid. If the same math produced a forecast LTV of $2.60, that is a 130% return and a clear signal to push more budget in. The point is that one week of data, run through a curve, let you decide.

How forecasts turn into budget decisions

A forecast is only useful if it changes what you spend. The first thing it sets is your maximum bid. If projected LTV is $2.60 and you want a 120% payback, your ceiling CPI is about $2.60 / 1.2, roughly $2.17. Above that, the cohort will not clear your target, so you cap the bid there and let the channel find volume underneath it.The second thing forecasting forces you to plan for is the funding gap. Because every cohort runs at a loss before it matures, scaling UA means carrying months of negative cash before the payback curve catches up. You have to know that gap and fund it deliberately, otherwise a healthy, profitable-on-paper campaign can still run you out of money mid-flight.Then it becomes a loop. You forecast a cohort, commit budget against the prediction, and watch the real ROAS curve land day by day. When actuals beat the forecast, you scale and raise bids. When they undershoot, you cut early instead of waiting six months to confirm it. That feedback loop, not any single number, is what makes UA a controllable machine.