Why retention matters in ad monetization

"An ad placement here? That's a really bad experience for users."

You look at this year's revenue target and realize your current ad placements can't get you there. Then you're poking around the app and it hits you: an ad placement right here at the bottom would be perfect. You've got a decent user base by now. Would they really hate one more placement that much? You open a spreadsheet, run a rough revenue simulation, and a placement that doesn't even exist yet already feels like your own child. So you schedule a meeting with the product team, make your case for the projected revenue, and what comes back is: "That's a really bad experience for users. Can you rethink it?"

This isn't about taking sides. But if you work at a company of any real size, you've been in this situation at least once. We've written before about balancing UX and revenue (The dilemma of app ad monetization: balancing UX and ad revenue), and about a case where Alarmy made the strategic call to give up revenue, for a time, to protect the product experience (Choosing usability over a 2x revenue lift).

When we experiment with adding an ad to the user journey, we read user retention alongside revenue: does the new ad damage the user flow, and are users feeling enough friction that they stop coming back? Today we want to extend that idea: why retention and LTV matter in ad monetization, and how to measure them.

‍

Should you acquire these users or not?

LTV stands for Lifetime Value: the revenue a customer generates from the moment they enter your service until they churn. To calculate ad LTV, convert a user's lifetime into days and multiply it by daily ad revenue (ARPDAU).

Once you measure ad LTV and start monitoring ad revenue with it, you gain two things: 1) you can measure the relationship between ad revenue and product usability in a single metric, and 2) you can measure the value of user acquisition far more precisely.

When you run an ad experiment with revenue as the primary metric and retention as the guardrail, the analysis often traps you in the revenue–retention tradeoff. It would be great if ad revenue rose and retention held flat. But in reality, if ad revenue is up 10% and retention is down 3 percentage points, should you keep the treatment group? How much revenue gain does a 1-point retention difference offset?

LTV makes this dilemma easier to work through. The "lifetime" in LTV is the average customer lifespan. If you already measure daily retention, you can substitute the sum of daily average retention for that lifespan. In other words, instead of retention over one fixed window, you use total retention to derive lifetime. Compute that difference, then compare the change in lifetime against the change in ARPDAU, and you can gauge whether a user's actual LTV is rising.

Go a step further. Once you measure the ad revenue a user generates across their lifespan, you can also build an efficient user acquisition strategy. Divide LTV by CAC (customer acquisition cost), and you can judge quantitatively whether you're currently losing or making money on each user you bring in. Break that down by channel or country, and you can form concrete hypotheses about which countries to focus on, or where to take action to raise daily ad revenue.

For example, if ROAS is high in a country, you can act to drive more installs there. If ROAS is low but CAC is also low (so users are relatively cheap to acquire), you can run more aggressive ads to push LTV up.

‍

So how do you calculate ad LTV?

Ad LTV = lifetime (days) × daily ad revenue (ARPDAU)

The hard part of ad LTV is the lifetime. Even if you shorten your measurement window to one or two months, you still have to wait out that period before the retention numbers land, which slows everything down. The metric ends up feeling un-actionable, a lagging indicator, which leaves a nagging unease.

To stretch the covered period as far as possible while minimizing the wait, treat lifetime as SUM(D+N day retention) and use a predicted retention value. The simple route is scipy's curve_fit function; or build your own model with linear regression, tree-based methods, or deep learning. None of it is hard. What matters here is identifying the points where your service's retention curve converges, and finding a function or model that predicts those points as well as possible.

Say your service has retention that drops sharply through D7, flattens from D14, and holds steady after D28. Then it's important to check that your chosen model predicts the D7 and D14 data points well, and that it predicts retention staying flat past D28.

The window varies by model, but with a little refinement you can predict the sum out to D365 fairly accurately from just D7 or D14, so you can compute a much longer lifespan from a small number of observations.

One more suggestion: measure retention by event date (green) rather than by cohort (red). This is the more actionable approach mentioned earlier. If you wait 7 days on a cohort basis and measure LTV against D7 retention, you're looking at metrics from 7 days ago, and it can feel too late to act.

Change this slightly. Compute LTV each day using that day's D1–D7 retention, and you're working with the freshest retention data. When LTV shifts, you can analyze the change in each retention metric the same way and trace the cause. At first glance this might seem like splitting hairs: same thing, different framing. But from both the perspective of presenting a metric and receiving it, recognizing that you're looking at the latest data raises the odds that a hypothesis turns into an action.

‍

Ad revenue and product usability

As noted, ad revenue and product usability are like the chicken and the egg. You can't separate them even if you want to. With those two always in tension, LTV makes a good compass. It can look hard to compute consistently and awkward to monitor regularly. But once you put in the effort to start measuring it, it becomes an excellent metric, not just for weighing revenue against usability, but for turning user acquisition into a higher-quality decision, and for choosing where to focus.

‍