Activity View vs LTV View in Mobile Measurement Partners (MMP): Attribution Cohorts and In-App Revenue Tracking

The Same Revenue, Filed Under Two Different Dates

The confusion these two reporting modes generate inside marketing teams is almost entirely resolved by a single realisation, and it is worth stating before anything else: they contain the same underlying transactions, filed against different dates. Nothing is being counted twice, nothing is missing, and neither view is more accurate than the other.

Consider one user. They install on 1 March through a paid campaign, spend nothing for a fortnight, then make a purchase on 15 March. In activity view that revenue appears on 15 March, because that is when the money moved. In LTV view the same revenue appears against 1 March, because that is when the user was acquired.

Now scale that to a real account. Any given day's activity-view revenue is the sum of purchases made that day by users acquired across every previous month and campaign. Any given day's LTV-view revenue is everything that cohort has ever generated, and it will keep growing for as long as those users remain active. The two reports describe the same business and will never produce matching daily totals, which is exactly as intended.

Once that is clear, the question stops being which view is right and becomes which question you are asking. If the question involves money that arrived, you want activity view. If the question involves whether an acquisition decision was a good one, you want LTV view. Almost every dispute about MMP numbers is two people answering different questions with equal confidence.

Why Marketing Decisions Require the Cohort View

The argument for using LTV view in marketing is not a matter of preference or convention — it is that the alternative produces arithmetic which cannot mean anything, however confidently it is presented.

Media spend is incurred on a specific day to acquire a specific set of users. To judge whether that spend was worthwhile you need the value those particular users produced. LTV view provides exactly that: spend on 1 March against revenue from users acquired on 1 March, however long that revenue took to arrive.

Dividing activity-view revenue by same-day spend produces a ratio with no defensible interpretation, because the numerator and denominator describe different populations. Today's revenue came largely from users acquired weeks or months ago; today's spend acquired users who have mostly not spent anything yet. The resulting number moves for reasons entirely unrelated to campaign quality — a payday, a seasonal peak, a promotion aimed at existing users — and teams that optimise against it end up chasing noise.

The characteristic failure this causes is worth naming because it is common and expensive. A team scales spend on a day when activity-view revenue happens to be high, attributing the revenue to the campaigns running that day. The revenue was actually generated by an older cohort. The scaled spend acquires users who convert normally, the ratio collapses the following week, and the campaign is judged to have deteriorated when nothing about it changed. The measurement created the illusion of both the success and the decline.

There is also a subtler benefit to the cohort framing: it forces an honest conversation about payback period. LTV view makes it immediately visible that a cohort has returned only a fraction of its acquisition cost after seven days, which is uncomfortable and true. Activity view lets a team avoid that conversation, because the daily revenue figure includes mature cohorts and looks healthier than the marginal economics actually are.

What Both Views Inherit From Attribution

Neither view is more accurate than the other, but it is worth being explicit that both rest on exactly the same foundation: the MMP's decision about which source deserves credit for each install. Everything downstream inherits whatever imprecision exists in that decision.

That foundation has become considerably less solid on mobile. Privacy frameworks introduced by the platforms restrict device-level identifiers, which means a meaningful share of installs arrive without the deterministic signal that attribution historically relied on. MMPs fill the gap with a combination of aggregated reporting from the platforms, probabilistic modelling, and self-attributing network data, each of which behaves differently from the others.

Three consequences are worth carrying into any LTV analysis. Install counts will not match across systems — your MMP, the ad network and the app store dashboard each count differently, and the differences are structural rather than errors. Some revenue is unattributable and lands in an organic or unattributed bucket that is not genuinely organic; it is measurement loss, and treating it as free acquisition overstates organic performance considerably. Self-attributing networks report their own contribution, which means comparing them against each other requires knowing that each is grading its own homework under its own rules.

None of this makes cohort analysis useless. It does mean the honest framing is that LTV by channel is a well-founded comparison rather than a precise measurement, and that a channel appearing to outperform by a small margin may simply be attributing more aggressively. Where a decision is large enough to matter, an incrementality test — holding a channel out in matched geographies — is the only method that settles it, and it is worth the cost precisely because attribution cannot.

Cohort Maturity and the Recency Trap

The single most common misreading of an LTV report, and the one most likely to trigger a bad decision, is treating cohorts of different ages as directly comparable. Because value accumulates over time, a cohort acquired last week will always look worse than one acquired last quarter, entirely regardless of quality.

This produces a predictable and damaging pattern. A weekly report shows LTV declining steadily as the dates approach the present. Someone concludes that acquisition quality is deteriorating. In reality the older cohorts have simply had longer to spend, and the apparent decline is a property of the calendar rather than of the campaigns.

The fix is to compare at fixed cohort age rather than across a date range. Day-7 LTV for the cohort acquired four weeks ago against day-7 LTV for the cohort acquired last week is a fair comparison; total accumulated LTV for both is not. Every meaningful cross-cohort or cross-channel comparison should specify the age at which it is measured, and reports that omit it invite the wrong conclusion.

Choosing which age to standardise on is a business question rather than a technical one. It should be the point at which a cohort has reached a reasonably stable proportion of its eventual value — early enough to make decisions quickly, late enough that the signal is real. Establishing that from your own historical data is straightforward: take mature cohorts, plot cumulative value by day since install, and find where the curve flattens sufficiently that the remaining growth would not change a decision. That day becomes your standard reporting horizon.

A related discipline is to hold reports open rather than treating them as final. Because LTV rows continue growing, an export taken today and an export taken next month will disagree about the same historical week, and both will be correct at the moment they were run. Date-stamp exports, and never compare a freshly pulled report against a saved one without accounting for the maturation that happened in between.

Payback Period and How It Constrains Growth

Once cohort revenue is being measured properly, a question emerges that daily activity reporting hides completely, and it is arguably the most important number in a growth business: how long does it take for a cohort to return what it cost to acquire? That number governs how fast a business can grow, and it is only visible in LTV view.

The mechanism is cash rather than profitability. Acquisition cost is paid immediately; the revenue arrives over weeks or months. A business whose cohorts break even at day ninety is funding three months of spend before seeing it back, and the faster it grows the larger that funding gap becomes. Growth consumes cash in direct proportion to payback period, which is why two companies with identical unit economics can have entirely different growth ceilings.

Reading payback from an LTV report means plotting cumulative revenue per cohort against cumulative acquisition cost for that cohort and finding the crossing point. Doing this by channel rather than in aggregate is usually where the useful finding sits, because channels differ substantially. A channel with a higher cost per install but a much shorter payback can be more valuable than a cheaper channel whose users take a year to return their cost, even when the eventual lifetime value is similar.

Two practical cautions apply. First, use contribution margin rather than gross revenue on the revenue side wherever you can — platform fees, payment processing and cost of goods all reduce what actually comes back, and a payback calculation on gross revenue is optimistic by exactly that margin. Second, be careful about extrapolating from immature cohorts. Fitting a curve to thirty days of data and projecting a year forward produces a number that feels rigorous and rests almost entirely on the shape you assumed.

The organisational value of getting this right is that it converts a recurring argument into a constraint everyone can see. Marketing wanting to scale and finance resisting is usually a disagreement about payback conducted without the number in front of either party. Once the cohort curves are on the table, the conversation becomes about whether the business can fund a known gap, which is a decision rather than a dispute.

Where Activity View Is the Right Tool

Activity view is frequently dismissed as the naive option once a team understands cohort measurement, and that dismissal is a mistake. Several important questions can only be answered on the date the activity occurred.

Financial reconciliation is the clearest and least negotiable case. Revenue recognition happens on the transaction date, so activity view is the only MMP report that can be compared against what finance sees. Attempting to reconcile LTV view against a bank statement is a category error that will consume a great deal of time.

Operational monitoring is the second. If purchases stop working after a release, activity view shows revenue collapsing today. LTV view spreads that same collapse thinly across every historical install date, which makes it far harder to spot and considerably harder to date. For incident detection, the date something happened is the useful date.

Measuring interventions aimed at existing users is the third and least obvious. A re-engagement campaign, a promotion, a seasonal event or a content release is intended to move revenue among people you already have. Its effect appears in activity view as a lift on the days it ran. In LTV view that same lift is scattered backwards across the install dates of everyone who responded, where it is effectively invisible. Anything targeting your existing base should be judged on activity view.

The practical arrangement most mature teams settle on is to run both continuously with explicit ownership: LTV view for acquisition decisions and channel comparison, activity view for finance, operations and lifecycle marketing. Writing down which report answers which question removes most of the recurring confusion, because the underlying problem was never the data — it was two teams reading different reports and assuming they described the same thing.