How AI Companion Apps Should Measure Acquisition, Trials, Subscriptions and Retention

The AI companion app market was worth around USD 10.8 billion in 2024 and is projected to reach roughly USD 290.8 billion by 2034, a compound annual growth rate near 39 per cent over the 2025 to 2034 forecast period, according to a market.us industry report (accessed 30 July 2026). Money and competition are arriving in the category faster than measurement discipline is, and that gap is expensive. Teams are buying installs for a product that monetises through trials and subscriptions, then wondering why cheap installs never turn into revenue.

AI companion apps do not behave like the games or utilities most attribution advice is written for. The funnel runs from acquisition to a very specific activation moment, then into a trial, a subscription, and a long retention tail that defines whether the economics work. Add ad-network policy constraints that other categories never hit, and generic install-first reporting becomes actively misleading. This guide lays out the complete measurement model for the category, event by event.

Why AI Companion Measurement Differs From Other Subscription Apps

An AI companion app should be measured as a subscription business with a conversation-based activation step, not as an install-driven app, because the install is the least predictive event in the entire funnel. What predicts revenue is whether the user has a first meaningful conversation, starts a trial, and stays.

Three things make this category distinct:

Activation is behavioural, not a screen view. The moment that matters is the first real conversation, not the first app open. A user who installs and never talks to the companion is worth nothing, and install-count reporting cannot tell the difference.
Revenue is recurring and delayed. Money arrives through trials converting to subscriptions and subscriptions renewing, often days or weeks after install.
Channel access is constrained. Ad platforms apply policy scrutiny to companion and relationship apps that other verticals never encounter, which shapes both creative and measurement. Because the paying event lags the install so heavily, optimising campaigns on installs buys volume that looks cheap and converts poorly. The category rewards teams that measure and optimise on downstream, revenue-bearing events instead.

The AI Companion Funnel and Its Event Model

The funnel has five stages, and each needs its own instrumented event. Treating the whole thing as "install then purchase" throws away the signals that actually let you optimise.

The event model for the category:

Install and first open: the baseline, tracked but never optimised against.
Activation, defined as first conversation: the user sends and receives their first real exchange with the companion. This is the true top-of-funnel quality signal.
Trial started: the user begins a free trial of the paid tier.
Trial-to-paid conversion: the trial converts to a paid subscription, with plan and value attached.
Subscription renewed and cancelled: the events that define retained revenue and churn. Attach revenue values to the subscription events and send them into your measurement stack as revenue events. The Linkrunner revenue tracking API is a representative way to capture subscription revenue rather than treating a subscription as a valueless conversion. Once activation is instrumented as first conversation, you can finally answer the only acquisition question that matters here: which channels bring users who actually talk to the product.

Measuring Trial Starts and Trial-to-Paid by Campaign and Creative

Trials are the hinge of the whole model. A channel can look excellent on cost per trial start and terrible on trial-to-paid, and only measuring both together tells you where to put budget.

What to measure at the trial stage:

Trial start rate by campaign and creative, so you know which acquisition sources produce users willing to try the paid tier.
Trial-to-paid conversion rate by the same cuts, because the gap between cheap trials and converting trials is where budget gets wasted.
Time from install to trial start, which shapes how long your attribution window needs to be to credit the right campaign. Across subscription-app audits, the consistent pattern is that campaigns optimised purely on trial starts pull in a wave of low-intent trials that never convert, while the same budget optimised on trial-to-paid finds fewer but far more valuable users. Send the trial-to-paid event back to the ad networks as the optimisation target where volume allows. The full mechanics of tracking trials, conversions, and churn are covered in our guide to attribution for subscription apps.

Session Depth and Retention as User-Quality Signals

In a companion app, engagement is not a vanity metric; it is the leading indicator of subscription survival. A user who has long, frequent conversations is a user who renews. One who opened once and left is churn waiting to be recorded.

Quality signals worth tracking as cohorted metrics:

Session depth and conversation length, as an early proxy for who will convert and renew.
Return frequency in the first week, which predicts trial-to-paid better than install source alone.
Retention rate** by acquisition cohort**, so you can see which channels bring users who stay engaged versus users who churn before the trial even ends. Read these by cohort analysis rather than as blended averages, because a blended retention number hides the exact channel differences you are trying to act on. The practical approach to reading retention beyond the standard D0, D7, and D30 snapshots is laid out in our performance marketer's guide to cohort analysis.

Instrumentation Pitfalls Specific to Companion Apps

A few measurement mistakes recur in this category specifically, because the product shape is unusual. Catch them at instrumentation time, not after a quarter of misreported revenue.

Treating first open as activation. The first open is worthless as a quality signal here. If your activation event is anything other than a genuine first exchange with the companion, every downstream cohort comparison is built on noise.
Logging subscriptions without plan or value. A subscription event with no plan tier and no revenue value cannot power LTV or payback analysis. Attach both at the point the event fires, not in a later cleanup pass.
Ignoring free-to-paid tier changes. Users often move between plan tiers. If upgrades and downgrades are not instrumented, your renewal and LTV figures drift away from billing reality.
Mixing chargeable and free conversations. If the product meters conversations, track chargeable interactions distinctly, because they are the ones that predict monetisation.
Not tagging trials by their entry point. A trial started from a hard paywall behaves differently from one started from a soft prompt. Tag the trigger so trial-to-paid can be read by entry point, not just by channel. The common thread is that a companion app has more meaningful in-app states than a typical utility, and each state that goes uninstrumented is a decision you will make blind later. Design the event model around the states that predict revenue, and validate that each one fires cleanly before spend scales.

Channel Notes: Policy, Creative and Targeting Constraints

Companion and relationship apps face ad-platform constraints that most verticals never think about, and these constraints shape measurement as much as creative.

What to plan around on the major channels:

Meta and TikTok policy scrutiny. Companion apps, particularly those with relationship or emotional framing, face heightened review. Creative that works elsewhere can be rejected here, so measure creative approval and delivery, not just performance.
Creative constraints drive measurement needs. Because winning creative is narrower, creative-level ROAS becomes essential; you cannot afford to scale a concept that quietly underperforms on trial-to-paid.
Targeting limits. Some audience and interest categories are restricted, which pushes teams toward broad targeting and makes downstream event optimisation more important, not less. The measurement implication is consistent: with fewer creative and targeting levers, the value-based event signal you feed the platform does more of the optimisation work, so getting trial and subscription postbacks right is not optional.

Cohort ROAS and Payback Windows for Subscription Pricing

Subscription economics live or die on payback. With recurring revenue, a cohort that looks unprofitable at D7 can be strongly profitable by D90, and only cohort ROAS over a proper window reveals it.

How to measure return for a subscription companion app:

Cohort ROAS over the payback window, not day-one ROAS, because most revenue accrues after the trial converts and across renewals.
Lifetime value (LTV)** by acquisition cohort**, so you can compare channels on the revenue they eventually produce rather than the installs they cheaply deliver.
Payback window by channel, which tells you how much you can afford to pay for a subscriber and how long you must fund the gap. The metrics that connect early events to eventual churn and LTV are covered in our rundown of the MMP metrics that predict churn and LTV. The headline discipline: never judge a subscription campaign on install cost or day-one return. Judge it on cohort ROAS measured across the full payback window.

Sanity Checks to Run in Your MMP Each Week

A weekly hygiene routine keeps the model honest as spend scales. None of these takes long; skipping them is how a quarter of misattributed revenue accumulates unnoticed.

The weekly checks:

Do subscription revenue events reconcile with your billing or RevenueCat data within tolerance? A widening gap means a broken revenue event.
Is trial-to-paid conversion stable by channel, or has a source quietly started delivering low-intent trials?
Are postbacks firing the trial and subscription events to each network, to the correct account?
Do cohort retention curves hold their shape, or has a recent campaign pulled retention down? If you are still choosing the platform to run these checks in, our comparison of the best MMP for AI companion apps in India covers the category-specific requirements. The rule is to run the checks before you scale spend, not after finance flags a discrepancy.

How to Validate the Model in Your MMP

The funnel model only pays off if your dashboard captures it faithfully. Run these validation steps before scaling spend, because subscription economics amplify measurement errors over the payback window rather than revealing them on day one.

Confirm activation fires as first conversation. Send a first message on a test account and verify the activation event fires once, at that moment, not on app open. If activation is a screen view, your top-of-funnel quality signal is meaningless.
Verify subscription revenue reconciles with billing. Start a test trial, convert it, and confirm the revenue event matches your billing or RevenueCat figure. A gap here compounds across renewals into a large reporting error.
Check trial-to-paid is readable by campaign and creative. Confirm you can segment trial-to-paid conversion by both cuts, because that segmentation is the entire budget-allocation decision for the category.
Confirm value-based postbacks send trial and subscription events. Verify Meta, Google, and TikTok receive the downstream events, on the correct account, so bidding trains toward payers.
Build the cohort ROAS view over the payback window. Confirm the dashboard can show ROAS maturing over 30, 60, and 90 days by acquisition cohort, not just day-one return. Tech Explainer: why day-one ROAS misleads for subscriptions. A subscription cohort earns almost nothing on install day. Revenue arrives when the trial converts, then again at each renewal, so a cohort that looks deeply unprofitable at D1 can clear payback by D60 and turn strongly positive by D90. If you optimise or cut campaigns on day-one ROAS, you will kill your best subscriber sources before they have had a chance to pay. The correct unit of judgement is cohort ROAS measured across the full payback window, with early engagement and trial-to-paid used as the leading indicators while the revenue matures.

Do not scale spend until these five checks pass. In a subscription model, a broken revenue event does not announce itself; it silently understates your winners for the entire payback window.

Frequently Asked Questions

What is a good trial-to-paid rate for an AI companion app?

There is no single benchmark that fits every pricing model, but the number that matters is the gap between your channels, not the absolute figure. Measure trial-to-paid by campaign and creative, then shift budget toward the sources with the highest conversion rather than the lowest cost per trial. A channel with fewer, higher-converting trials usually beats a cheaper one.

Which events should fire to ad networks?

Send downstream, revenue-bearing events, primarily trial-started and trial-to-paid conversion, as value-based postbacks to Meta, Google, and TikTok. Optimising on installs buys cheap, low-intent users in this category; optimising on trials and subscriptions trains the platforms to find users who actually pay.

How long should the attribution window be?

Long enough to capture the lag from install to trial start and from trial to paid conversion, which is often days rather than hours. A short window will credit the wrong campaigns or drop the conversion entirely, so set the window to match how long your users actually take to convert.

Building the Measurement Model Before You Scale

The AI companion category is growing fast enough that acquisition budgets are climbing before measurement catches up, and that order is exactly backwards. Instrument activation as the first conversation, track trials and subscriptions as valued events, read retention and ROAS by cohort over the real payback window, and feed trial and subscription signals back to the ad networks so they optimise for paying users instead of cheap installs.

If you want cohort views showing trial-to-paid by campaign and postbacks trained on subscription events rather than installs, that is the workflow platforms like Linkrunner are built to run, and you can request a demo from Linkrunner to see it against your own funnel. Start by mapping your five funnel events, defining activation as the first real conversation, and pointing your postbacks at the events that pay.