See what your young cohorts, campaigns and experiment variants will earn at 1, 3, 6, 9 and 12 months — predicted from your own renewal history, with anonymized platform data filling the gaps while your history is short.
A subscription business earns most of its money after the first payment. A cohort that started last week has paid for one week, so every report that looks at it — cohort revenue, ROAS of a new campaign, an A/B test that ended yesterday — shows only the beginning of the story.
LTV Predictions fill in the rest. For every cohort, web2wave predicts how much revenue it will have earned at each milestone of the cohort report, up to 12 months, and uses those predictions everywhere you make decisions on revenue: the cohort reports, the UTM report with pLTV and pROAS, the Overview dashboard, experiments and, if you want, the purchase events sent to Meta.
How predictions are built
Cohorts, not individual users
Predictions are made for cohorts — groups of subscriptions that started in the same period — based on how your earlier cohorts actually renewed. A prediction is always "what this group is expected to earn", never a guess about one particular person.
The model accounts for the things that make subscription revenue behave differently — your plans and billing periods, trials and intro offers among them — so a weekly trial and a full-price yearly plan are never predicted alike.
Predictions are recalculated every day. As soon as a cohort reaches a new milestone, the actual number replaces the prediction and the rest of the curve is updated.
Your data first — platform data only where yours is not enough
How much a prediction relies on your own history depends on how much history you have for the plan and the milestone in question:
- When there is enough data for your prices and subscriptions, predictions are built on your own data only: your plans, your offers, your renewals.
- When your own data is not enough yet — a new project, a new plan, or a milestone your cohorts haven't reached often enough (for example 9 or 12 months) — web2wave uses a mix of your data and anonymized data from the whole platform for that part of the curve. The platform part reflects how comparable subscriptions typically renew across the platform.
- As your cohorts mature, the weight of your own data grows automatically, and the platform part fades out. Nothing needs to be switched — the next daily recalculation simply relies more on you.
The same project can therefore be predicted from its own data at short milestones (where it has plenty of history) and from the mix at long ones (where it doesn't yet). The confidence of each prediction tells you which case you're looking at — see Confidence and range.
For uncommon plan types, where there is no reliable platform-wide pattern, predictions are made from your own data only once your project has enough history; until then those cells stay empty.
Privacy: anonymized data and opting out
- Only anonymized, aggregated statistics are used. No personal data (emails, names, user IDs, payment details, quiz answers) goes into the shared model.
- No project can see another project's data. Data is combined across many projects, so no single project's name, prices or numbers can be seen or derived from it.
- Regions stay separate. Data of projects hosted in the EU is not combined with data of projects hosted in the US.
- You can opt out. If you don't want your project's data to be used to improve the shared model, contact support (via the in-app chat or your account manager) and ask to disable LTV training on your project's data. After the opt-out your project's data is no longer used for anyone else's predictions.
Opting out of training also means your project is predicted from its own data only. It no longer uses platform data either, so cohorts and milestones where your own history is short get no prediction until your data grows.
Your own predictions are always computed for your project only and are never shown to anyone else.
Availability and switching it on
LTV Predictions are available on the Advanced plan.
- Open Projects, open your project, and go to the Analytics tab.
- Turn on Enable LTV Predictions and save the project.
You can also switch them on directly from the reports: the cohort tables, the UTM report and the Overview dashboard show an Add LTV Predictions button while predictions are off. On a plan below Advanced, the button explains the upgrade instead.
The first predictions are calculated within a few minutes of switching on — refresh the report to see them. After that they are updated daily.
Turning the setting off removes the predictions from all reports. Values you send yourself through the API (see Your own projections via the API) are never changed by the platform.
Confidence and range
Every predicted number comes with a confidence level and a likely range.
| Confidence | What it means |
|---|---|
| High | Built on your project's own history. |
| Medium | Built on your project's history, complemented by platform data where yours is thinner. |
| Low | Built mostly on platform data: your project has little history at this milestone yet. Treat it as a rough estimate. |
The range shows where the actual value is likely to land, based on how accurate past predictions turned out to be. The further ahead a milestone is, the wider the range.
A few things lower the confidence on purpose:
- Small cohorts. A cohort with few subscriptions is noisy no matter how good the model is, so its confidence is capped. For small projects, cohort reports open on the Month grouping by default, and a hint suggests it whenever weekly cohorts are too small to be steady.
- Filtered views. When a report is filtered (by quiz, UTM, country, platform, experiment and so on), predictions are less precise and are shown as Low at best.
- Recent changes. If a plan's renewals recently changed noticeably (a price change, a new offer), the confidence is lowered until things settle.
Where you see predictions
Cohort reports
In Analytics → Cohorts, the cells of cohorts that haven't reached a milestone yet are no longer empty:
- Predicted cells are shown in italic with a
~sign and a colored underline: green for high, orange for medium, red for low confidence. - Hover a predicted cell to see the confidence, the likely range, what the prediction is built on and how many subscriptions the cohort has.
- Predictions are saved every Monday. The tooltip shows what was predicted for this cell earlier and how the prediction has moved since — and, once the cell is actual, how close that earlier prediction was. This lets you check the accuracy on your own data.
- In the cohort charts, predicted points are hollow and connected with a dashed line.
Predictions appear in every cohort table — Cohort LTV per paid user, Cohort Revenue, the revenue multipliers and Cohort ROAS (with a connected Meta Ad Account) — for weekly, monthly, quarterly and yearly groupings. The day grouping shows actual values only.
UTM report: pLTV and pROAS
With LTV Predictions on, the UTM report shows prediction columns automatically:
- pLTV 3 mo / 6 mo / 9 mo / 12 mo — revenue the row's subscriptions have earned so far plus the predicted rest, at 92, 184, 279 and 367 days.
- pROAS 3 / 6 / 9 / 12 mo — predicted revenue divided by ad spend. Needs a connected Meta Ad Account.
The columns work for every grouping and drill-down level: source, campaign, ad set, ad, country, platform, quiz and paywall. Spend is known per UTM only, so groupings by quiz, country or platform show pLTV but no pROAS.
A pLTV cell shows "–" when too few of the row's subscriptions have a prediction at that horizon to give a reliable total. The newest subscriptions are included as well, so pROAS of the last few days isn't understated.
Overview dashboard
The Overview dashboard adds prediction tiles:
- Cohort Revenue 12m — what the subscriptions created in the selected period will have earned 12 months after they started: revenue so far plus the predicted rest. Empty while too few of them have a prediction.
- pROAS 12m — Cohort Revenue 12m divided by the Meta Ads spend of the period.
- Payback — the day after which the period's subscriptions are predicted to have earned back the period's ad spend. Lower is better.
pROAS and Payback need a connected Meta Ad Account.
Experiments
A/B test results include pLTV per visitor at 3, 6 and 12 months for every variant: revenue so far plus the prediction, divided by the variant's visitors. The relative difference against the control is highlighted when it is statistically significant. This lets you judge a paywall or pricing test by long-term revenue, not just first payments — useful when variants differ in plans, trials or intro prices. pLTV can also be plotted on the experiment's graph over time. See Experiments.
Meta: predicted LTV with purchase events
You can send the predicted LTV of each new subscription to Meta together with the purchase event, so that Meta campaigns can optimize for long-term value instead of the first payment. See Send predicted LTV to Meta.
What is and isn't predicted
- Revenue is gross — the amounts charged, before payment fees and taxes, converted to USD.
- Horizon — milestones of the cohort report up to 12 months (367 days). Nothing is predicted beyond a year.
- Plans with too little history — when there isn't enough data for a reliable prediction, the cell stays empty rather than showing a made-up number.
- Mature cells are never touched. Once a cohort has reached a milestone, the report shows the actual value.
- Predictions are not written to your subscriptions. They live in the reports only. The subscription API, webhooks and exports keep showing actual revenue and, if you send them, your own projections.
Your own projections via the API
If you calculate LTV yourself (for example in your app's backend), you can keep sending it via the API in the subscription fields projected_revenue_32d, projected_revenue_62d, projected_revenue_92d, projected_revenue_184d, projected_revenue_279d and projected_revenue_367d (cumulative revenue in cents at 1, 2, 3, 6, 9 and 12 months).
- With LTV Predictions off, the cohort reports and the UTM report use your values: such cells are marked with a blue underline and the tooltip Projected from the values your project sends via the API.
- With LTV Predictions on, the reports use the platform's predictions. Your values are kept as they are and are still returned by the API — the platform never overwrites them.
Common questions
Why did a prediction change since yesterday?
Predictions are recalculated daily with the latest data: when a cohort reaches a new milestone, the actual number replaces the prediction, and fresh data refines the rest.
Why is a cohort shown with Low confidence when my project is large?
Confidence is per cohort and per milestone. Long milestones (9–12 months) need cohorts that have already lived that long — a project that launched a plan recently has no such history yet and relies on platform data there. Small weekly cohorts and filtered views are also shown at a lower level.
Why are some cells still empty with LTV Predictions on?
There isn't enough data for that plan yet. The cells fill in automatically once there's enough data.
Will a new project get predictions right away?
Yes, for common plans. A new project is predicted from a mix of its own early data and anonymized platform data, with Low confidence at first. As its cohorts mature, the predictions move to its own data and the confidence grows.
Can I see how accurate the predictions were?
Yes. Predictions are saved every Monday; hover any cell that has become actual to see what was predicted for it earlier and how far off it was.
Related pages
- Connect a Meta Ad Account — spend and ROAS — spend for pROAS, Payback and Cohort ROAS.
- Analytics Integration Guide — sending predicted LTV with Meta purchase events.
- Experiments — pLTV per visitor in A/B test results.
- Advanced analytics and event tracking — revenue value settings and the UTM report.