> Site index: https://usehardal.com/llms.txt
> Every content route also serves markdown at <url>.md

# What Is Predicted Lifetime Value?

Predicted lifetime value estimates revenue a customer or install may bring later. Ad platforms can use it in value-based bidding before that revenue arrives.

Source: https://usehardal.com/glossary/pltv

---

Predicted lifetime value (pLTV) is an estimate of the revenue a customer, install, or player will generate over a future window you choose. Historical lifetime value adds up what already happened. pLTV is the forward number, produced early enough that an ad platform can bid with it.

The estimate is only as good as the events and the horizon behind it. It is not cash received.

## How does it work with value-based bidding?

[Value-based bidding](https://usehardal.com/glossary#value-based-bidding) optimizes toward conversion value rather than conversion count. If the value you send is "install equals 1," the platform finds more installs. If the value is expected revenue, it can shift spend toward the installs that look more valuable, provided the values arrive in time and stay consistent.

A typical mobile loop looks like this.

1. The install or first open is recorded with a campaign identifier your measurement setup actually has. On iOS that may be a SKAN postback rather than a user-level click.
2. Early events, such as a tutorial finish or a first purchase, update a model the advertiser or the ad platform maintains.
3. The platform receives either an observed `value` on a `purchase` event or a predicted value on an event you both agreed to treat as a forecast.
4. Later purchases, ad revenue, and refunds are the score you compare to the prediction. They do not automatically rewrite yesterday's bid.

Keep predicted and observed values in different fields or different events. A report that adds them together will show the same future dollar once as a forecast and again as revenue.

## A concrete example

A game sends two events for a new player.

- `install` with `transaction_id` left empty, `campaign_id` from the measurement partner, and no revenue.
- On day 2, `pltv_update` with `user_id` (only where you are allowed to store one), `currency` `USD`, `predicted_value` populated by your model, `horizon_days` `30`, and `value_type` `predicted`.

A purchase the same week is a different event: `purchase` with `value` set from the validated store price, `currency` `USD`, `transaction_id` from the store, and `value_type` `observed`.

The ad platform's value optimization is pointed at `predicted_value` only if that is the conversion you configured. Pointing it at `purchase.value` trains it on collected revenue instead, which arrives later and misses players who will pay after the optimization window. Neither setup is free of error. They answer different questions, and the event names should say which one you chose.

If iOS campaign reporting depends on SKAN, the prediction has to be folded into a [conversion value](https://usehardal.com/glossary/conversion-value) band. The postback will not carry `predicted_value` as a decimal.

## What are the limits?

A model trained on installs that never see day-30 revenue will copy that blindness. Refunds and store chargebacks lower collected revenue. If the prediction is not reduced, bidding keeps chasing a number you will not keep.

pLTV is not comparable across vendors unless the horizon, the revenue components (purchases, ad revenue, subscriptions), and the refund rule match. "Higher pLTV" from a new model can be a longer window.

Predicted value does not create tracking permission. Sending it server-side does not bypass App Tracking Transparency or consent. If the identifier you join on is not allowed, the join is not allowed because the payload is a forecast.

Small cohorts make the estimate noisy. Do not treat a one-day swing in average prediction as a change in creative performance.

## How Hardal handles it

Hardal does not ship a lifetime-value model and does not invent a `predicted_value`. Your model, or the ad platform's, produces the number. Hardal can collect the early events and the observed `purchase` and ad revenue events, then forward the value field you mark as safe to send.

Label `value_type` so a destination does not treat a forecast as settled revenue. Send refunds against the same `transaction_id`. The [MMP revenue signals](https://usehardal.com/blog/mobile-game-mmp-revenue-signals) article describes why value-based and pLTV bidding only learn from the revenue you actually pass through, and [Hardal for Gaming](https://usehardal.com/solutions/gaming) is where that pipeline is packaged.

**Related terms:** [Impression-level ad revenue](https://usehardal.com/glossary/ad-revenue-ilrd), [Cost per install](https://usehardal.com/glossary/cost-per-install), [SKAdNetwork](https://usehardal.com/glossary/skadnetwork), [Attribution](https://usehardal.com/glossary/attribution)
