Predictive LTV (pLTV)

What Is Predictive LTV?

予測LTV is a form of predictive analysis. It is a metric used to estimate a user’s value in the future, before the actual value has been generated. It is related to traditional lifetime value (LTV), which looks at past data and measures revenue that a user or cohort has already produced. Unlike traditional LTV, pLTV is a forecast. It uses the data available from early lifecycle signals to infer and estimate how much revenue users of a cohort are projected to generate at a later date. 

pLTV is extremely useful for user acquisition teams, who have to make decisions quickly. Marketers can’t always wait weeks or months for cohorts to mature before deciding whether to add more budget and scale, maintain, or pause a campaign. For example, a user acquired on Day 1 may eventually generate value through:

• アプリ内課金
• Advertising revenue
• Subscription payments
• 更新
• Other monetization events

A good pLTV metric is able to estimate and project value based on certain signals in the algorithm. 

Predictive LTV vs. Historical LTV

Historical or traditional LTV measures realized value. Predictive LTV estimates future value. Let’s consider two user cohorts seven days after acquisition.

キャンペーンA

• Day 7 realized LTV: $1.50
• Predicted future LTV: $4.20

キャンペーンB

• Day 7 realized LTV: $1.70
• Predicted future LTV: $2.40

If we examine realized value, then Campaign B looks stronger than A. However, looking at the pLTV projections, Campaign A is the winner. The result is based off of the early signals from Campaign A. They could have had more sessions, better retention, higher engagement, stronger purchase intent, or a combination of all of these. pLTV is a metric that helps identify early signal patterns and communicate their strengths before revenue is fully there. 

How Does Predictive LTV Work?

Predictive LTV models typically combine early user behavior with historical patterns from mature cohorts. Some of the possible signals include:

• Early revenue
• セッション
• Retention
• 広告インプレッション数
• Purchase activity
• Engagement depth
• Acquisition source
• 地理
• プラットフォーム
• App behavior

The job of the predictive model is to learn relationships and correlations between early signals and later projections. The historical data for an app could show that users who completed 5 sessions, viewed 10 rewarded ads, and made a purchase are more likely to generate more LTV. So, when a new cohort of the same app starts to display similar behavior, the model can estimate a higher forecast of LTV. 

Every predictive analytics approach is a bit different, each MMP has their own algorithm. Every system uses their own models, signals, and forecast windows, so it's very important that marketers who use pLTV understand what the metric represents. 

What Is Tenjin's Predicted LTV?

Tenjin includes Predicted Lifetime Value, or pLTV, as part of its revenue dashboard. The current N-Day 全pLTV（広告メディア + IAP） metric combines ad mediation revenue and in-app purchase signals into one machine-learning forecast. According to ドキュメント, Tenjin’s pLTV model:

• Uses cohort sizes from Day 1 through Day 30
• Combines ad mediation and IAP signals
• Uses matured cohort and historical data
• Processes early behavioral and monetization signals
• Has approximately 90% average accuracy
• Can use up to 78 input signals

To add the metric to your revenue report, simply go to the add metrics section in the dashboard and search for pLTV. Because it is available directly from the dashboard, marketers are able to immediately evaluate their pLTV metric according to cohorts alongside their spend and LTV. 

Predictive LTV and Subscription Apps

Predictive LTV is highly relevant when it comes to subscription apps because the subscription lifetime cycle takes longer to develop compared to many hypercasual and gaming apps. A newly acquired subscription cohort may only show some installs, trial starts, and some first payments. A month later, the same cohort could generate renewals, cancellations, reactivations, and reached a matured subscriber LTV. 

It is a similar pattern and timing issue that is observed with hybrid monetization models. And when under pressure, marketers may need to make a budget decision before the full subscriber lifecycle is visible.

The early signals for subscription LTV take into account: 

• Trial start rate
• Trial-to-paid conversion
• Plan selection
• First renewal behavior
• Early engagement
• Cancellation timing

Early signals differ for subscriber LTV by focusing on the value generated by paying subscribers. It could show that mature annual subscribers trend $80 on average. However, the predictive model may estimate a newly acquired cohort has the likelihood of generating $70. 

Similar to traditional or historical LTV, the subscriber LTV describes the value whereas the pLTV describes a projected subscriber value. Both metrics are useful for deciding on campaign economics. 

pLTV vs. pROAS

If pLTV is a projected estimate of user or cohort value, then pROAS would be a projected estimate of ROAS, or the revenue earned relative to ad spend. In order to calculate pROAS we use the simple formula: 

Predicted ROAS = Predicted Revenue or LTV ÷ Spend × 100

例えば：

• Campaign spend = $10,000
• Revenue = $6,000
• pLTV = $14,000  

In this example, the current realized ROAS is 60% and the pROAS is 140%. The pLTV projection suggests the campaign could earn beyond break-even as the cohort matures. This adds important context compared with early revenue alone. Rather than stopping the campaign based on its current ROAS, the team could continue running the campaign and monitor whether performance continues according to the projection. 

This shows that predictive LTV can be more useful than early revenue signals alone. Early signals are noisy: a small number of whales could make a new cohort seem stronger than it actually is. Likewise, a cohort with slow but steady purchases could look weak at the start, and end up with more long-term potential. The same principle applies to hybrid monetization and subscription apps: the first payment does usually reveal renewal quality.

Good predictive models take into account a wider range of signals and add a trained metric to the decision context. 

What Makes a Good Predictive LTV Model?

A useful LTV prediction model should have:

Relevant Inputs
The model needs signals that actually relate to future value.

Enough Mature Historical Data
The model needs examples of what happened to similar users in the past.

Regular Recalibration
User behavior, monetization, creative strategy, and acquisition channels change over time.

Clear Forecast Definitions
Marketers should know what revenue streams and time horizon are being predicted.

Validation Against Actuals
Predictions should be compared with realized cohort performance as the data matures.

A model that cannot be validated is difficult to trust for budget decisions.

How Can an MMP Use Predictive LTV for Campaign Optimization?

An MMP already has acquisition context such as:

• Network
• キャンペーン
• クリエイティブ
• 国
• プラットフォーム
• Spend

When predictive revenue is available in the same reporting environment, marketers can compare forecast value with acquisition cost much earlier.

This supports decisions such as:

• Scale a campaign showing strong predicted value
• Maintain spend while waiting for more data
• Reduce spend on a cohort with weak predicted economics
• Compare creative quality beyond CPI
• Prioritize channels expected to deliver higher long-term value

Tenjin's pLTV is designed for this use case: bringing a forecast into the same UA workflow where marketers already evaluate spend and actual LTV.

Predictive LTV and Hybrid Monetization

Predictive LTV becomes especially valuable when users monetize in more than one way. However, there’s a huge lack of those out there supporting apps that have a monetization mix. Many platforms continue to treat app monetization as if it exists as a single stream of IAA or IAP or subscription, never a hybrid or mix. This can turn into a measurement issue if you’re using more than one revenue stream.

For example, if a hybrid monetization game generated both IAA and IAP revenue and the model only predicts one stream, the total value could be understated. That’s why Tenjin's pLTV combines ad mediation and IAP signals and it reflects a broader measurement principle: user value should be evaluated across the monetization streams that contribute to the business. 

Read more about this on our blog: ハイブリッド収益化における予測LTV：4つの課題を解決.

Best Practices for Using Predictive LTV

Treat the prediction as an estimate.
It supports decisions but does not replace realized performance.

Know what revenue is included.
Different pLTV products can forecast different monetization streams.

Compare predictions with actuals.
Validate the model as cohorts mature.

Use pLTV alongside spend.
Predicted value becomes more useful when translated into expected ROAS or profitability.

Segment by campaign and channel.
Blended predictions can hide meaningful acquisition differences.

Avoid comparing cohorts at different definitions.
Make sure forecast windows and revenue inputs are consistent.

Use enough data.
Very small cohorts can produce unstable signals.

関連用語

• Lifetime Value
• Predictive Analytics
• Return on Ad Spend
• Subscriber Lifetime Value
• Subscription Events
• 広告収益のアトリビューション
• In-App Purchase
• ハイブリッド型収益化

よくある質問

What is predictive LTV?

Predictive LTV is an estimate of how much revenue a user or cohort is expected to generate in the future based on early and historical data.

What is the difference between LTV and predictive LTV?

LTV can describe realized value that has already accumulated, while predictive LTV forecasts value that has not yet occurred.

How is predictive LTV used in mobile marketing?

Marketers use it to evaluate campaign quality and expected ROAS earlier, before cohorts have fully matured.

What does Tenjin's pLTV include?

Tenjin's current N-Day All pLTV combines ad mediation and in-app purchase signals into a machine-learning forecast for mobile UA analysis.

Is predictive LTV accurate?

Accuracy depends on the model, inputs, app, and cohort. Tenjin's current documentation reports approximately 90% average accuracy for its pLTV metric, but predictions should still be validated against actual performance as cohorts mature.