Table of Contents
Predictive LTV bidding is a user acquisition method that feeds ad platforms a predicted lifetime value for each prospective user, so the auction optimizes toward the people most likely to generate long-term revenue rather than the cheapest installs. Instead of telling Meta or Google “find more installs,” predictive LTV bidding (also called pLTV bidding or value-based bidding) tells the algorithm “find more of the users who will still be paying you in 90 days.” This is the difference between buying volume and buying value, and for subscription and in-app-purchase apps it is the single biggest lever on profitable scale.
At Admiral Media, a performance marketing agency that has managed over €500M in mobile ad spend across 150+ brands, the shift from volume bidding to value-based bidding is the recurring pattern behind durable growth. This guide explains the mechanism of pLTV modeling, how proxy events feed value signals into the auction, how value-based bidding works on Meta and Google, and the cohort economics that decide whether your acquisition is actually profitable.
What Is Predictive LTV Bidding?
Predictive LTV bidding is the practice of optimizing ad delivery against a modeled future value for each user instead of a binary install or a single early purchase. The ad platform receives a value signal, predicts the return on ad spend a given impression is likely to produce, and bids higher for users with higher predicted value.
Three terms anchor the concept. Lifetime value (LTV) is the total revenue a user generates over their lifecycle with the app. Predicted LTV (pLTV) is a model’s early estimate of that lifetime value, produced within hours or days of install from behavioral signals, before the real revenue has materialized. Value-based bidding is the bid strategy that consumes those values: on Meta it is value optimization and the ROAS goal, on Google it is Target ROAS (tROAS) bidding for App campaigns.
The reason this matters is economic. Two users can cost the same to acquire and have wildly different value. A volume-optimized campaign treats them identically. A value-optimized campaign learns to find more of the profitable one. The Admiral Media team has seen this distinction decide whether an app can scale spend at all, because volume buying eventually drowns a profitable cohort in unprofitable installs.
Why Install-Volume Optimization Caps Your Growth
Install-volume optimization caps growth because cost per install says nothing about whether a user is worth acquiring. The bottleneck is not the price of the install, it is the value distribution of the users behind that price.
When a campaign optimizes for installs or cost per install (CPI), the algorithm hunts for the cheapest humans who will tap “install.” Early on, those installs look efficient. As budget scales, the platform reaches deeper into low-intent inventory, CPI creeps up, and the average value per user falls because the incremental users are progressively less likely to subscribe or purchase. The campaign hits a ceiling: spend more and ROAS collapses, spend less and growth stalls.
Value-based bidding breaks that trade-off by changing the objective. The platform is no longer rewarded for cheap installs, it is rewarded for predicted value. In Admiral Media’s campaigns across subscription, gaming, fintech, and dating apps, reframing the goal from “more installs” to “more value” is what unlocks the ability to scale spend without watching payback periods blow out. For a deeper treatment of how LTV modeling survived the iOS privacy changes, see Admiral Media’s analysis of whether LTV is still relevant in the post-iOS 14 world.
| Dimension | Volume bidding (CPI / installs) | Value-based bidding (pLTV) |
|---|---|---|
| Optimization objective | Lowest cost per install | Highest predicted lifetime value |
| Signal sent to platform | Install event (binary) | Modeled conversion value per user |
| Behavior at scale | CPI rises, average value falls | Holds value as spend grows |
| Right for | Early seeding, broad awareness | Subscription and in-app-purchase economics |
| Primary risk | Unprofitable volume | Mis-specified value model |
| Decision metric | CPI, install count | Cohort ROAS, payback period |
How Predictive LTV Models Actually Work
Predictive LTV models work by mapping early, observable user behavior to a future revenue estimate, then passing that estimate to the ad platform as a conversion value. The model does not wait for lifetime revenue to arrive, it predicts it from the first signals available.
Proxy events: the inputs to the model
A proxy event is an early in-app action that correlates with long-term value: completing onboarding, reaching a paywall, finishing a tutorial level, starting a trial, or hitting a usage milestone in the first session. Because true revenue often takes weeks to accumulate, and because iOS measurement windows are short, the model leans on these proxies to estimate value fast. The closer a proxy event correlates with eventual payment, the more accurate the predicted value. Choosing the right proxy events is the highest-leverage modeling decision in value-based UA.
Modeled conversion values: the output the auction consumes
Once the model estimates value, it assigns a modeled conversion value to the user and sends it to the platform. On Android and web, where signal is richer, these values can be granular. On iOS, the value is compressed into the limited postback structure that Apple’s privacy framework allows. Apple’s SKAdNetwork and its successor AdAttributionKit deliver coarse, privacy-thresholded conversion values rather than user-level revenue, so the model must encode value into a small set of buckets. Admiral Media’s AdAttributionKit iOS measurement playbook covers how to design those conversion value schemas without losing the signal that bidding depends on.
Cohort LTV: the truth the model is judged against
Cohort LTV is the actual revenue a group of users acquired in a given week or month generates over time. It is the ground truth that validates or invalidates the prediction. A predicted LTV model is only useful if its early estimates track real cohort revenue. The Admiral Media team treats cohort LTV curves as the scoreboard: if predicted value and realized cohort value diverge, the proxy events or value mappings get re-specified before more budget goes in. Target ROAS bidding in particular needs a steady stream of valued conversions to learn from, which is why Google’s own Target ROAS documentation stresses conversion value tracking as a prerequisite.
The Admiral Media Value-First Acquisition Framework
Value-based bidding fails when teams flip the switch without building the value foundation underneath it. The Admiral Media Value-First Acquisition Framework sequences the work so the model, the signal, and the bidding strategy reinforce each other instead of fighting.
The Admiral Media Value-First Acquisition Framework
- Define the value event: Decide what “valuable” means for the app: trial start, subscription, repeat purchase, or a revenue threshold. The entire system optimizes toward this definition, so it must reflect real profitability, not a vanity action.
- Map proxy events to value: Identify the early in-app behaviors that predict the value event, and confirm the correlation with historical cohort data before trusting them. Weak proxies produce confident but wrong predictions.
- Build the conversion value schema: Translate predicted value into the conversion values each platform accepts, designing iOS schemas around SKAdNetwork and AdAttributionKit constraints so coarse postbacks still carry usable signal.
- Seed the learning phase with clean signal: Give each platform enough valued conversions to exit learning before judging performance. Starve it of value events and the algorithm reverts to noise.
- Bid to value, not volume: Switch the objective to value optimization or Target ROAS, set a defensible ROAS goal, and resist the urge to read CPI as the success metric.
- Validate against cohort LTV: Compare predicted value to realized cohort revenue on a rolling basis. When they diverge, fix the model upstream rather than chasing the bid.
- Scale the profitable cohorts: Once predicted and realized value agree, increase budget into the segments, geos, and creatives producing the strongest cohort ROAS, and hold the value bar as spend climbs.
This sequence is deliberately model-first. Admiral Media’s experience managing user acquisition across 150+ apps is that the bidding strategy is the easy part. The durable advantage comes from a value definition and proxy-event map that actually predict revenue.
Value-Based Bidding on Meta and Google
Meta and Google both support value-based bidding, but they consume value signals differently. On Meta, value optimization and the ROAS goal let the system bid for the highest predicted purchase value rather than the most conversions. Meta requires a baseline of valued events to operate: per its ROAS goal documentation, value optimization needs an active signal source sending purchase values, with a minimum threshold of recent valued conversions before it can deliver reliably. On Google, Target ROAS bidding for App campaigns predicts the value of each potential conversion and bids to hit an average return target, which depends entirely on the conversion values you feed it.
The mechanism is the same underneath both: the platform’s model multiplies its conversion-probability estimate by your supplied value, and bids proportional to expected value. That is why the quality of your value signal, not the cleverness of the bid setting, determines the outcome. A precise value schema with mediocre creative will outperform brilliant creative bidding on installs, because the auction is being steered toward the right people.
| Bidding approach | What it optimizes | What it needs to work | Best fit |
|---|---|---|---|
| Cost cap / CPI | Volume at a target cost | An install event | Cold-start seeding |
| Meta value optimization | Highest total purchase value | Purchase events with values, recent valued-conversion volume | Apps with variable purchase values |
| Meta ROAS goal | Average ROAS around a target | Stable value signal and patient budget | Established value models |
| Google Target ROAS (App) | Conversion value at a return target | Conversion value tracking, sufficient valued conversions | UAC scaling on value |
For apps whose revenue lives behind a subscription, the value model has to account for trial-to-paid conversion and renewal, which is why Admiral Media pairs value-based bidding with funnel design. The agency’s work on web-to-app funnels for subscription apps shows how the value signal and the funnel have to be engineered together.
Proof: What Value-Led User Acquisition Looks Like in Practice
Value-led acquisition produces compounding gains across the funnel, not just a lower install price. Three Admiral Media case studies show the pattern.
Admiral Media managed NeuroNation’s Google App Campaigns with a structured creative testing approach and a custom performance-ranking methodology called pRank, achieving a 117% increase in ROAS alongside a 39% reduction in CPI, a 66% lift in installs, a 32% increase in purchases, and a 42% rise in net cohort revenue over the measured period. The net cohort revenue gain is the value-led signal that matters most: more installs at a lower cost is good, but more revenue per cohort is what proves the acquired users were actually worth more.
For PURE, a dating app targeting a competitive CPI and a high D7 ROAS on US Android, Admiral Media tested a programmatic DSP channel against an established self-attributing network. The DSP channel delivered a CPI of $2.44, roughly four times lower than the self-attributing network’s $9.43, while exceeding D7 ROAS goals, which produced a 74% reduction in CPI and triggered expansion into new market launches. Acquiring D7-ROAS-positive users at a quarter of the cost is exactly what value-aware channel selection is meant to find.
With Clark, a German insurance fintech app, Admiral Media optimized toward mid-funnel proxy events rather than only top-of-funnel installs, recognizing that users explore the app before converting. By optimizing for lower-funnel events that sit outside the obvious purchase path, the team cut cost per lead by 50% and cost per level achieved by 47%, while lifting conversion rate by 41% and installs by 18%, with CPI down 29%, comparing month three to month one. Clark is a clean illustration of the proxy-event principle: optimizing toward a value-correlated mid-funnel action, not just the cheapest install, moved every downstream metric. See the full Clark case study for detail.
Across all three, the through-line is consistent with how Admiral Media approaches performance marketing generally: define what value means, send that signal to the platform, and let the auction do the targeting work.
Common Pitfalls in Predictive LTV Bidding
Most predictive LTV bidding failures trace to the value foundation, not the bid setting. The recurring mistakes are predictable.
The first is optimizing for a proxy event that does not actually predict revenue. If “tutorial complete” feels valuable but does not correlate with subscription, the model will confidently acquire users who never pay. The second is switching to value bidding before the platform has enough valued conversions to learn, which leaves the algorithm guessing and produces volatile, disappointing early results that get the strategy abandoned prematurely. The third is reading CPI as the scorecard after moving to value bidding: CPI often rises under value optimization because the system is deliberately paying more for better users, and a team watching CPI will “fix” a campaign that was working. The fourth is letting the iOS conversion value schema go stale, so the coarse signal AdAttributionKit returns no longer maps to current value. The fifth is ignoring cohort validation entirely and trusting the predicted value indefinitely, even as user quality drifts.
The discipline that prevents all five is the same: treat the value model as the product, validate predictions against real cohort revenue, and judge campaigns on cohort ROAS and payback period rather than install economics. For teams sizing the investment, Admiral Media’s view on user acquisition economics is that value-led UA is where agency leverage compounds, because the modeling work pays off across every channel at once.
Frequently Asked Questions
What is predictive LTV bidding?
Predictive LTV bidding is a user acquisition method where ad platforms optimize delivery against a modeled future lifetime value for each user instead of a single install or early conversion. A model estimates each user’s likely long-term revenue from early behavior, sends that value to the platform, and the auction bids higher for users with higher predicted value. The goal is to acquire users by value rather than by volume, which protects profitability as spend scales.
How is pLTV bidding different from cost per install optimization?
Cost per install optimization rewards the platform for finding the cheapest installs, while pLTV bidding rewards it for finding the most valuable users. Under CPI optimization, value per user typically falls as budget grows because the algorithm reaches into lower-intent inventory. Under value-based bidding, the objective itself is value, so the system keeps targeting profitable users even at higher spend. CPI often rises under value bidding, and that is expected, not a problem.
What are proxy events in value-based user acquisition?
Proxy events are early in-app actions, such as completing onboarding, reaching a paywall, or starting a trial, that correlate with long-term value. Because true lifetime revenue takes weeks to accumulate and iOS measurement windows are short, predictive models use proxy events to estimate value quickly. The accuracy of the prediction depends heavily on how well the chosen proxy events correlate with eventual payment, making proxy selection the most important modeling decision.
Does predictive LTV bidding work on iOS after the privacy changes?
Yes, but it requires designing value signals around Apple’s privacy framework. SKAdNetwork and AdAttributionKit return coarse, privacy-thresholded conversion values rather than user-level revenue, so the value model must encode predicted value into a small number of buckets. Value-based bidding still works on iOS when the conversion value schema is built carefully, which is why schema design is a core part of any serious iOS value strategy.
How much conversion data do you need for value-based bidding to work?
Each platform needs a baseline of valued conversions before its model can deliver reliably. Meta’s value optimization requires an active signal source sending purchase values plus a minimum volume of recent valued conversions, and Google’s Target ROAS depends on conversion value tracking with enough valued conversions to learn from. Switching to value bidding before that threshold is met is a common cause of early failure, because the algorithm is forced to optimize on insufficient signal.
What metric should I track instead of CPI under value-based bidding?
Track cohort ROAS and payback period rather than CPI. Cohort ROAS measures the revenue a group of acquired users generates relative to their acquisition cost, and payback period measures how long it takes to recover that cost. These metrics reflect whether the users you bought were actually valuable, whereas CPI only reflects what you paid to acquire them. Judging value-based campaigns on CPI leads teams to undo profitable optimization.
Should an app run volume bidding or value bidding?
Most apps need both at different stages. Volume bidding is useful for cold-start seeding when there is not yet enough value signal for a model to learn from. Once an app has a working value definition and sufficient valued conversions, value-based bidding becomes the lever for profitable scale, especially for subscription and in-app-purchase economics. The transition point is when predicted value reliably tracks realized cohort revenue.
