Roi Shivek
Ad networks deliver meaningful performance by attracting users who are likely to complete key business actions, such as subscribing to a service or purchasing a product. To avoid wasting resources and advertising budgets, they continuously optimize ad delivery to reach users most likely to convert. To do this effectively, they must analyze past users, identify common characteristics among those who convert, and bid on similar users. This follows the standard exploration/exploitation approach: bringing in a diverse set of users, observing who converts, and refining bidding based on those insights.
In performance marketing, the shift from exploration to exploitation is known as "exiting the learning phase." This article is the first in a series examining the role of conversion signal health in campaign optimization. In this piece, we’ll focus on the key conditions an ad network must meet to successfully move from the learning phase into sustained optimization.
For a network to learn and optimize towards a business action or goal (e.g., a purchase event, a subscription, or a value prediction event), the following conditions must be met:
Once the network has gathered enough meaningful actions from a sufficient number of users, it will begin actively bidding on users who are more likely to complete the desired action. Each ad network has its own criteria for transitioning from exploration to exploitation.
As of writing, Meta requires at least 50 conversion events per week to exit the learning phase, whereas Google Ads requires fewer. Other networks have their own signal volume thresholds and timeframes for this transition.
It is crucial to measure signal matching volume directly within the ad platform. A healthy signal means the ad network can accurately match conversion signals to users it recognizes. This is achieved by linking various identifiers that advertisers share when posting back conversion events. These identifiers may include device characteristics, device IDs, and user IDs.
Importantly, using user IDs to establish a healthy signal does not conflict with data privacy or PII concerns, as long as they are implemented in a privacy-preserving manner. Advertisers should rely on standard hashed or pseudonymized identifiers rather than raw PII data. Examples include hashed emails, user pseudo IDs, device identifiers like IDFA/IDFV etc.
pLTV introduces an additional layer of consideration in ad network exploration, particularly in conversion optimization-based campaigns. To help the network improve the quality of users it bids on, pLTV must selectively report back only those users who not only complete the desired action but are also predicted to have a higher long-term value for the advertiser’s business.
To illustrate the difference, let’s briefly compare a standard conversion event to a pLTV conversion event across the two main campaign optimization categories:
Given enough conversions, the network should be able to learn and optimize toward users who are likely to complete the targeted conversion action—whether that’s making a payment, subscribing to a trial, or another key business goal.
However, the further down the funnel the conversion action is, the lower the potential number of conversions. For example, if a subscription app optimizes for paying subscribers, it must first acquire a large volume of trial users—only a fraction of whom will eventually convert to paying customers. The network needs enough paying subscriber conversions within the optimization window to effectively learn and optimize. This can be challenging if the total number of conversions is too low.
This is why many advertisers optimize for higher-funnel actions (e.g., trial signups) rather than final conversion events (e.g., payments), ensuring the network has enough data to learn from while still driving somewhat valuable outcomes.
pLTV can address different challenges in conversion optimization:
Scaling Issues – If your campaign lacks enough conversion events within the optimization window (e.g., paying subscribers), pLTV can predict conversion likelihood early on and report events for high-potential users, increasing learning signals and improving scale.
ROAS Challenges – If early conversions are abundant but some conversions are by customers who do not mature to become high value customers, pLTV filters events based on predicted long-term value, reporting only high-value customers above a set threshold. This reduces event volume but improves campaign efficiency.
Given enough conversions with value data, the network should be able to learn and optimize toward short-term customers.
Because value is involved, you should aim for a large number of conversions with different values so the network can not only optimize for converters but also target and bid based on its prediction of a user’s potential value. Instead of only predicting the likelihood of a conversion action, the network also predicts the potential value of each user.
If you have enough conversions with value data for the network to learn and optimize for short-term customers using your standard conversion event, introducing pLTV should not affect its ability to optimize for long-term ones.
In value optimization campaigns, where every conversion is assigned a value, pLTV simply replaces the reported short-term deterministic value with a long-term predicted value for each user. Since every conversion is still shared, the network continues learning as before, but now optimizes bids based on predicted future value rather than immediate returns.
For ad networks to transition from exploration to exploitation, signal health is key. A strong, privacy-compliant signal ensures accurate conversion tracking, enabling networks to learn and optimize bids effectively. This is even more critical with pLTV, where predicted long-term value guides optimization. Without a healthy signal, networks struggle to move beyond the learning phase, limiting scale and ROAS improvements. Advertisers must invest in signal quality to fully unlock ad network optimization.
Your data warehouse has incredible value. Our causal AI helps unlock it.