By Roi Shivek
Many brands build their own pLTV-based UA optimization solutions but face modeling and activation challenges. When DIY efforts fall short, they often abandon pLTV, missing a key hyper-optimization lever. This article explores why DIY fails, its hidden costs, and why if you’re still taking the DIY route, benchmarking against a specialized solution is the smarter approach.
Before we begin, it's important to distinguish between two types of pLTV models: those used for insights and actions, such as budget optimization and campaign targeting adjustments, and those that generate user-level predictions, which are sent to ad networks as custom events and used as conversions for campaign optimization. This article focuses on the latter.
Building a predictive lifetime value (pLTV) model in-house for automatic UA campaign optimization may seem logical. You retain full control of your data, get a tailored implementation, and expect cost savings. However, many businesses struggle, not because pLTV doesn’t work, but due to unforeseen challenges. Models drift, ad networks fail to optimize effectively, learning cycles are expensive, and results can be disappointing.
When DIY efforts fall short, companies often abandon pLTV entirely, missing out on a proven UA strategy. To prevent this, we advise against DIY, but if a business chooses to proceed, they should at least test their in-house approach against a specialized solution.
We have explored the topic of pLTV with hundreds of businesses and found that they often underestimate the complexity of successfully activating a pLTV signal while overestimating the reliability of their models. Here are the key pitfalls that often derail in-house pLTV implementations.
Which predictions from your amazingly accurate PLTV models should you send to the ad platform to deliver good ROAS? Sending pLTV signals to ad networks isn’t just about plugging in numbers, it requires unique ad network experience, continuous adaptation and robust infrastructure.
Learning from your mistakes and improving your pLTV based optimization is costly and slow by design, because UA campaigns operate on delayed feedback loops. Ad platforms optimize bids based on historical performance, meaning miscalibrated pLTV signals, for example, such as overestimated predicted values, can lead to aggressive overspending. By the time you spot an issue, after investing budget and waiting weeks for cohorts to mature, inefficient bidding has already done its damage, forcing you to fix and iterate slowly and at a cost.
Ad networks constantly evolve, making UA optimization a moving target. It’s not just about setting up a model, learning from mistakes, and expecting it to work long-term. You need to continuously adapt to stay effective. As ad platforms refine their auction dynamics, attribution models, and bid optimization methods, past performance benchmarks can become obsolete. This means that staying competitive isn’t just about building a strong mode. It requires constant monitoring, updating, and re-evaluating how your signals deliver performance with the ever-changing rules of ad network optimization.
Successfully delivering a pLTV model with a low average error doesn’t guarantee good performance in UA campaigns. Ad auctions are adversarial, meaning that over-optimistic predictions don’t just introduce noise, they actively work against you. When you overestimate a user’s future value, you’re far more likely to win the auction for them, simply because other advertisers who bid based on real purchases, avoid them. This means your model isn’t just making a mistake; it’s systematically pushing budget toward users who are unlikely to convert.
If pLTV helps expand your audience, it must be accurate about who to target. You’re bidding on future potential, while others bid on actual revenue. Without causal modeling, it's easy to pick the wrong users—those who look valuable at first but never convert. This leads to wasted budgets on users your model overestimated.
Building a pLTV model is one thing. Turning it into a fully operational, mission-critical system is an entirely different challenge. A functioning pLTV pipeline isn’t just about getting the model right; it requires real-time infrastructure, robust data pipelines, continuous monitoring, and fail-safe alerting. If something breaks, whether it’s a feature drift issue, a delay in updates, or a miscalibrated signal - you’re not just dealing with a bad report, you’re actively feeding ad platforms the wrong signals, directly impacting bidding decisions on high-budget campaigns. In performance-driven UA strategies like tROAS, a flawed pLTV system doesn’t just reduce efficiency - it burns money at scale.
Churney provides a proven, automated pLTV optimization solution that delivers immediate value without the complexity, cost, and risk of internal development. Our causal machine learning models predict user lifetime value within the first 36 hours, driving double digit improvements ROAS while continuously adapting to the ever changing landscape. But beyond execution, we’ve also become exceptionally good at qualifying whether pLTV is right for a business—before they invest a single dollar. Our free Impact Explorer tool helps companies determine if pLTV will work for them, avoiding costly missteps that lead to false conclusions about its effectiveness. If your business still chooses to build in-house despite the challenges, at least test it against Churney to ensure you’re not wasting resources, misfiring on execution, or abandoning a strategy that could significantly improve UA performance.
Your data warehouse has incredible value. Our causal AI helps unlock it.
