Churney predicts each new customer’s long-term value and sends it to Meta, Google and TikTok, so they bid for the customers who pay back.
Not sure it fits? Start with the free Growth Predictability Test.
Trusted by leading marketing and data teams globally
.svg)
The problem
Meta, Google and TikTok optimize toward whatever your conversion event tells them. If that event fires before customers show their value, the platform learns the wrong lesson.
The window
Most of your customer value arrives after that: repeat orders, renewals, upgrades.
The proxy
Installs, trial starts and first orders fire inside the window, so that is what bidding learns from.
The cost
You see it in cohort reports a quarter later, after the platform has already scaled the wrong audience.
The pilot
Every pilot is a test against your business-as-usual setup. The budget per arm, the primary metric and the uplift that counts as success are agreed in writing before anything goes live.
Weeks 1–4
Ad accounts, data warehouse and conversion events connected and validated.
Weeks 5–6
We build your model, present the opportunity sizing and sign off the test design with you.
Weeks 7–8
The Churney signal runs live for a week so volume and match quality are checked before launch.
Weeks 9–14
Churney vs. business as usual: a two-week learning phase, then four measured weeks.
Weeks 15–17
Cohorts mature for two weeks, then we read out the result against the agreed criteria: scale up, extend or stop.
%20(1).svg)
"Churney helped us scale our campaigns without sacrificing efficiency. By focusing on predicted long-term value instead of early conversions, we doubled new customer volume and improved ROAS by 190%. It’s now a core part of how we run performance marketing."
"Working with Churney allowed us to test a smarter approach to campaign optimization, one that focuses on users’ long-term value rather than just early conversions. It’s an exciting direction for improving efficiency and user quality in performance marketing."
"Many companies promise improved ROAS, but Churney actually moves the bottom line. We saw immediate improvements in efficiency that allowed us to scale without second-guessing."
"When we evaluated Churney, we focused on three criteria: Could they construct a predictive signal with appropriate calibration? Could they integrate it reliably into our data pipeline? And could they measure the outcome without bias? They delivered on all three."
"Churney helped us move past short-term signals and tap into our users' real long-term value."
"Implementing Churney’s predictive LTV model has been a game-changer for our user acquisition strategy. By identifying high-potential subscribers early, we’ve not only increased the lifetime value of our customers but also significantly improved retention rates and overall campaign performance."
"Churney has improved our advertising efficiency. By optimizing for predicted long-term value, we increased ROAS by 33% while gaining scale. The impact was so clear that we’ve now rolled out Churney’s conversion action across our Google account. Beyond that, Churney has enabled us to launch and scale new product categories that we couldn’t advertise before. It’s transformed the way we approach growth."
“The key to success is to identify potential high-value customers swiftly and channel marketing budgets towards acquiring them. Our partnership with Churney was crucial in fueling Meta’s machine-learning algorithm with pLTV values.”
"The most meaningful day-to-day change so far is that I'm checking the split a bit more closely than usual in order to learn as much as possible about the signal, for example when it leaves the learning period, when it will show the first good ROAS signals."
Build or buy
pLTV for user acquisition needs two different models. One belongs in your data team. The other is what Churney runs.
Build it
Finance planning and payback curves run on cohort predictions. Your data team should own that model, and we give you a free generator to build it.
Build yours at cohorts.churney.io →That is Churney
Bidding needs a value for every new customer, sent within hours, calibrated for how each platform learns and proven in a test against business as usual. That is a control problem, not only a prediction problem.
Why two models are required →Beyond ad platforms
Meta, Google and TikTok are the largest learners in production, and they learn from the value we predict. The same prediction works as a reward for fine-tuning LLMs and agents: each response is scored by the renewal, repeat order or churn it leads to.
Outcome rewards
Predicted long-term value as the reward for any training run or agent decision.
Continual calibration Design partners
Every reward is checked against the outcome it predicted and retrained, so it holds up under optimization.
Learning from logs Research preview
Train on the interactions and outcomes you already hold, corrected for how past decisions were made.
We have run the same RFT tasks on Vertex AI, Together, Fireworks, Tinker and SageMaker.
See the reward layer →Illustrative example. One prediction, two consumers: ad platforms and training runs.
Book a 30-minute call. We look at your setup and tell you plainly whether pLTV is worth testing.
