How Coinbase Uses Signal Engineering to Bid on pLTV in a Market That Moves by the Hour

Two years is roughly how long it takes a new Coinbase user to earn back what it cost to reach them, and a lot can happen in crypto during that time. A bull run can lift revenue from everyone who signed up in the last 24 months, and a drawdown a few weeks later can change how new users behave. Scott Ridout, Coinbase’s Group Performance Marketing Manager, needed to tell Google which users were on track to pay back within days of sign-up, not years later.

At Google Rethink Apps 2026, Scott walked through how his team got there, through several iterations of pLTV measurement and optimization that anyone with a long payback window will recognize. He shared the results and the lessons he’d pass on to teams making the same move. 

‍

Same first trade, wildly different revenue paths

‍
“For six years, we've been trying to answer one deceptively simple question: how much is a new user actually worth? Two users can complete the same first trade event, but their future value to the business can be dramatically different. This can be especially difficult in crypto, a volatile industry and a high-consideration risk asset.
‍
We stopped trying to predict lifetime value and tried to predict something in 30 days. We paired that with user-level predictions in partnership with Voyantis and moved from target CPA to target ROAS on Google App campaigns.
‍
On Android app campaigns, initial experiments against tCPA showed improvements in cost per first trade, trade volume, and day-30 ARPU. We're now rolling out value-based bidding across every Android app campaign globally.”

‍
Was it the ad or the price of bitcoin?

‍
“Predicting user value is genuinely one of the hardest problems facing consumer marketing.

People do not behave like they're buying a new shirt. Engagement depends on trust, financial capacity, experience, and anything that is happening in the market at any given time.
‍
Crypto is a high-consideration, high-volatility risk asset. A user who signs up in a bull run and a user who signs up a few weeks into a drawdown can have wildly different trajectories, even if the campaign, creative, and conversion event are exactly the same.
‍
Market conditions don't just affect new users. A bull run can lift the revenue of everyone who has signed up with us over the last 24 months. When performance marketing looks great, how much of it was the ad vs. the price of bitcoin?
‍
So, we had a bidding blind spot.
‍
For years we were using target CPA, flat CPA bidding that treated everyone the same. You can imagine that someone living here in Manhattan has a lot more purchasing power than someone living in Phoenix. By using tCPA, we were overpaying for low-intent users and underpaying for high-value traders. In a volatile market, this is an expensive way to be wrong.”

‍
Moving from two-year LTV to 30-day payback

‍
“In order to get to value-based bidding, we first had to refine measurement, and this happened in three different attempts.

In 2021, we started with a predictive ARPU model: 12-month ARPU extrapolated forward. It had a few flaws. It assumed linear trading behavior, and it also looked at the campaign level. So all users in one campaign got the same value, and we were bidding on averages and nothing else.

In our second attempt in 2023, we built a machine learning predictive LTV model. It had all this fancy data science stuff: a two-stage gradient boosting model, macro conditions built into the outputs, and three market scenarios. However, for bidding we could only use it at the channel and geo level.

The model also needed months of historical data to make the prediction, so our LTV predictions always lagged the market. We were bidding today on a picture of the world from two months ago, and in crypto, two months is an era.

So in our last attempt in 2025, we changed everything.

We threw away pLTV and predictive ROI as optimization metrics and replaced them with something called payback ratio. Instead of trying to predict two years of lifetime revenue, we tried to predict how much revenue a user needs to have made in the first 30 days to give us a more accurate prediction of what their two-year LTV is going to be.

This is essentially realized revenue from a user divided by pacing revenue, which is a function of market cost. So a payback ratio of 1 means you're on track. If you have one of 0.8, it means you're about 20% behind. We also deliberately locked this metric at 30 days, which helped isolate ad performance from the market. This is the measurement solution that ultimately allowed us to unlock value-based bidding.”

How signal engineering unlocked value-based bidding for Coinbase

“So in 2026, we partnered with Voyantis to generate user-level value predictions. These are fired at the “know your customer” (KYC) step in our funnel and optimized against payback ratio.

Voyantis looked at a variety of signals from our sign-up flow and ran regression analysis against the past five years of our user revenue data in order to generate predictive models specifically for Google Ads value-based bidding. After a few rounds of backtesting and a blessing from our data science team, we were ready to test tCPA against tROAS.

So what were the results?

First trades up 19%, cost per first trade down 20%, day-30 ARPU up 90%, day-30 ROAS up 114%. We were clearly driving higher-value users at a lower cost.

Since then, we've started rolling out tROAS across all of our Android app campaigns, and next we're going to continue improving the feedback loop between early behavior, predicted value, and realized payback.”

‍
Scott's advice for growth teams moving to tROAS

‍
“The takeaway is not simply that tROAS outperformed tCPA. The deeper lesson is that value-based bidding only works when the value signals are timely, granular, and connected to the economics of the business.

We spent years trying to build a more accurate prediction of a distant future. What actually unlocked performance was building a faster, more granular prediction of the near future – one that we could feed back to Google while it still meant something.”

Scott's talk matched the message and roadmap Google delivered. It’s time to get fluent in ROAS and max bidding. For teams still bidding on a flat CPA, Coinbase's results show what that shift can look like when the value signal behind it is tied to how the business makes money.

If you're bidding against a long payback window and want to see what user-level predictive signals could do for your Google campaigns, talk to our team, or get the playbooks from apps that have already made this shift, including Credit Genie, Upside, Klar, and Rappi.