How Klar’s Meta Campaigns Unlocked 22% More Revenue with pLTV Bidding

Two people get approved for a Klar credit card on the same day, through the same Meta campaign. One activates the card, uses it every week, and makes every payment on time. The other never uses it or falls behind on payments within a few months. With campaigns built around a CPA target, Meta’s algorithms chased both just as hard because each conversion was priced the same. 

That’s the promise and the risk of Meta’s bidding engine, which our co-founder and CPO, Eran Friendinger, calls a relentless genie. Ask for conversions, and you get conversions – a lot of them. Ask for customers, and you get customers.

So what does asking for customers look like in practice?

During a recent Meta Growth Accelerator, Daniel Benavides, Klar’s CRO, and Eran explained how Klar moved from conversion optimization to pLTV bidding. They covered the customer data that sat in Klar’s warehouse but never reached its campaigns, and why paying more per approval brought in 22% more revenue.

‍

Read on for the Q&A with Daniel and Eran from the session.

Q: What made you think there was a better way to grow than your existing setup?

‍

Daniel: It's a very interesting market, but it's also super competitive, and we're judged on how fast we grow and how efficiently we grow. In the end, the board is going to judge us on our revenue and our ROAS. At the time we were running all campaigns on target CPA, and our campaign goal was optimizing for credit card approvals. Credit card approvals are a proxy of what's going to happen, but they're not the reality of what's going to happen in terms of revenue. A client can get a credit card approved but never activated. 

Q: How do you think about lifetime value, and where did standard campaign setups fall short?

Daniel: A credit card approval is very different from a credit card activation, or a deposit in a debit account. The revenue really starts when we reach clients who are going to be credit approved, who really activate their card, stay with us longer and start purchasing.

‍

Our North Star right now is LTV at 120 days, because that's when clients usually start making their first purchases. The problem before was that with target CPA on credit approvals, all those approvals count the same.

What used to happen was that riskier clients, who tend to be more delinquent, were easier and faster for the campaigns to get. That doesn't translate into value for us. The clients we wanted were harder to get, more expensive to get, and sometimes harder to activate.

We had all this information on the client's transactions, credit scores, credit limits and revenue patterns. It was all living in our data warehouse, but it was not being translated to the campaigns or signals.

So that's what we worked on together with Voyantis: how to bring all these signals through a predictive LTV system in order to align the campaigns with our business goals.

Q: Eran, how did Voyantis help with the technical development?

Eran: I like to compare Meta's AI engine to a relentless genie. It literally grants your wishes. But when you make a wish, you need to be very particular about it, and you better be careful what you wish for. It will do whatever you ask it to do. So yes, we wanted to move from tCPA to value optimization, but it's not enough to move to value optimization. 

For example, we could have looked at the first transaction somebody makes on a credit card, before you even know if they will pay it back or how many more transactions they will make. That is a value. If that value is not well correlated to your long-term value per user, you're going to fall into the same trap as tCPA, or a similar one.

So the first step was identifying the window. If you were a good customer at 120 days, you were a good customer, period, for Klar. We identified the data points and the way to build a model, and that's what we sent over to the network.

LTV modeling was the first part, but a prediction isn't a signal yet. You also need to make sure you're delivering it to the network in the way that maximizes Meta's ability to act on it. That's a lot of trial and error, which we've done over a lot of customers over a lot of time, and we partnered deeply with Meta's product team to understand the nuances.

Q: What did the test show?

Eran: On a very clean 50/50 split, where one side was optimizing for approval and the other was optimizing on predicted 120-day LTV, the pLTV side returned 22% more revenue. Revenue per paying customer was up 87%. And the cost per approval rose. That's actually the good signal. As Daniel said, it's harder to get to those higher-value users. It's going to cost a little bit more, but you're going to make much more money, and that's what the business cares about.

‍

Q: What's the next growth challenge you're hoping to solve?

Daniel: There are two we're working on right now. The most important one is how we bring delinquency as a signal into the acquisition decision itself. This is a tricky one, because you usually see delinquency at least 45 days after the client is activated, and many times a lot later. That lag makes it really hard to predict and bring into the acquisition engine.

So we're working with Voyantis on how to understand those signals early, so we can adapt the pLTV model and the acquisition campaigns to also account for probable delinquency.

The second one is lifecycle marketing. Not all clients behave the same. About 70 to 75% of our clients activate during the first two or three days, and there's an important percentage that don't. Right now we do lifecycle marketing in a pretty standard way, sending the same promotions and activation efforts to all of our clients. 

We're working on bringing what we learned from the pLTV models into lifecycle marketing, to offer specific campaigns and offers depending on the value of the client, and not spend money on clients we already know are going to activate. It's just a matter of time for them, versus those who really need that extra push.

Give Meta a Target Worth Chasing

Klar’s approvals campaign was doing its job. It found approvals quickly and cheaply, but the easiest approvals to get were often the riskiest customers, the ones more likely to fall behind on payments.

Changing the bid strategy alone wouldn’t have fixed that. 

Choosing a window that matched how Klar makes money (120 days), predicting it early enough for Meta to use, and keeping that signal current as Klar’s risk models and approvals change made the difference. 

If you're bidding against a long payback window and want to see what user-level predictive signals could do for your campaigns, talk to our team, or get the playbooks from brands that have already made this shift to pLTV bidding, such as Semrush, Miro, and ZipRecruiter.