We stopped a Google Ads campaign from dying over its CPA

The mistake of setting target CPA on the first sale alone

Almost every Google Ads account we review starts in the same place, and I think that is fine: target CPA gets set off the margin on the first sale. That is, if an average €120 order leaves you €40 clean, you do not pay more than €40 per conversion. So far so good, an argument like that is easy to defend and it is "logical".

The problem with that reasoning is that the €40 you have set may differ from what the market says. If your competitor knows the same customer is worth €180 over twelve months, they can pay much more per conversion and still make money. Voilà! They are not beating you on creative or on the famous Quality Score, they are doing it because they are playing with different numbers.

The 90-day wall: why full LTV does not work for bidding

Making the natural jump to bidding on full lifetime value (LTV) sounds logical, but it runs into an instrument failure. An LTV is an aggregate built on retention, repeat-purchase and margin assumptions over that user's ENTIRE lifetime. And of course, some of our customers have been with us for 7 years. Those same customers were acquired at different prices and against different competitors. A bidding algorithm cannot learn from a figure like that.

There is also a hard limit on the platform side. The Google Ads click-through conversion window maxes out at 90 days, so revenue that lands in month 8 can no longer be imported against the click that produced it. You can have a beautiful LTV in a spreadsheet: bidding will optimize on whatever fits inside its attribution window, not on your full projection for that user.

The three rules for measuring repeat-purchase cohorts honestly

Against theoretical estimates, what does work is to stop estimating and start measuring real cohorts under three conditions almost nobody meets:

1. The window runs from each customer's first purchase, not from today. A customer acquired a year ago has had twelve months to spend; one acquired three weeks ago has had three weeks. Averaging them together does not measure retention, it measures age.
2. Only closed cohorts count. For a 90-day multiplier, only customers whose first purchase happened more than 90 days ago qualify. Throw in last week's and you drag the average down and understate yourself. The sample size changes with the window, and it should: that is the sign the measurement is honest.
3. The multiplier is computed on margin, not on revenue. It is not what the customer bills, it is what is left after VAT and the variable costs of each purchase (e.g. shipping costs, etc.).

With those three rules in place, this is what came out of a real account with 1,870 customers acquired through advertising:

The first purchase leaves €15 of margin. That is the ceiling target CPA gets set against. At 90 days, with money already collected and closed cohorts, that same customer has left €47.65. It is not a projection: it is collected revenue, from customers who have already had those 90 days to come back.

From 3.17x to 6.33x: the real value that stays out of the auction

Looking at how this cohort behaves, the jump from 90 days to all time goes from 3.17x to 6.33x, revealing exactly what the LTV debate usually skips. That customer ends up worth €95.87 in margin, but half of that value lands after the 90-day window and never enters any auction, ever. It is not a problem with your measurement or your integration: it is the platform's technical limit.

My takeaway on the real multiplier for your Google Ads CPA

So the operational question is not "how much is my customer worth?", it is "how much is my customer worth inside the horizon where Google still knows which click that money came from?". And here the answer is 3.17x. The rest sits in your books, it justifies the business and pays the salaries, but it is not a bidding input.