Google Ads and Meta do not know what your customer is really worth. They know what you tell them. Most stores send one number: the order value.
The platforms optimize toward ROAS calculated from the value you pass in. Feed them transaction revenue and the algorithm learns to hunt for the next high-value order, whether or not there is a valuable, repeat customer behind it. Budget flows toward high immediate revenue, which is not always the same place as high value over time.
A one-time buyer and a subscriber look identical
At the moment of purchase, a customer who buys once for PLN 150 and a customer who starts a subscription with a PLN 150 order are the same conversion, the same value, the same signal to the platform.
Their real worth can differ several times over. For some products we see a subscription customer generate roughly 2x CAC in margin after 90 days and 3x CAC after six months. The algorithm cannot see that, so it leaves it out of the optimization. The budget goes where the immediate revenue is, and misses where the long-term value is.
The data that separates one customer from another is usually already in your first-party data: LTV, subscription status, the historical value of the products in the first basket. The catch is that it lives in the reporting layer and never reaches the ad platform’s algorithm at the moment it could act on it.
Methodology note: the 2x and 3x CAC figures are directional benchmarks observed across selected subscription products, not a guarantee. Your own ratios depend on margin, retention and product mix.
What BFA does in Sublime Analytics
BFA (Backend-for-Analytics) is a data-modification layer that turns business data from Sublime Analytics into an adjusted signal sent to the ad systems. It has one job: attach the customer’s real value to the purchase event before the algorithm makes its next decisions.
Instead of sending the order amount alone, BFA recognizes who the customer is and what they bought, pulls the values you already calculated (LTV, CAC, subscription status), and returns a conversion value that reflects that customer’s potential rather than a single purchase.
A few examples of how this changes the signal:
Subscriber vs one-time buyer. Two orders of PLN 150. One is a single purchase, the other opens a subscription historically worth PLN 700 after six months. Sent as identical PLN 150 values, they tell the algorithm these are the same people. BFA raises the subscriber’s value, so the algorithm starts looking for people who behave like them.
A cheap high-LTV product vs an expensive one-time buy. One customer spends PLN 100 on an entry product that historically leads to repeat orders and high LTV. Another spends PLN 300 once on a heavily discounted product and never returns. On revenue alone the algorithm prefers the second. On value over time the first is worth more. BFA flips that signal and stops overpaying for expensive but one-off purchases.
Damping weak segments. This works both ways. A segment with a high return rate or low repeat behavior can be signaled below its nominal purchase value, so the algorithm stops spending budget on customers who look good at checkout and poor three months later.
You change the goal, not the campaign
Most of the levers marketers use are indirect: budget, creative, audiences, bidding strategy. All of them still chase the same target, revenue at the moment of purchase. That target is almost always fixed, and almost no one touches it.
The value you send is the target the algorithm learns toward. Change it and you change the definition of success for the whole system, without touching a single creative. It is one of the few ways to genuinely influence which customers the algorithm treats as good in the first place.
Few teams do it, because it takes three things at once: reliable data on customer value, the ability to calculate it, and a way to inject it into the platform at the moment of conversion. Most brands have the data, but it stays locked in reports, and the loop never closes.
The potential is easy to picture: the same budget and the same campaigns, but the algorithm starts hunting for your best customers and stops chasing your weakest.
Two signals, two roles
There is a line worth keeping. BFA does not overwrite the transaction value. The standard purchase and the reported revenue stay untouched, and your attribution reports keep showing what actually happened.
The enriched value runs on a separate track: as a dedicated conversion meant for optimization, not in place of real revenue. That way you teach the algorithm a new signal without corrupting the data you base your own decisions on. When reliable data is missing for a given customer, the signal falls back to the normal purchase value rather than a guess.
The standard question is: “how do we raise ROAS on the first order?” That is a question about a single transaction.
The better question is: “are we teaching the algorithm to acquire our most valuable customers?” BFA does not change how many customers you have. It changes which ones the algorithm learns to look for.
If you want your campaigns to optimize for customer value over time, book a demo. We will map the BFA mechanism to your business model and point out the data needed to put it in place.
