Why your approval rate is leaking eight points — and how to find them

Most operators I meet quote their approval rate as a single, almost mystical number. "We're at 87." As if that figure were a property of the universe rather than the cumulative result of dozens of decisions made by their PSP, their issuer, their 3DS provider, and themselves.

It almost never is. In a typical first audit I find between five and ten percentage points of approval that the merchant is leaking without realising it.

Start by segmenting

The single number is a trap. The interesting truth lives one level below it. Split your traffic by:

• Issuer country and BIN range. A 92% global rate can hide a 71% rate on a major market.
• Card scheme and product type. Premium credit cards behave very differently from prepaid.
• Channel. Recurring rebills, one-click checkout, and guest checkout each have their own physics.
• 3DS outcome. Frictionless, challenge, and bypassed all carry different decline patterns.

Once you segment, you usually find that the leak is concentrated. Two or three issuer-country pairs are responsible for most of the loss.

Read the response codes properly

The second leak is interpretive. Soft declines — "do not honour," "issuer unavailable," "insufficient funds" — are often retried badly or not at all. Hard declines are sometimes retried, which makes things worse.

Build a response-code map that classifies each code as:

• Retry now (transient network or issuer issue)
• Retry later (insufficient funds, velocity)
• Do not retry (stolen card, fraud)
• Route differently (PSP-specific issue)

Most PSPs ship a default that is wrong for your traffic. Override it.

Test the alternative path

The third leak is invisible until you create the comparison. If you only use one PSP, you have no counterfactual. Even a 5% test split through a second provider on the same traffic will reveal whether your incumbent is genuinely at the frontier or simply familiar.

What to expect

In the projects I have led, a structured approval-rate program typically delivers a 10 to 15 percentage point lift within two quarters, with most of the gain landing in the first eight weeks. The work is unglamorous: it is data, response-code maps, and weekly readouts. But it pays for itself many times over.

The first step costs nothing. Pull your last 90 days of attempts, segment them four ways, and look at the histogram. The leaks tend to be obvious once you stop staring at the average.