Crypto card cashback after fees: findings from a 63-program audit

CARDEFI examined the first listed variant of each of 63 programs marked live in its compiled dataset of 12 September 2026. Of those records, 28 contained a verified positive reward rate on a tier labelled as having no requirement. Thirteen of the 28 had an undisclosed cap in the structured data. Across all 63 records, 33 had an undisclosed or missing conversion-spread field.

Those are findings about the coverage of this dataset. They do not mean 33 providers hide a fee, or that all 28 reward offers are available to every applicant. They explain why a precise-looking annual cashback figure can depend on assumptions the reader needs to see.

What the sample measures

The compiled file contained 68 programs: 63 live, four closed and one on a waitlist. We excluded the latter five and used the first listed variant of each remaining program, giving every program one observation. Regional variants were not counted separately.

This is a catalog audit, not a representative survey of every crypto card. It includes different countries, funding models and reward types. Some recorded facts predate the compilation date. We checked selected official examples again on 22 September 2026; we did not reverify the entire catalog on that date.

The project makes its compiled data available through CARDEFI’s data page. The counts below refer to the frozen September 12 extract, rather than whichever version is downloadable later.

“Verified” describes the stored field’s evidence status, not a guarantee that its value remains current. “Undisclosed” describes an unresolved value in the collected material. Additional official documentation may resolve it.

The conversion field had the largest combined count of undisclosed and missing entries among these four fields: 33 of 63, or 52.4%. A missing value cannot establish a zero-cost conversion.

A reward rate can be known while its ceiling is not

The 28-record reward subset was selected mechanically: the first variant needed at least one tier with a positive verified rate and a requirement type of none. Where several tiers qualified, we used the first one. This label does not rule out conditions described elsewhere, such as account eligibility, a particular funding asset or merchant restrictions.

Within that subset:

• 11 records had a verified cap with an explicit period.
• 13 had an undisclosed cap.
• 4 recorded the cap as not applicable.

The 13 unresolved records make up 46.4% of the subset. They require further source work before an uncapped annual projection can be treated as established terms.

There is a concrete example in KAST’s membership agreement. Its Standard clause gives a reward allowance without stating a reset period in that clause. A monthly allowance and a one-time allowance would produce different annual results; the time unit must be confirmed rather than inferred.

This is also why the audit retains lower-confidence records for coverage counts. Excluding incomplete entries would make the catalog look more complete. We do not use those counts to rank the cards by annual returns.

The purchase currency does not describe every conversion

A purchase in the card’s settlement currency may still involve selling the asset used to fund it.

The OKX EEA fee table, checked on September 22, lists no added FX fee but a 0.1% spread for conversion from USDC or USDG to euros. Those are different cost categories. Treating “no FX fee” as “no conversion cost” would omit the latter.

Its cashback FAQ gives non-VIP users a 2% rate on eligible USDG-funded purchases, subject to a normal €50 calendar-month cap. This reward condition is narrower than simply being able to spend a supported stablecoin.

The sequence matters: identify the funding asset, establish which purchases earn rewards, then apply each charge to the amount it actually affects. A crypto-to-fiat spread should not disappear from a calculation just because the merchant charged in the card’s settlement currency.

Reward balances also need a unit

Another example is Bitrefill. Its cashback guide describes rewards for both cryptocurrency deposits and card spending, redeemable as Bitrefill store credit. Its separate funding-fee page lists a fixed 1.99% charge plus a variable exchange cost.

A store-credit balance can be useful to someone who would otherwise make purchases there. Its practical value is different for someone who needs money for a bank transfer. Adding reward percentages together and subtracting a fee percentage would conceal both the different calculation bases and the different uses of the reward.

For a comparison, show the reward’s form next to its rate. Any decision to value a token, point or restricted credit as cash is an assumption, and should be labelled.

How a spending pattern changes the result

To isolate the arithmetic, we ran a hypothetical card through CARDEFI’s calculation core, version 1.2. This is a controlled example, not an observed card offer or the average of the 63 programs.

Assume $12,000 of annual eligible purchases, 2% cashback with no cap, a 1% FX fee, no conversion spread, no stake and no other costs. Vary only the share of purchases incurring FX, then add a fixed fee in the final scenario.

Nothing happened to the advertised 2%. The annual outcome nevertheless moved from $210 to zero as the spending pattern and fixed charge changed.

These examples explicitly set all other costs to zero. A real calculation may need funding fees, network charges, ATM use, issuance or delivery, interest and transaction-specific exclusions. Rewards may also change in value or stop when a promotion ends. The model is only as complete as the costs and conditions supplied to it.

Readers can inspect CARDEFI’s methodology before using its estimates. In particular, a default assumption is not an issuer-confirmed price.

What belongs beside the headline rate

This audit points to three useful additions to a cashback comparison: the reward’s eligible spending base, its cap and reset period, and the charges along the funding-to-payment route. The reward’s redemption form belongs there too.

The immediate research task for this sample is to resolve the 13 cap records and 33 conversion-spread records before making stronger claims from them. Until then, a bounded calculation with visible unknowns gives the reader more to work with than an annual return presented to the nearest dollar.