Why Your Credit Card Rewards Are Worth Less Than the Calculator Says

Run an average US household's spending through a rewards optimizer and you get a clean answer. Three cards, $34,800 of spending, about $1,384 a year in net value. Every step of that is reproducible in the tool.

You will not receive $1,384.

Not because the arithmetic is wrong. The arithmetic is the easy part, and it has been solved for years. What has not been solved is everything between the arithmetic and your bank account: whether you reach for the right card, whether the merchant codes the way you assumed, whether you remember the credits, whether the program still works the way it did in January.

See what your own wallet actually keeps →

This post puts a number on each of those gaps.

What is actually solved

Given a fixed set of cards, a known spending profile, published earn rates, and a point valuation, picking the best card for each purchase is a small optimization problem. Picking the best portfolio is a slightly larger one, and CardSavvy solves it with a mixed-integer program that assigns every spending category to exactly one card and models shared reward caps.

Both which card for this purchase and which cards to hold are computable in well under a second on a laptop.

The problem is that the optimizer never observes the world it is optimizing against. It sees your estimate of future spending, today's published rules, and an assumed point value. Every one of those inputs can be wrong by the time the year ends.

Gap 1: Interest, which can make the whole thing negative

This one is a gate, not a caveat.

The CFPB's most recent Consumer Credit Card Market Report, published in December 2025 and describing 2024, puts the average APR on general-purpose cards at 25.2%. Around half of accounts revolved a balance. About 15% of general-purpose cardholders paid only the minimum. Consumers were assessed $160 billion in interest that year, plus $31.3 billion in fees, against $47.5 billion in rewards earned.

Do the arithmetic on a single purchase. Carry $1,000 for one month at 25.2% and you pay roughly $21. A 2% card earns $20 on that same $1,000.

One month of interest wipes out the reward, and the interest keeps running after the reward stops. If you carry a balance, the correct move is to stop optimizing rewards entirely and pay the balance down. Any honest model checks this first and refuses to give a reward recommendation until it clears.

Gap 2: You will not always reach for the right card

No published source measures how often people use the wrong card for a bonus category. We looked. That number does not exist, and we are not going to invent one.

What does exist is adjacent evidence that people fail to optimize across cards even when the stakes are obvious. Gathergood, Mahoney, Stewart and Weber studied 1.4 million UK cardholders with multiple cards and found they match repayment shares to balance shares, sending money to each card in proportion to what it holds rather than to whichever charges the most interest. The heuristic costs the average two-card holder about £65 a year. Crucially, the misallocation held steady as the stakes grew, which rules out the "people are optimizing quietly and we cannot see it" explanation.

Reward routing is a harder problem than repayment routing, done more often, with less feedback. There is no reason to expect better compliance.

The honest way to model this is to ask you. Count your last ten bonus-category purchases and see how many landed on the wrong card. That is the input the calculator below asks for, because it is the only version of this number anybody can defend.

Gap 3: The merchant does not code the way you think

You think in terms of what you bought. Your card issuer pays based on how the transaction was coded, and those are different things.

The merchant category code is assigned by the merchant's acquiring bank or payment processor when the merchant is onboarded, following Visa and Mastercard classification rules. You have no input. Your issuer has no input. The merchant cannot change it unilaterally.

Chase says so plainly in its rewards category FAQ: "Even though a merchant or some of the items that it sells may appear to fit within a rewards category, the merchant may not have a merchant code in that category. When this occurs, purchases with that merchant won't qualify for rewards offers in that category."

American Express warns that a restaurant inside a hotel may be recognized as a hotel purchase, and that transactions submitted through a third party or a mobile card reader may not earn category bonuses at all. Capital One's own card pages go furthest: "Please note, Capital One is not responsible for codes used by merchants."

Three issuers, three different accounts of who assigns the code. Capital One tells consumers the merchant does it, which contradicts Visa's rulebook. If the issuers cannot describe the mechanism consistently, a category-level spending model is an approximation, and it is worth knowing which direction it errs.

There is a second-order problem too. A restaurant that switches point-of-sale vendors can start coding differently with no announcement and no change you would notice.

Gap 4: A credit is not cash

A $20 monthly credit is worth $20 only if you would have made that exact purchase, at that merchant, at that price, without the credit.

CardSavvy publishes what it thinks the realistic figure is. Across ten premium cards carrying $9,960 of marketed credits, the credit tracker estimates a realistic optimizer captures about $4,418, or 44%. The rubric is published alongside it: broad annual credits with no activation score 80% to 100%, brand-specific monthly credits score 40% to 70%, and niche subscriptions requiring an existing service score 10% to 30%, with deductions for use-it-or-lose-it months, portal-only booking, and per-period activation.

Two things to be clear about. That 44% is an assumption set built credit by credit under a stated rubric, verified as of May 2026. It is not a measured breakage rate, and nobody publishes one of those. And it varies enormously by card: Amex Gold scores 59% on its $424 of credits, while the Amex Platinum scores 38% on $3,114.

Value a credit at the lower of its face value and the spending it naturally displaces.

Gap 5: Points are not dollars until you redeem them

Cash back is easy to model. Transferable points are a claim on a future redemption at an unknown rate.

The CFPB's 2024 data shows consumers accruing rewards faster than they redeemed them, with more than $190 sitting in the average rewards account. In each quarter of 2024, about 2.8% of general-purpose cardholders forfeited some portion of their rewards balance. Read that figure carefully: it counts the share of people who lost something, and says nothing about how much they lost. Among cardholders with subprime credit scores, the rate has run more than double the overall figure since late 2021.

The gap that matters here is between the value you earned against and the value you take. Earn against 2.0 cents assuming transfer partners, then book the travel portal at 1.5 cents, and a quarter of the reward is gone before you notice. The points calculator uses conservative benchmarks for exactly this reason, and our post on valuing points walks through the floor-versus-upside distinction.

Gap 6: The rules change under you

This is the gap nobody prices in, and it is measurable.

Here is what changed across major programs in the thirteen months to July 2026.

Roughly one material change every eight weeks, across every major issuer and two hotel programs.

Note the Marriott entry. That devaluation shipped with no announcement at all and was discovered by third parties comparing prices. "Read the terms" is not a defense against a change that was never published.

Note the Sapphire Preferred entry too. It is usually described as a refresh that added categories. It also cut Hyatt transfer value and killed the anniversary bonus, on staggered dates for new and existing cardholders. Card changes are rarely purely additive, and the parts that get the press release are rarely the parts that cost you.

The CFPB ran into this problem itself. It launched an interactive credit card comparison tool in December 2024 and retired it the following year, stating that it "does not have timely source data to effectively support the purposes for which consumers might want a card comparison tool." The underlying survey still publishes twice a year. The consumer-facing interface to it does not.

An optimizer is partly a model and partly a continuously maintained knowledge base. The model can be perfectly correct against stale inputs and still give you the wrong answer.

Gap 7: The reward may change what you spend

This one gets overstated constantly, so here is the careful version.

A 2026 study by Agarwal, Ang, Wang and Zhang in the Journal of Banking & Finance examined cardholders at a large U.S. financial institution who enrolled in a 1% cash-back program. Enrollees raised spending on that card by 32% and debt by 8%, with larger responses among less financially literate and more liquidity-constrained cardholders. Both numbers belong in the same sentence: a 32% jump in volume on one card is compatible with a much smaller change in actual consumption, and the 8% debt figure is the one that costs money.

The broader claim that paying by card makes people spend more is weaker than the internet suggests. A 2024 meta-analysis by Schomburgk, Belli and Hoffmann pooled 392 effect sizes from 71 papers and found the cashless spending effect real but small, with a mean Hedges' g of 0.135, enormous heterogeneity across studies, and a statistically significant decline over time.

Rewards can change what you spend, for some people, in some settings, by an amount nobody can pin down for you specifically. Small as that sounds, it swamps the optimization. Moving $10,000 from a 2% card to a 3% card gains $100. Spending 1% more because a purchase feels cheaper costs $100.

Adding it up

Take the published $34,800 wallet. Amex Gold at a $325 annual fee carrying $424 of credits, an optimized earn rate of 4.17%, and a flat 2% card as the fallback.

Assume two of every ten bonus purchases land on the wrong card, credits get captured at the published 44%, and points are redeemed the way they were valued.

The gap is $388, or about 21% of the gross value on paper. You keep roughly four dollars in five.

Two things worth noticing. The realized $1,162 still beats the $696 a flat 2% card earns, so optimizing paid for itself here. And CardSavvy's own optimizer reports $1,384 for this wallet, which sits between the two columns, because it already discounts credits before reporting a net value. Treat the left column as raw face-value arithmetic, deliberately undiscounted so the leakage below it has something to bite into.

Run your own numbers.

What to do about it

Pay in full, every month. Everything above is irrelevant until this is true.

Benchmark against a flat-rate card, not against zero. A card that generates $400 of rewards is worth $200 if your existing 2% card would have generated $200 on the same spending. The number that matters is what a card adds over the next-best card you already hold, which is also the number that tells you what to cut.

Value credits at what they displace. Not at face value, and not at the number in the marketing copy.

Use your own redemption history, not a travel blogger's best case. If your last four redemptions averaged 1.3 cents, plan at 1.3 cents.

Stop adding cards when the incremental value stops covering the incremental ways to leak. We measured this ladder on an average budget: the second card added $640, the third added $48, and the fifth destroyed $417. Every card after the second buys you less value and more chances to reach for the wrong one.

Who this is for, and who it is not

Optimization pays for people who clear the statement every month, have a cash buffer, hold enough card-eligible spending for percentages to matter, and will follow a routing rule once they set one. A simple two-card setup captures most of the available value for most households, and it is the right recommendation far more often than the enthusiast forums suggest.

It does not pay for anyone carrying a balance, paying down high-interest debt, sitting on unstable near-term liquidity, or preparing for a mortgage application where new accounts are unhelpful. It also does not pay for people who find the tracking stressful. Stress is a real cost, and a system you abandon in March returns nothing.

Execution decides this, and plenty of financially sophisticated people execute badly. The only question that settles it is whether the system leaves you better off in dollars.

What we cannot know

No optimizer knows whether you will be approved, whether a targeted offer will appear, when the next devaluation lands, how a given merchant will code next year, or what you would have spent in a world without the reward.

And one more, which is worth saying because our competitors do not: there is no peer-reviewed research estimating whether multi-card reward optimization is worth the effort. We searched hard for it. The literature covers whether rewards change payment choice, whether card payment changes spending, and whether people misallocate repayment. Nothing measures the return to the optimization itself. Every "$500 to $1,500 a year from optimizing" figure in circulation traces to blog or issuer content, including, until this post, some of ours.

We also publish where our own optimizer is wrong, because a model you cannot audit is a model you should not trust.

Bottom Line

The optimization is solved. The realization is not.

Your inputs are estimates, the rules move under you, merchants code in ways nobody tells you, credits expire quietly, points are worth what you take rather than what you assumed, and you will occasionally reach for the wrong card. None of that makes rewards a bad deal. It makes the advertised number the wrong number to plan against.

Optimize for what you keep. Then stop, because the returns run out faster than the complexity does.

Once you have captured the rewards, the harder question is what those dollars should do. Summitward is our sister site, built by the same team, for net worth, financial independence, portfolio risk, and taxes. Rewards are the smallest lever in your financial life, and we have written about where they rank against everything else.

Frequently Asked Questions

Why do I earn fewer points than the calculator said I would?

Four things sit between the two numbers. Some purchases go on the wrong card or code outside the category you expected. Statement credits go unused. Points get redeemed below the value they were earned against. And card rules change mid-year. On a typical three-card wallet those gaps run 15% to 25% of the value shown on paper.

What is the difference between theoretical and actual credit card rewards?

Theoretical rewards are what your wallet earns with perfect card selection, full credit usage, and redemptions at the value you assumed. Actual rewards are what survives after real execution. On a $34,800 wallet earning $1,550 on paper, a reader who misses two bonus purchases in ten and captures 44% of their credits keeps about $1,162.

How much of my statement credits will I actually use?

CardSavvy publishes a capture rate of about 44% across ten premium cards, covering $9,960 of marketed credits. That figure is an assumption set built from a stated rubric, credit by credit, not a measured breakage rate. Broad annual credits with no activation score 80% to 100%; narrow monthly ones score far lower.

Do credit card rewards make people spend more?

Sometimes, and the size of the effect is contested. A 2026 study of one U.S. institution found cardholders who enrolled in a cash-back program raised spending on that card 32% and debt 8%, though enrollees chose to join. A 2024 meta-analysis of 71 papers found the general cashless spending effect small (Hedges' g = 0.135), highly variable, and weakening over time.

Is optimizing credit card rewards worth the effort?

No peer-reviewed research answers this. Every dollar figure circulating online traces to blog or issuer content rather than a study. What the arithmetic does support: the first two cards capture most of the available value, and each card after that adds less while adding more ways to leak.

Who should not bother optimizing credit card rewards?

Anyone who carries a balance. At the 25.2% average APR the CFPB reported for 2024, one month of interest on $1,000 costs about $21, which already exceeds the $20 a 2% card earns on the same purchase. Rewards optimization should be switched off, not merely discouraged, until the balance is gone.