Where the Margin Actually Hides

Key takeaways

• AutoNation's 2025 filing prices the point exactly: $2,564 of gross profit on an average new vehicle that sold for $52,075, and $2,769 of gross profit on the finance and insurance products attached to the same transaction.
• Costco's merchandise operation cleared $5.06 billion after SG&A in fiscal 2025; membership fees added $5.32 billion, so more than half of the $10.38 billion operating income came from a fee rather than from selling anything.
• Starbucks recognised $222.4 million of stored-value and loyalty breakage in fiscal 2025 (money customers loaded onto cards and is not expected to be redeemed) while holding a $1.75 billion customer-funded balance at year end.
• Delta's 2025 remuneration from American Express was $8.2 billion, roughly 1.4 times the airline's entire $5.82 billion of operating income for the year.

The blended margin is a lie of averages

A company-level gross margin is a weighted average of businesses that often have nothing in common. Report it as one number and you hide the only fact that matters: which revenue line is carrying the others. Almost every ordinary business is quietly a portfolio, and almost nobody looks at it that way.

AutoNation's 2025 Form 10-K breaks the car business apart line by line, and the result is worth sitting with. Across 528,822 retail vehicles sold, the average new vehicle brought in $52,075 of revenue and produced $2,564 of gross profit, or 4.9 cents on the dollar. The average used vehicle brought in $26,967 and produced $1,555. Finance and insurance products sold alongside those same transactions produced $2,769 per vehicle retailed. The paperwork is worth more than the car.

Run that to the company level and the shape gets starker. Finance and insurance was $1,464.4 million of revenue, net, which is 5.3% of the company's $27.63 billion top line, and $1,464.4 million of gross profit, because that line is reported net of its own cost. That is 29.6% of the company's $4.95 billion of total gross profit. It also exceeds the company's entire $1,239.9 million of operating income for the year. Parts and service, the unglamorous back of the building, turned $4,835.4 million of revenue into $2,355.1 million of gross profit, a 48.7% margin, roughly ten times the margin on a new car.

So the business a customer thinks they are transacting with, a place that sells cars, is not the business that generates the profit. The dealership is a customer-acquisition machine for two much better businesses attached to it. Notice too what the same table reveals about fragility: new-vehicle gross profit per unit fell from $4,342 in 2023 to $3,045 in 2024 to $2,564 in 2025, a 41% decline in two years, while finance and insurance held between $2,612 and $2,769 across the same period. The high-margin attachment was also the stable one.

The operator's move is mechanical and most people have never done it: take last year's revenue, split it by line, and compute gross margin on each line separately, allocating only the costs that genuinely vary with that line. You will usually find one line doing most of the work, one line roughly break-even, and one line you have been proud of that is worse than it looks. The risk in doing this badly is allocation games: pushing shared overhead around until every line looks acceptable. Allocate only what is directly traceable, leave the rest unallocated at the bottom, and read the result honestly.

When the fee is the actual business

Some businesses sell goods at close to cost and make their money on the right to buy those goods. Once you have seen the arithmetic, you cannot unsee it, and it explains behaviour that otherwise looks irrational.

Costco's fiscal 2025 results, for the 52 weeks ended 31 August 2025, are the clearest published version. Net sales were $269,912 million and merchandise costs were $239,886 million, so merchandise gross profit was $30,026 million, or 11.1% of net sales. Selling, general and administrative expense was $24,966 million. Subtract one from the other and the entire merchandise operation, on $269.9 billion of sales, cleared $5,060 million: an operating margin of 1.9 cents on the dollar.

Membership fees were $5,323 million. Add them and operating income is $10,383 million. A fee with essentially no cost of goods against it supplied 51.3% of the company's operating profit. The merchandise is not the profit centre; it is the reason the fee is worth paying.

That structure changes every downstream decision, which is why the same filing contains a sentence most retailers could never write: that the company may hold prices steady despite cost increases instead of passing them on to members, knowingly damaging gross margin in the near term. A merchant whose profit came from merchandise margin cannot do that. A merchant whose profit comes from renewals can, because the price restraint is what buys the renewal. Renewal rates were 92.3% in the US and Canada and 89.8% worldwide at the end of 2025.

The fee also arrives before the service is delivered. Membership revenue is deferred and recognised ratably over the one-year membership period, and deferred membership fees stood at $2,854 million at the end of fiscal 2025, customer money on the balance sheet, funding the business, at no interest.

The lever for an ordinary operator is not to bolt a membership onto anything. It is to ask which of your customers would rationally pay for the right to keep buying from you, and what you would have to give up on unit margin to make that trade genuinely good for them. The failure mode is charging the fee and keeping the margin too: customers do the arithmetic faster than operators expect, renewal falls, and you have converted a stable revenue line into a churn problem.

Money you were paid for something you will never deliver

Breakage is revenue from an obligation that will never be claimed. Gift cards nobody spends, loyalty points nobody redeems, credits nobody uses. It carries no cost of goods, no labour and no delivery, which makes it the highest-margin revenue in almost any business that has it, and the most quietly estimated.

Starbucks discloses the mechanics precisely. In its company-operated markets, including the US, stored value cards do not expire and no service fee erodes the balance. Based on historical redemption rates, a portion of loaded value is not expected to be redeemed, and that portion is recognised as breakage over time in proportion to actual redemptions. For fiscal 2025, ended 28 September 2025, the company recognised $200.4 million of breakage in company-operated store revenues and $22.0 million in licensed store revenues: $222.4 million in total, against $37,184.4 million of net revenues and $2,936.6 million of operating income.

The honest way to read that ratio is with the denominator's history attached. In fiscal 2024 breakage was $187.6 million plus $20.0 million, or $207.6 million, against $5,408.8 million of operating income, or 3.8%. In fiscal 2025 the same line was 7.6% of operating income, but almost all of that move came from operating income falling, not breakage rising. Operating margin dropped from 15.0% to 7.9% on restructuring. Breakage grew 7.1%. A ratio that improves because the denominator collapsed is not an improvement, and this is exactly the kind of number that gets quoted without its context.

Underneath the income statement sits the larger prize. The stored value card and loyalty program liability was $1,751.7 million at the end of fiscal 2025, up from $1,718.7 million. During the year $15,245.8 million was deferred as customers activated cards, reloaded them and earned Stars, and $15,199.5 million was recognised as those balances were redeemed or broke. The balance barely moved, which is the point: every individual dollar is temporary, and the pool is permanent. That is a customer-funded, interest-free balance roughly the size of a mid-cap company's entire market value, sitting on the books because people load cards before they drink coffee.

For a small operator the same physics is available at small scale (prepaid packages, punch cards, deposits, unredeemed credits) and the same three cautions apply. First, the money is an obligation, not income, until it is either delivered or genuinely broken. Second, breakage is an estimate built on your own redemption history, so a change in customer behaviour revises prior revenue, not just future revenue. Third, unclaimed property law is real: Starbucks names remittance to government agencies under those laws as an input to its own redemption rates, and in many US states unredeemed balances escheat to the state rather than to you. Design the programme with your state's rules in front of you, not after.

Selling the customer instead of the product

The highest-margin thing some businesses own is not their product. It is the fact that a large number of people have a reason to keep interacting with them, which a third party will pay for directly.

Delta's 2025 Form 10-K states it plainly: remuneration from American Express under the co-brand credit card relationship totalled $8.2 billion in 2025, an increase of approximately 11% over 2024, and the company expects it to reach $10 billion over the next few years. Set that against the airline's own income statement for the same year, total operating revenue of $63,364 million and operating income of $5,822 million, and the co-brand payment stream is about 1.4 times the entire operating profit of the airline. Separately, the filing reports total cash sales from marketing agreements related to the loyalty programme of $8.0 billion in 2025, $7.4 billion in 2024 and $6.9 billion in 2023; that is a different measure covering the sale of miles itself, allocated between travel and other performance obligations, and it should not be added to the remuneration figure.

What the credit card company is buying is not seats. It is a reason for a consumer to put ordinary spending (groceries, fuel, everything) on one specific card. Delta manufactures that reason at a cost it controls, because the miles it sells are redeemed against seats that would frequently have flown empty. In 2025, 12% of revenue miles flown on Delta were award travel, across approximately 35 million award tickets.

The balance sheet shows how much of this business is customer-funded. Air traffic liability, meaning tickets sold and travel not yet taken, was $7,157 million at the end of 2025. Loyalty programme deferred revenue was $4,876 million current and $4,386 million noncurrent, $9,262 million in total. Together, $16.4 billion of obligations funded by customers and partners, against $8,342 million of cash generated by operating activities in the year.

The transferable idea is not "launch a loyalty programme." It is that if you have recurring attention from a defined group of people, someone whose economics are better than yours may pay you more for access to that attention than you can earn selling to it yourself. A gym with 2,000 members, a trade newsletter with 40,000 readers, an equipment dealer who sees every contractor in a county: each has an asset that a financing partner, an insurer or a supplier may value more highly than the operator does.

The risk is the mirror image of the reward, and it is concentration. A payment stream larger than your operating income, from one counterparty, on a contract with a renewal date, is a single point of failure dressed as a profit centre. Know the renewal date. Know what the business looks like at 70% of that payment. And understand that the partner is buying the quality of your relationship with your customers, which means anything that degrades that relationship to squeeze the partner's payment eventually reduces it.

Pricing power is the only lever with no cost attached

Every other route to profit costs something. More volume costs acquisition spend and capacity. Lower input costs cost negotiating leverage or quality. Efficiency costs management attention and usually capital. A price increase that customers accept costs nothing at all, which is why its arithmetic is so lopsided, and why so few operators run the arithmetic.

Illustrative only, with round numbers chosen for legibility. A services business does $1,200,000 of revenue at a 62% gross margin, so direct costs are $456,000 and gross profit is $744,000. Fixed operating costs are $620,000. Operating profit is $124,000, or 10.3% of revenue.

Raise prices 3% and hold every unit. Revenue becomes $1,236,000. Direct costs do not move, because you are delivering the same work: still $456,000. Gross profit becomes $780,000 and operating profit becomes $160,000. A 3% price move produced a 29.0% increase in operating profit, because the entire $36,000 landed on a $124,000 base with nothing subtracted from it.

Now the number that decides whether to do it. At the 3% higher price, gross profit per unit of work is higher, so you can afford to lose volume. Set the new gross profit equal to the old: you can lose 4.62% of your volume and be exactly where you started. Anything less than that and the price increase wins. Run the same machinery in reverse and the discounting trap appears: cut price 3% and gross profit falls to $708,000, so you need volume up 5.08% just to return to $744,000. The asymmetry between 4.6% and 5.1% is small in isolation and enormous in habit, because most businesses discount reflexively and raise prices reluctantly.

The levers that make a price increase survivable are unglamorous: raise on new customers first and let the existing base roll over at renewal; change the package alongside the price so the comparison is not like-for-like; move the smallest, least price-sensitive segment first and watch what actually happens rather than what you feared; and never announce an increase without also naming what got better.

The risk worth naming is that pricing power is a property of your position, not of your courage. Costco's filing describes deliberately not passing cost increases through to members, a company with enormous scale choosing to spend its pricing power on retention because its profit arrives as a fee. If your customers can switch in an afternoon, if your offer is undifferentiated, or if your buyer is a procurement department with three quotes on the desk, the 4.6% volume-loss cushion in the example above will be consumed instantly. Test small before you believe the arithmetic.

What breaks first

Hidden margin fails in specific, repeated ways, and each has a different early warning. Knowing which one you are exposed to is more useful than knowing the general principle.

Deferred balances get spent as if they were profit. Costco's $2,854 million of deferred membership fees and Starbucks' $1,751.7 million stored value liability are obligations that happen to be denominated in cash. Large companies hold them against a stable inflow. A small operator who books a year of prepaid retainers, spends it in four months and then has eight months of delivery obligation and no cash has run a private reserving crisis. The test is simple and most people avoid it: how much of your current cash balance is money you already owe in service?

Estimates get revised backwards. Breakage is calculated from your own redemption history. When behaviour changes, whether an app makes redemption easier, a loyalty rule shifts, or a state tightens escheat, the estimate changes and the correction hits reported earnings. Revenue built on an estimate of what customers will never do is real revenue and volatile revenue at the same time.

Attachment margin is the first thing a regulator or a competitor takes. High-margin add-ons attached to a low-margin core sit in exactly the place scrutiny lands: they are lucrative, poorly understood by buyers, and sold at the moment of least resistance. AutoNation's finance and insurance line held up while new-vehicle gross profit per unit fell 41% in two years, which is also the reason a business structured that way should assume the attachment is the part under pressure next, not the part that is safe.

Mix drifts without a decision being made. A business that quietly grows its lowest-margin line looks like it is growing. Costco's own filing notes that gasoline and e-commerce both carry lower gross-margin percentages than warehouse operations, so a higher share of either mechanically lowers the blended figure without anyone doing anything wrong. If you only ever look at total revenue and total margin, mix drift is invisible until it is structural. Track margin by line, monthly, and treat a falling blended margin with rising revenue as a mix question first and a pricing question second.

And concentration hides inside every one of these. A fee model concentrates on renewal. A partner-funded model concentrates on one contract. A breakage model concentrates on one behavioural assumption. The margin you cannot see is also the margin your customers, your partners and your regulators cannot see, right up until one of them looks.

Put it to work

Split last year's P&L by revenue line and compute gross margin on each. Rank them. Then find the three margin pools you are not yet running: cash you could collect before delivery, an attachment sold alongside the core product, and a price you have not raised. Model a 3% price increase against a 4.6% volume loss. Book the review quarterly.

Sources & references

Linked entries open the named source directly. Entries without a link say exactly what kind of reference they are — and how to check them yourself.

• AutoNation, Inc. — Form 10-K for fiscal year 2025 (gross profit per vehicle retailed, finance and insurance, parts and service)
• Costco Wholesale Corporation — Form 10-K for fiscal year 2025 (membership fees, merchandise costs, SG&A, operating income, renewal rates, deferred membership fees)
• Starbucks Corporation — Form 10-K for fiscal year 2025 (breakage revenue, stored value card and loyalty program liability, operating income)
• Delta Air Lines, Inc. — Form 10-K for fiscal year 2025 (American Express remuneration, loyalty deferred revenue, air traffic liability, award travel share)
• Worked pricing example is illustrative — The $1,200,000 revenue, 62% gross margin and $620,000 of fixed cost in the pricing section are made-up round numbers chosen to make the arithmetic legible. The 29.0% profit increase, the 4.62% break-even volume loss and the 5.08% volume gain needed to offset a 3% discount all follow from those figures and no others. Replace them with your own.