A discount is not a discount. It is a permanent downward revision to the price your customer believes your product is worth, and you sell it to yourself for one period's volume. You book the revenue and feel clever. You have also repriced every future transaction with that buyer, in the direction that costs you money, indefinitely. The sale is temporary. The anchor it sets is not.
This is the part the spreadsheet hides. The promotion model asks one question — did the volume lift cover the margin I gave up this month? — and answers it in isolation. The real transaction is intertemporal. You are not selling units at a lower price. You are selling a piece of your future pricing power at an exchange rate you never quoted yourself.
The buyer doesn't store a price. They store a reference point.
People do not evaluate a price against a product's intrinsic worth, because they mostly can't compute that. They evaluate it against a reference point — the last price they saw, the price they expected, the price the category has trained them to assume. This is the load-bearing finding of prospect theory: outcomes are coded as gains or losses relative to a reference, not as absolute states of the world. Change the reference and you change how the identical number feels.
A discount moves the reference point down and leaves it there. Before your sale, $100 was simply the price. After a $100 product sits at $80 for two weeks, $80 is the price the customer now believes is real, and $100 has been re-coded from fair into a loss — a $20 penalty for buying at the wrong time.
Then loss aversion does the disproportionate damage. Losses loom larger than equivalent gains by a factor Tversky and Kahneman estimated at roughly 2.25. The $20 you saved a customer during the sale registered as a modest gain; the same $20 gap felt at full price now registers as a loss weighted about 2.25 times as heavily. You bought a little goodwill on the way down and manufactured a lot of resistance on the way back up. You didn't discount the product. You revalued every subsequent full-price offer into a thing that stings.
The exchange rate is worse than it looks
Start with the part that isn't even behavioral — just contribution margin. Take a $100 product at 60% gross margin, so $40 of cost and $60 of contribution per unit. Run a 25% sale. New price $75, contribution now $35. To hold total contribution flat you need volume to rise by 60/35 − 1, about 71%. A quarter off your price demands a seventy-percent volume lift just to break even on the current period. Most sales don't clear that bar, which is why "the promo was a success" so often decodes to "revenue went up and profit went down."
Now stack the second-order cost the model never charged you for. Some fraction of your sale buyers would have paid full price and did — you handed them $25 each for free and taught them to wait next time. Another fraction merely pulled forward a purchase they'd have made next month anyway — you cannibalized your own future revenue and paid 25% for the privilege. The genuinely incremental buyers, the only ones the promotion was supposed to be for, are a slice of the total, and they are disproportionately the buyers who came for the price and will leave for a price.
So the true exchange rate is: give up 42% of per-unit contribution, demand a 71% volume lift you probably won't hit, subsidize the customers who'd have paid anyway, pull revenue backward in time, and — the item with no line on the invoice — lower the reference price for the whole market. That last one doesn't clear at the end of the quarter. It compounds.
A recurring discount is a price cut with extra steps
Do it twice and you've stopped running promotions and started publishing a schedule. Your customers are not passive; they're inference engines with memory. The second sale teaches them the first wasn't an event, it was a pattern, and rational buyers respond to patterns by timing around them. Your best, most engaged customers — the ones who watch you closely, who'd have been your least price-sensitive segment — become the most sophisticated at never paying full price. Engagement, which you wanted, converts directly into discipline about waiting, which you didn't.
Meanwhile the discount runs adverse selection on the intake. A price cut is a beacon aimed precisely at the most price-sensitive tail of the market: the buyers with the lowest willingness-to-pay, the weakest loyalty, and — the cruel correlation — often the highest support cost per dollar of revenue. You lower the gate and the people who walk in are, on average, the customers you'd have wanted to select against. The number on the tag decides who walks through the door, and this is not a small effect; your price is one of the sharpest selection mechanisms you have, sorting for the customers you get before they ever contact you. A discount aims that mechanism at the wrong tail.
Put the two together. You've trained your good customers to wait and recruited your worst ones to arbitrage. Run that quarterly and you've executed a permanent price cut, except worse than an honest one, because an honest price cut at least doesn't teach the market that your price is a moving target they can game.
A cheap signal says "weak"
There's a reason a fire sale reads as distress and never as confidence. Price carries information, and a voluntary, repeated markdown is information about you. It says: at the number I first quoted, I couldn't move this, so here's what I'll actually take. That is a confession, and everyone in the market can read it.
This is the exact inversion of how a credible quality signal works. The marketing that actually persuades sophisticated buyers is the kind that would be irrational to run if the product were bad — the expensive, hard-to-fake gesture only a confident seller can afford. A discount is the anti-signal: costless to fake, available to any desperate seller, and precisely because it's cheap to send it carries no credible information about quality — except the one unflattering inference that you have margin to give up and a reason to give it. Where trust is the binding constraint, this turns actively dangerous. Building Kommerce for cash-on-delivery buyers in trust-scarce markets — where the customer inspects the goods at the door and can refuse on the spot — I watched discounts read not as generosity but as admission: if it's suddenly cheaper, the first price was a markup on something lesser, and I was nearly the mark. A discount there doesn't lower the barrier to purchase. It raises refusal at the door.
When it's actually fine
The claim is not that price should never move. It's that a discount is toxic exactly when the buyer will generalize from it to your real price. The exceptions are the cases where they won't.
A genuine one-time reason the customer can't extrapolate from. A store relocating, a product line retired, a legitimately time-bound event with a legible cause — these read as local, non-repeating, tied to a state that visibly won't recur. The buyer prices the exception as an exception and leaves the reference point intact.
True penetration pricing where the market is winner-take-most. If the value of your product to each user rises with the number of users — real network effects, economies of scale strong enough to make early share strategically decisive — then buying volume cheaply to reach a tipping point can dominate protecting per-unit price, because the whole game is installed base. This is a narrow condition, and most businesses that invoke it don't have the network effect they're claiming. Be honest about whether you're one of them.
Clearing the perishable. Inventory that expires, seats on a flight about to leave, the last units of a discontinued SKU — anything whose alternative to a discounted sale is a total loss. You're not repricing the franchise; you're recovering salvage value on a unit about to be worth zero. Different math entirely.
Add value at the top instead of cutting the anchor at the bottom
The goal is to improve the ratio of value to price without touching the denominator the customer has memorized. Three moves do that.
Bundle instead of mark down. A bundle changes the unit of comparison, so there's no clean per-item anchor left to reset. The customer can't compute a reference price for "the product plus onboarding plus a year of priority support," which means you can be generous inside the bundle without teaching anyone what your core product's number should drop to.
Time-box a founding cohort, not a sale. Early customers can pay less — but tie it to a legible, non-repeating state: the product is early, you're trading price for their feedback and their logo, and the price steps up as the product matures. The difference from a discount is the inference the buyer draws. A sale says the price will fall again if you wait. A founding-cohort price says this specific window closes and won't reopen, because the reason for it — your immaturity — is something you're actively destroying.
Or hold the line and raise delivered value. Add the feature, ship the faster onboarding, extend the guarantee. You've improved value-over-price by lifting the numerator, and the reference price — the thing your best customers are quietly tracking — never moves.
The reflex to discount comes from treating price as a lever you pull to move volume this month. It's actually a number your customer writes down and defends against you forever. The promotion ends Sunday. The reference price it set does not.