Discount Engineering: When Promotions Destroy LTV

TL;DR: A discount is a transfer of revenue from the company to the customer. It creates value only if it brings a customer the company would not otherwise have, and only if the customer it brings is worth more than the discount cost across a sensible horizon. The default discount, the broad-based promotion to the existing list, often fails both tests at once: most of the redemption is from customers who would have purchased anyway, and the lifetime value of the few incremental customers is depressed by the price anchor the discount establishes. The honest framework involves four diagnostics: incrementality, anchor migration, loyalty contamination, and stacking exposure. This essay walks through the empirical literature on each diagnostic and proposes an operating discipline that distinguishes discounts that compound from discounts that destroy.

A note on the named companies. Amazon, Sephora, and Starbucks appear throughout as well-known archetypes of distinct discount strategies. Quantitative figures (incrementality rates, anchor migration, churn-after-discount) come from advisory work with anonymized partner operators in the same archetypes, not from those companies themselves.

---

The Discount Looks Cheap Until You Price the Aftermath

A team running a discount promotion typically sees three numbers immediately: the redemption rate, the conversion lift over the comparable non-discounted period, and the gross revenue from the promotion. All three numbers tend to look good. The redemption rate is high because the offer was attractive; the conversion lift is positive because of course it is, the price went down; the gross revenue is healthy because volume grew. The team logs the promotion as a win, and a second promotion is scheduled for the next quarter.

What the three numbers do not show is the cost of the aftermath. The aftermath includes four effects that the immediate metrics do not capture: most of the redemption is not incremental (the customer would have bought anyway), the customers acquired through the discount have lower lifetime value than organic acquisitions, the existing customer base has been trained to wait for the next discount (the loyalist-discounting trap), and the discount has shifted the reference price downward, making the next non-discounted purchase feel expensive. Each of these effects is well-documented in the academic literature; each of them is silent in the immediate promotion metrics; each of them shows up in the next twelve to twenty-four months as something the team will attribute to other causes.

This essay is a framework for telling the difference between a discount that acquires incremental value and a discount that destroys it. The framework is not a formula. It is four diagnostic questions, with the empirical literature behind each, and an operating discipline that turns the diagnostics into a process rather than a post-mortem.

---

The Four Diagnostics

The four diagnostic questions, in operating order:

Diagnostic 1: incrementality. What fraction of the redemption represents customers who would not have purchased without the discount? This is the most important question and the one most often skipped. It is also the most technically demanding, because answering it requires either a holdout cell (some customers do not receive the promotion) or a quasi-experimental design (a comparable population, comparable period, without the promotion). Without one of those, the incrementality number is a guess dressed as a metric.

Diagnostic 2: anchor migration. How much has the discount shifted the customer's internal reference price downward, and how much margin is destroyed on subsequent non-discounted purchases? This is the Janiszewski and Lichtenstein (1999) range theory effect: the customer's reference price is constructed from the prices they have recently seen, and a discount becomes part of the price range they evaluate future prices against. The next full-price purchase feels expensive because the discounted price is now in the customer's evoked range.

Diagnostic 3: loyalist contamination. What proportion of the redemption came from the most loyal segment of the existing customer base? Discounts that disproportionately reach the customers who would have stayed loyal anyway are pure margin loss with no acquisition compensation. This is the loyalist-discounting trap, and it is endemic in promotion programs that target the email list rather than the acquisition funnel.

Diagnostic 4: stacking exposure. How does the discount interact with other discounts, with the loyalty program, with the affiliate program, with the platform fees, and with the long-run promotion-decay literature? Each stacking interaction either compounds the cost or creates an attack surface where the discount is exploited beyond its intended use. The Mela, Gupta, and Lehmann (1997) finding that long-run promotion intensity increases consumer price sensitivity is the broadest version of the stacking concern: each promotion shifts the population's average price elasticity, making subsequent promotions less effective and full-price sales harder.

The four diagnostics are independent; a discount can pass one and fail others. A flash sale to acquisition prospects might score well on incrementality and loyalist contamination, but poorly on anchor migration if it is run on the same SKUs that the company sells at full price the rest of the year. A loyalty-program discount might be defensible on stacking grounds (it is the only promotion that segment sees) but fail catastrophically on loyalist contamination by definition (it goes to the most loyal customers).

---

Diagnostic 1: The Incrementality Question

The incrementality question is unglamorous because answering it honestly requires foregoing some revenue that the team is convinced is real. The team is asked to set aside a holdout cell: some fraction of the population (typically 5 to 20 percent) does not receive the promotion, the rest does, and the team measures the differential conversion between the two cells over the promotion period and the post-promotion period.

The honest version of the test includes the post-promotion period, not just the promotion window. A discount that lifts conversion during the promotion and then suppresses it for the next four to eight weeks (because customers who would have bought next month bought now) has approximately zero true incrementality, just timing displacement. The displacement is sometimes operationally useful (a discount that pulls Q4 purchases into Q3 to smooth manufacturing capacity is doing useful work), but it is not an acquisition discount and should not be analyzed as one.

The empirical patterns we have observed in advisory work, across roughly two dozen partner promotion analyses in retail and DTC archetypes, fall into a fairly consistent distribution.

The pattern that surfaces from the table: incrementality is strongly correlated with how upstream of purchase intent the discount reaches the customer. A first-time-buyer discount gated to genuinely new customers has the highest incrementality, because the gating mechanically restricts the population to one that mostly was not going to purchase otherwise. A sitewide percent-off blasted to the email list has the lowest, because the email list is concentrated in customers with prior purchase intent. Operationally, this is the most consequential insight from the diagnostic: discount the upstream populations heavily, discount the downstream populations lightly or not at all.

The technical complication is that gating a discount to first-time buyers requires identity infrastructure: the system must be able to tell whether a customer is genuinely new or merely using a new email. Determined arbitrage (creating fresh accounts to repeat first-time-buyer redemptions) is a measurable cost on any gated discount, and the engineering work to suppress it (device fingerprinting, payment-instrument matching, shipping-address de-duplication) is the operating overhead that makes gated discounts harder to run than sitewide ones. Most operators we have worked with underinvest in this infrastructure and then complain that the gated discount did not perform.

---

Diagnostic 2: Anchor Migration and the Reference-Price Effect

The reference-price effect is the most empirically robust and least operationally appreciated discount cost. The customer's perception of "fair price" for a product is not stable; it is constructed from recent observations, including from the company's own discount history. A customer who has seen the product at $40 (full price) and $32 (discounted) more often than at $40 alone will eventually treat $32 as the reference, and $40 as the markup.

Janiszewski and Lichtenstein (1999) formalized this as range theory: the reference price is determined by the range of prices the customer has seen, with the low end of the range pulling the reference downward when the customer is exposed to repeated discounts. The earlier behavioral-pricing literature (Monroe, Kahneman and Tversky's prospect-theory framing of price perception) had reached similar conclusions through different paths.

The operational consequence is that a discount on a SKU shifts the reference price for that SKU, and the shift persists past the discount window. The next full-price sale of the same SKU encounters a customer whose internal reference is no longer $40 but somewhere between $32 and $40, depending on how recent and how frequent the discount exposure was. The customer either does not buy at $40, or buys with reduced perceived value, which surfaces later as lower repeat-purchase rates and weaker word-of-mouth.

The chart illustrates a composite reference-price recovery curve. A SKU at full price of $100, after a 20 percent discount promotion that ended at week zero, shows an internal reference of $80 immediately post-promotion (the customer's most recent observation), drifting upward as the customer accumulates full-price observations or simply forgets the discounted price. Recovery to within 5 percent of full price takes approximately twenty-four weeks in this composite, and to within 2 percent takes around thirty-six weeks. The exact rate of recovery depends on purchase frequency, product category, and the breadth of the promotion. For frequently-purchased categories the recovery is faster; for high-consideration goods the original discount sticks longer.

The implication is that a discount imposes a margin tax on subsequent full-price sales for months after the discount ends. The size of the tax is hard to measure directly but is observable indirectly: the conversion rate at full price after a recent promotion is typically 5 to 15 percent lower than the pre-promotion baseline, in our partner data, for the first eight to twelve weeks. This is the unbooked cost of the promotion. It does not appear on the promotion's P&L line because the lost margin is distributed across the post-promotion sales, where it looks like normal variance rather than a promotion aftermath.

---

Diagnostic 3: Loyalist Contamination

The loyalist-discounting trap is the failure mode where a discount disproportionately reaches the most loyal customers, who would have purchased at full price, and produces minimal incremental volume in exchange for substantial margin loss. The trap is structurally easy to fall into because loyal customers are the ones most likely to open promotional emails, the most likely to be on the company's most active retention lists, and the most likely to have purchase intent during any promotional window. The promotion's redemption funnel is heavily biased toward exactly the segment whose business the company already had.

The Bain loyalty research, anchored by Reichheld and Sasser's 1990 Zero Defections paper and the subsequent Reichheld loyalty work, established that the most loyal customers are the most profitable per unit of activity, partly because they cost less to serve (lower acquisition cost amortized over more transactions, lower support cost per transaction as familiarity accumulates, lower price sensitivity from established habit). A discount that lands on this segment reverses several of those advantages: it teaches the loyalist that price sensitivity is rewarded, it depresses the per-transaction margin on the segment with the lowest variable cost, and it primes the loyalist to wait for the next discount before purchasing.

The diagnostic for loyalist contamination is the redemption distribution by tenure. Healthy promotions that target acquisition have a redemption distribution heavily weighted toward zero-tenure customers (the first purchase). Unhealthy promotions targeting acquisition but reaching loyalists have a redemption distribution heavily weighted toward existing customers with multi-year tenure.

The composite shows the same headline discount, run through two channels. The acquisition-targeted promotion reaches mostly first-time buyers (68 percent of redemptions). The email-blast version of the same discount reaches mostly customers with more than a year of tenure (53 percent of redemptions are at 1-2 years or 2+ years). The two promotions look identical in their pricing-page presentation. They have very different incrementality, very different anchor-migration consequences, and very different loyalist-contamination exposure.

The operating defense against loyalist contamination is not to abandon loyalty discounting; it is to design loyalty rewards that produce behavioral value rather than margin transfer. A loyalty benefit that unlocks an early-access window, a higher-quality customer-service tier, or a referral bonus produces incremental behavior (the customer references the brand, spreads it, deepens engagement). A flat 10-percent-off-everything loyalty benefit produces a margin tax with no behavioral lift, because the customer would have purchased anyway.

---

Diagnostic 4: Stacking Exposure and the Long-Run Promotion-Decay Problem

The stacking problem is the operating-complexity counterpart to incrementality and anchor migration. A discount system that runs on a single promotional channel is simple to reason about. A discount system that has promotions running in parallel (the email list, the loyalty program, the affiliate channel, the new-customer gate, the abandon-cart trigger, the segment-specific holiday push) interacts with itself in ways that compound the costs of each individual promotion.

Three stacking interactions are worth flagging.

Stacking interaction 1: customer-side stacking. A single customer is eligible for multiple promotions and applies them sequentially or together. The 15 percent email coupon plus the 10 percent loyalty discount plus the free shipping over $50 stack to a 25 to 30 percent effective discount the team did not authorize and did not budget for. The fix is per-customer eligibility tracking (a customer who has used a discount in the last N days does not receive another) and explicit non-stacking rules at checkout.

Stacking interaction 2: promotion-on-promotion temporal stacking. Each promotion contributes to the population's expected promotion frequency. After a year of monthly promotions, the customer base does not perceive the next promotion as a special event; it is a regular feature of the brand. The customer's response is to defer non-urgent purchases to the next anticipated promotion, which is now structurally close. This is the long-run promotion-decay mechanism from the Mela, Gupta, and Lehmann (1997) finding: the population's price elasticity rises over time as promotions become more frequent, making each subsequent promotion less effective. The fix is promotion frequency discipline: a documented limit on the number of promotions per quarter, with the limit set below what the team would naturally run.

Stacking interaction 3: affiliate / channel arbitrage. Affiliates and coupon sites publish the company's promotional codes broadly. A code intended for the email list ends up posted publicly on coupon aggregators, where it is found by customers who would have purchased anyway. The intended audience targeting collapses, and the discount turns into a sitewide discount with extra steps. The fix is single-use codes, gated landing pages, or honor-system codes that do not appear in the URL.

The stacking interactions compound. A team running six promotional channels with weak per-customer eligibility tracking, no documented promotion-frequency policy, and codes that leak to aggregators is running, in effect, a sitewide discount most of the year. The blended margin is well below the published list price; the customer base has been trained to wait for the next discount; the incrementality on any single promotion is low; and the only way out involves a politically difficult de-escalation that will look like a price increase to the customer base.

The fourth row is the failure state that many DTC operators slide into without explicit choice. Once the customer base has been trained to expect monthly 20-percent-off promotions, the list price stops being the price. The honest move is to recognize the state, restate the list price downward, and rebuild a promotion ladder from the new floor. The dishonest move (continuing to publish a list price the customer no longer believes) is the more common one because it is operationally easier, and it is the state that the McKinsey discount-effectiveness literature describes as the discount trap.

---

A Decision Framework: When to Discount, How Deep, How Often

The diagnostic framework points toward an operating discipline rather than a formula. The discipline has four components.

Component 1: separate discount budgets by intent. The team should run separately budgeted discount programs for acquisition, retention-of-at-risk-customers, inventory clearance, and seasonal volume smoothing. Each program has different incrementality expectations, different audience targeting, different anchor-migration profiles, and different success metrics. Running them as one budget is what produces the email-blast pattern: all discount activity reaches the same broadly-engaged segment, which is rarely the segment the discount needs to reach.

Component 2: gate aggressively, audience first. The most reliable lever for incrementality is the audience gate. First-time-buyer gates are the strongest; acquisition-channel gates (paid social only, partner-specific links) are the next strongest. The email list and the loyalty list should be the lowest priority for percent-off promotions, because the redemption from those lists is disproportionately customers who would have purchased anyway. The right discount for the loyalty list is rarely a percent off; it is a behavior-unlocking benefit that produces revenue rather than reducing it.

Component 3: measure incrementality with a holdout cell. Every meaningful promotion should run with a 5 to 15 percent holdout that does not receive the discount, and the post-promotion period should be measured for at least eight weeks to capture timing displacement. The holdout costs the team some redemption revenue (the holdout customers would have redeemed had they been eligible) but produces the only honest incrementality number. Without the holdout, the team is post-rationalizing.

Component 4: budget the long-run promotion-decay cost. The team should model an explicit per-promotion contribution to the population's expected price sensitivity and the anchor-migration cost. The model does not need to be precise to be useful; even a directional reservation against promotion budgets ("each promotion reduces next-year full-price conversion by 0.5 to 1.5 percent at the segment level") forces the team to confront the long-run cost rather than booking it as zero.

---

Reading the Margin Aftermath

The honest post-mortem on a promotion takes longer to write than the team usually allows. The promotion itself runs for two to four weeks. The aftermath runs for twelve to twenty-four weeks. The team writes the post-mortem at the end of the promotion window, before most of the cost has shown up.

The post-mortem should answer five questions, and most teams answer at most two of them by default.

1. What was the incrementality (versus the holdout cell)? Most teams skip this because there was no holdout cell.

2. What was the customer-tenure distribution of the redemptions? Most teams have the data but do not report it.

3. What is the full-price conversion rate in the eight weeks after the promotion, compared to the eight weeks before? Most teams do not track this as a promotion cost line.

4. What is the second-purchase rate for the customers acquired through the promotion, compared to the second-purchase rate for organic acquisitions in the same period? Most teams compare cohort-to-cohort but do not segment by acquisition method.

5. What is the next-promotion-redemption rate of the redeemers? A customer who redeems a discount once and never returns is a different outcome than a customer who becomes a serial discount-redeemer; the second outcome is operationally worse because it suggests the customer base is concentrating in discount-dependent segments.

A promotion that scores well on all five questions is a discount the company should run more often. A promotion that scores poorly on three or more is a discount whose cost is concentrated in the aftermath the team is not measuring, and the right corrective is not to refine the next promotion but to question whether the channel and audience should receive promotions at all.

---

When the Discount Is the Right Tool

The framework above is mostly cautionary. The honest counterpoint is that there are situations where discounts are the right tool, and where the alternatives (more advertising spend, more sales hires, more product investment) would produce lower returns.

Acquisition discounts gated to first-time buyers, on a defined budget, with clear arbitrage suppression. This is the cleanest case. The incrementality is high by construction, the loyalist contamination is zero by gating, and the anchor-migration cost is limited because the customer's reference price is being established for the first time at the discounted level. The risk is that the company is over-paying for customers whose LTV does not justify the acquisition cost, which is a separate analysis (the LTV-versus-CAC test) that should run alongside the promotion-margin analysis.

Inventory clearance for genuinely surplus inventory. Discounting inventory the company would otherwise write off is a clear margin gain over alternative dispositions. The honest framing is that the discount is selling units the company has already decided are off-strategy, and the comparison is to disposal cost rather than to full-price margin.

Bundling discounts that promote attached behavior. A discount that produces basket expansion (the second product gets 20 percent off when bought with the first) generates incremental revenue and habit-formation that pure percent-off promotions do not. The bundling has to be honest (the second product is genuinely complementary, not random) for the lift to persist.

Win-back discounts to lapsed segments. A customer who has lapsed (no purchase in twelve months) is operationally equivalent to a prospect, and a win-back discount is structurally similar to an acquisition discount. The incrementality is moderate to high, and the anchor migration is limited because the customer has not been actively transacting at full price.

The pattern across the four legitimate cases: the discount has a clearly defined audience, a clearly defined incrementality story, and a clearly bounded cost. The pathological case is the broad-list, untargeted, repeatedly-deployed percent-off promotion that the team runs because the calendar suggests it should.

A discount creates value when it brings a customer the company would not otherwise have. It destroys value when most of the redemption is from customers the company already had, and the price the customers will accept at the next full-price moment has been quietly shifted downward.

---

A Diagnostic Checklist

For an operator running a discount program today, the questions worth running through, in order, as a documented quarterly review:

1. What proportion of last quarter's promotional budget went to acquisition gates versus broad-list and loyalty-list channels? Healthy programs in our partner data run 50 to 70 percent of promo budget through acquisition-targeted gates; programs in trouble run 20 percent or less.

2. What is the incrementality of the most recent broad-list promotion, measured against a holdout cell? If the answer is "we did not run a holdout," the incrementality is unknown, and the budget for that channel should be treated as expenditure rather than investment.

3. What is the redemption distribution by tenure? Is the loyalty program absorbing more of the promotional spend than it should?

4. What is the post-promotion full-price conversion rate, eight weeks after the most recent promotion ended, compared to the eight weeks before? A persistent gap is the anchor-migration cost.

5. What is the year-over-year trend in promotional frequency? A frequency that has doubled in twenty-four months is on a trajectory toward the permanent-discount state.

6. What is the second-purchase rate for the most recent quarter's promotion-acquired customers, compared to organic-acquired customers? A gap of more than 25 percent suggests the discount is acquiring a different (worse) population.

7. What would happen, in the team's honest estimate, if all promotions paused for ninety days? If the answer is catastrophic, the team is in the trap, and the corrective is gradual de-escalation rather than continued operation in the current mode.

The checklist is not a formula. It is a forcing function. Most teams who run through it discover that one or two of the questions have not been asked, the data exists, and the answer changes the next quarter's promotional plan.

---

Key Takeaways

1. A discount is a transfer of revenue from the company to the customer. It creates value only if it brings a customer the company would not otherwise have, and only if the customer it brings is worth more than the discount cost across a sensible horizon.

2. The four diagnostics are independent: incrementality (would the customer have purchased anyway), anchor migration (has the reference price shifted), loyalist contamination (is the discount reaching the wrong audience), and stacking exposure (do the discounts compound in ways that were not budgeted). A discount can pass one and fail others.

3. Incrementality is the most important and the least measured. A holdout cell of 5 to 15 percent is the operational minimum; without it, the incrementality number is a guess. In partner data, the same headline discount, applied to an acquisition gate vs an email list, can produce five to eight times the incremental revenue per dollar of discount.

4. The reference-price effect, from the Janiszewski and Lichtenstein (1999) range theory literature, imposes a tax on subsequent full-price sales for weeks to months after a promotion ends. The unbooked cost typically extends across twelve to twenty weeks of suppressed full-price conversion. Most promotion post-mortems do not measure this.

5. Loyalist contamination is the failure mode where discounts disproportionately reach the most loyal customers, who would have purchased anyway. The structural fix is not to abandon loyalty discounting but to design loyalty rewards that produce behavioral value (referrals, bundling, category expansion) rather than flat percent-off-everything benefits.

6. The Mela, Gupta, and Lehmann (1997) finding establishes that long-run promotion intensity increases the population's price sensitivity, making each subsequent promotion less effective. This is the long-run promotion-decay mechanism, and the operational defense is documented frequency caps below what the team would naturally run.

7. Stacking interactions compound. Customer-side stacking, temporal stacking, and channel arbitrage each erode the targeting that the original promotion design assumed. The fix is per-customer eligibility tracking, frequency caps, and single-use or gated code distribution.

8. The honest promotion post-mortem takes twelve to twenty-four weeks to write. The team writing the post-mortem at the end of the promotion window is measuring the easy part and missing most of the cost.

9. Discounts are the right tool for first-time-buyer acquisition, inventory clearance, bundle expansion, and win-back of lapsed segments. They are the wrong tool for routine retention of an existing customer base, where the loyalist contamination and anchor migration are largest.

10. The diagnostic question for a discount program in trouble: what would happen if you paused promotions entirely for ninety days? A team that answers "catastrophic revenue loss" is in the trap, and the corrective is gradual de-escalation, not the next promotion.