Case Study 2: The Deal That Filled the Room
Daily deals, deep discounts, and the incrementality test almost nobody ran
Background
Between roughly 2009 and 2012, American and European small businesses were offered something that looked like free marketing and was in fact the most expensive customer acquisition many of them would ever buy.
Groupon, founded in 2008, built a business on the daily deal: a merchant offered a steeply discounted voucher, Groupon sold it to its subscriber list, and the two split the proceeds. The company grew with extraordinary speed and went public in November 2011 in one of the largest internet offerings of its era. Competitors appeared immediately — LivingSocial and a long tail of imitators — and for about three years, "should we run a deal?" was a live question in essentially every independent restaurant in the country.
The typical structure, as widely reported at the time and as merchants generally described it: the consumer paid about half the face value of the voucher, and the platform retained something on the order of half of what the consumer paid. Terms were negotiated and varied — larger merchants got better splits, and some categories were treated differently — but the merchant commonly received roughly a quarter of the original menu value.
Restaurants were among the heaviest users of the model and became among its loudest critics.
The operating issue
The pitch was simple and, on its face, reasonable: you are paying for customer acquisition, the discount is your acquisition cost, and the customers you acquire will come back at full price. Every restaurant that ran one had heard some version of that argument, and it is the same argument this chapter has been making about the first cover.
The difference is arithmetic.
The per-redemption math
Work it on Bellwether's own numbers, because the structure transfers to any full-service restaurant.
A voucher for \$50 of food, sold for \$25, with the platform retaining half of the \$25:
PER REDEMPTION [constructed teaching example]
Menu value delivered to the guest $50.00
Guest pays the platform $25.00
Platform retains (≈50% of $25) $12.50
RESTAURANT RECEIVES $12.50
Variable cost of $50 of menu value at 60% $30.00
──────────────────────────────────────────────────────────────────
CONTRIBUTION PER REDEMPTION −$17.50
At a 40% contribution margin, \$50 of menu value carries \$30 of food, beverage, and variable labor cost. The restaurant receives \$12.50 and spends \$30. Every redemption destroys \$17.50 of contribution, before counting the seat it occupied, the server's time, and the dining-room capacity consumed on a night that might have been full anyway.
Run it through the identity
§27.7 gave the break-even condition for a discount offer: incrementality $r \ge D \div C$, where $D$ is the revenue forgone per redemption and $C$ is the contribution the visit would otherwise produce.
Here, $D = \$50.00 - \$12.50 = \$37.50$ of forgone revenue, and $C = 40\% \times \$50 = \$20.00$ of contribution:
$$r \ge \frac{\$37.50}{\$20.00} = 1.875$$
The identity returns 187.5%, which is not a number. It is the arithmetic telling you that no incrementality rate can save this offer, because even a perfectly incremental redemption — a guest who would absolutely never have come otherwise — arrives at a loss. When $D$ exceeds $C$, volume does not fix the problem. Volume is the problem.
That is the general rule this case exists to teach, and it applies far beyond daily deals: any time the effective discount exceeds contribution per cover, there is no volume at which the promotion works. Test that condition before you test anything else.
The two effects nobody modeled
Even operators who understood the per-redemption loss usually made the case on two grounds, and both proved fragile.
"They'll come back at full price." This is the retention assumption, and it is the assumption the model most needed to be true. But the guest a deep-discount platform delivers is, by construction, selected for price sensitivity — they subscribed to a list of discounts. A restaurant paying \$17.50 per redemption to acquire a guest whose defining characteristic is that they buy on price has made a bet against the population it purchased.
"It's free advertising." It was not free; it was the most expensive line in the business for the months it ran, and it was invisible on the P&L because it never touched the marketing line. It landed as discounts and comps.
The reputational cost
There was a third effect, which almost nobody anticipated and which is well documented.
A widely cited academic analysis of the period — Byers, Mitzenmacher, and Zervas, examining businesses that ran Groupon offers against their Yelp ratings — found that ratings tended to decline around the deal period. The authors examined several possible explanations, including deal-driven guests being harsher raters and the operational strain of servicing a redemption surge.
Read that alongside §27.3's offset identity and the cost compounds. A restaurant running a deal that loses \$17.50 a redemption was also, plausibly, damaging the free asset that determines whether full-price guests choose it — and at a 4.6 average, each one-star review generated during the surge requires nine five-stars to undo.
The operational version of this is easy to picture. A deal delivers redemptions in a lump, weighted toward the expiration date. Ticket times blow out. The guests in the room are the ones paying the least and expecting the most. Regulars find they cannot get a table. One of the most widely reported episodes of the era involved a small UK bakery that offered a discounted cupcake deal expecting a few hundred takers and received redemptions in the thousands, reportedly losing a substantial sum and consuming months of production capacity to honor them.
Outcome
The daily-deal category contracted sharply. Groupon's share price fell steeply in the years following its offering, and the company moved away from the pure daily-deal model toward marketplace and local commerce products. Merchant participation, particularly among restaurants, dropped as operators compared their second deal to their first.
The model did not disappear, and it is worth saying that it was not universally destructive. It worked best where the structure was different: businesses with genuinely high margins and low variable cost, real excess capacity at specific times, and a plausible route to repeat purchase. Restaurants had the worst version of all three — a 60% variable cost, excess capacity concentrated on exactly the nights deal-seekers were least likely to use, and a price-selected guest population.
The lesson
Run the identity before you run the campaign. Three checks, in order, and the first two take a minute:
- Is $D \le C$? If the effective discount exceeds contribution per cover, stop. No volume, no retention assumption, and no story about exposure makes it work.
- What incrementality does it need? $r \ge D \div C$. Write the number down before the campaign and be honest about whether you believe it.
- How will you know? A code that must be rung, and a hold-out group. Without both, you will learn nothing and will run the same campaign again next year.
And three structural warnings that generalize past this case:
Watch which line it lands on. A discount-funded campaign appears in comps, not in marketing. A restaurant can run a \$40,000 promotional program while reporting a \$12,000 marketing budget, and the P&L will not object. Reconcile them monthly (Chapters 31 and 34).
Beware acquisition channels that select for the wrong guest. The point of Chapter 23's arithmetic is that a guest is worth \$220.80 if they come four times a year for three years. An acquisition channel that systematically delivers guests who will not do that has not acquired anything worth \$220.80, and applying the average lifetime value to them is the single most common error in marketing arithmetic.
A promotion that strains the kitchen is a promotion that damages the asset that was working. The free review average, built over a year of good nights, is easier to lose in one surge than to rebuild in a quarter.
Discussion questions
- Work the per-redemption arithmetic for a fast-casual restaurant with a 70% contribution margin and a \$14 check, under the same deal structure. Does the conclusion change? What does that tell you about which businesses the model actually suited?
- A restaurant argues that the daily deal was worth it because it "got 900 people in the door who had never been." What evidence would you require before accepting that as a defense, and what evidence would settle it against them?
- The Byers, Mitzenmacher, and Zervas finding is a correlation with several plausible causes. Name three, and design a way an individual restaurant could distinguish between them using its own data.
- Is there a version of a deep-discount promotion that a full-service restaurant should run? Specify the conditions — daypart, capacity, margin, and mechanism — under which the identity comes out favorable.
- Compare the daily-deal offer to Bellwether's \$3,500 loyalty discount line, which needs 54.3% incrementality. Both are discounts. Why is one testable and the other not, and what specifically about the design makes the difference?
- The chapter argues that access-based rewards have no incrementality bar because $D = \$0$. Is that entirely true? What does an access reward actually cost a restaurant, and where would it appear?