The offer brought in sales… or would they have come anyway?

1The offer that “worked”

Nobody asked the awkward question: how much of that would have come in anyway?

The week before Ramadan is no ordinary week. According to Arab News reporting of Saudi Central Bank (SAMA) data, food and beverage spending jumped 74.9% week on week in the week before Ramadan 2025.¹ In one Ramadan week, the same source reported restaurant and café spending down 28.3%.² Then, in the week after Eid, restaurants and cafés were the only category with a positive change.³

The calendar alone moves sales hard. Any offer launched in those weeks will look brilliant or useless depending on which week it landed in, not on the offer itself.

2The Friday queue

A restaurant next to a big mosque. Before the prayer, the owner hangs a “Friday deal” sign. After the khutbah, people pour out, the restaurant fills up, and the owner says: “See? The offer works.”

But that queue was coming anyway. Friday comes every week, and people have lunch after the prayer. The sign might have added ten customers. It might have added none. The only way to know is to see a similar Friday without the sign.

That’s incrementality: not how much you sold during the offer, but how much you sold on top of what you’d have sold without it. And the owner doesn’t pay the discount only on the ten new customers. He pays it on the whole queue.

3The idea beneath the number

In 2015, economists Randall Lewis and Justin Rao published a study of 25 large online advertising experiments. They found that the effect of ads on sales was tiny compared with the natural swings in sales, so tiny that the confidence intervals on return on ad spend were more than 100 percentage points wide, even in huge experiments.⁴ Put simply: the noise is louder than the signal. Look at a sales chart alone and you can’t tell whether your ad made or lost money.

The same year, Thomas Blake, Chris Nosko and Steven Tadelis published what happened when eBay switched off paid search ads on its own brand name. Traffic barely moved. People typing “eBay” into a search engine were going to get there anyway. The researchers estimated the return on those ads at roughly minus 63%.⁵

Careful: those are American, online numbers, and they don’t transfer to a restaurant in Riyadh. What transfers is the lesson: an offer may be paying for a customer who was coming anyway, and only a comparison will tell you.

An honest note: I looked for a published Saudi study measuring the incremental effect of restaurant or retail promotions against a control group, and found none. I have no Saudi number to give you. That in itself is the story: any “successful promotion” figure you hear that doesn’t mention a control group is most likely a before-and-after comparison, not an experiment.

The fix isn’t a complex statistical model. You have something online businesses don’t easily have: branches. Split your branches into two groups that are alike in sales, neighbourhood type and city. Run the offer in one group and leave the other alone, in the same weeks. That’s your holdout group. The gap between the two is the offer’s effect, because Ramadan, payday and the rain fell on both groups alike.

As a group grows and channels multiply, there are open-source marketing mix models, such as Google’s Meridian and Meta’s Robyn.⁶ But they don’t replace a simple experiment. They need one to calibrate their numbers.

4Where the money leaks

A discount isn’t paid on the new customer alone. It’s paid on every discounted order, including those from customers who would have paid full price.

That’s why the doctrine’s rule is plain: an offer is funded only if incremental orders × contribution per order exceeds its cost on all discounted orders, including those that would have come without it.⁷ The same logic applies to points and loyalty programmes: part of the points goes to customers who were coming regardless.

The second leak is worse because it’s invisible: the wrong lesson. An offer that “worked” thanks to the calendar gets repeated every season, until the discounted price becomes the price the customer remembers. At that point you’re no longer buying new orders. You’re paying to stand still.

5The Sunday-morning question

One question, one metric, one rule

7A Friday without the sign

Sales during an offer aren’t the offer’s effect. Its effect is the gap between you and a version of you that didn’t run it. In a branch network, that version exists: in the next neighbourhood.

The idea belongs to its authors. My share is translating it into branch language. Measuring incremental effect through experiments comes from the work of Lewis and Rao, Blake, Nosko and Tadelis, and many others. I’ve only moved it from eBay’s screens to the Friday queue.

Now suppose the test says yes, the offer really did bring new demand. One question remains before you roll it out: can the branch serve it? That’s the next episode.