Do You Know What Actually Drives Growth for Your QSR?

Sometime in the next few weeks someone is going to ask you whether the last promotion worked, and you're going to have to answer with confidence.

You'll have plenty to say about the units that moved and the redemptions that came in. But what you likely won't be able to say, at least not honestly, is what would have happened if you hadn't run it — and that's the only version of the question that matters, because it's the one that tells you whether or not to run it again.

If you can't separate the promotions that brought people in from the ones that discounted people who were already on their way, you have no way of knowing what’s really driving growth. And with traffic driven decline increasing almost 10% for QSRs in the past year, the ability to cleanly answer what’s driving your growth is no longer a nice-to-have.

The reason this question is hard to answer has nothing to do with anyone hiding anything. It's that you can't run the quarter twice. To know what a promotion caused, you'd need to see the version of the quarter where you didn't run it, and that version doesn't exist. Every measurement problem in this business is some form of that one.

To accommodate this contradiction, the industry measures the nearest available thing, which is appeal. Technomic can tell you that limited-time offers grew from 17,790 launches in 2020 to 36,830 in 2024, up 134% in five years, tracked to the unit. The standard measures of whether one of those succeeded are constructs like draw and craveability, derived from asking consumers how likely they'd be to buy something after reading a description of it. That's a genuinely useful input to a menu decision. It is not evidence that anything grew.

And it can't be, because the thing you most need to see is invisible in that frame. It's the guest who was coming in on Thursday the way she was always going to, and who traded down from a combo she'd have paid full price for to the LTO she saw on her app. Her redemption gets recorded, her visit doesn't move your transaction count, her check lands lower than it would have, and the margin on it is worse — and every report you get still shows a win.

The first place to look is closer than most teams think, and it's already on your desk. Same-store sales decomposes into transactions and average check, and the useful question about any promotion is which of those two halves moved. A promo that lifts the comp on check while transactions sit flat isn't growth; it's a mix shift you paid to create. One that lifts transactions while check comes down might be exactly the right trade, if those are guests you didn't have before. Shake Shack was unusual this summer in reporting the split out loud — 3.5% comps, of which two points were traffic and one and a half were price and mix. Most brands report the top number and stop. Blended dayparts hide it further, because breakfast and late night are different businesses with different competitive sets and different price sensitivity, and one lift number averages the daypart where an offer opened a door with the daypart that was already full and just got cheaper.

That analysis capability sharpens the question considerably. You stop asking whether the promotion worked, which isn't really answerable, and start asking whether the specific movement you can now see — the transactions that showed up at breakfast, the check that slipped at dinner — would have happened without it. That version has an answer. Getting to it is where the conversation usually stalls, because it seems to require a formal test program: budget, an analytics hire, and the politically impossible act of holding back support in markets that need it. Nobody wants to go dark in a market to prove a point.

Here's the thing: you probably don't have to. Your system has almost certainly already run the experiment.

Restaurant systems are not uniform and never have been. Co-ops don't all opt into the same calendar. Media weight varies by market because budgets vary by market. Franchisees adopt at different speeds, and some of them don't adopt at all. Every one of those inconsistencies — the ones that make your job harder all year — is a naturally occurring test group that a CPG brand would have to pay real money to manufacture.

McDonald's demonstrated this in public over the summer, though not on purpose. Their Everyday Affordable Price architecture went into market alongside a pullback in digital offers and the end of Buy One, Add One, and US comps decelerated from 3.9% growth in the first quarter to 0.8% in the second. The company attributed two-thirds of its traffic shortfall to that value program — and then disclosed the detail that should interest anyone trying to measure anything: only 60 to 65% of US restaurants actually executed the recommended pricing structure.

A third of the system didn't run the program. That is a control group. Nobody designed it, nobody randomized it, and it existed anyway — and it's the difference between knowing the value architecture underperformed and merely suspecting it. Chris Kempczinski's own read on where promotions sit was blunter than anything I'd write: "I mean there's certainly a role for [LTOs], but you're not going to promo your way to long-term value creation." If that's true, then knowing which promotions earn their place stops being an analytics nicety and starts being the whole job.

Natural variation is not a clean randomized test. The markets that didn't execute may differ from the ones that did in ways that also affect sales with different operators, different trade areas, and different competitive pressure, which all needs to be considered in the analysis, but this alternative isn't more rigorous. It's just faster.

And there's a version of this you can start on before your next launch, which costs nothing. Write down, in advance, which markets and which franchisees are getting the promotion, at what weight, and which ones aren't, and why. That's it. The reason most brands can't read their own natural experiments isn't sophistication; it's that nobody wrote down the conditions while they were happening, so six weeks later the variation has to be reconstructed from memory and email. Documented in advance, it's evidence. Reconstructed afterward, it's an argument.

Do that three or four times and you stop guessing. You'll know which of your markets respond to price and which respond to product news, which dayparts have room and which are just getting cheaper, and which promotions on last year's calendar were growth and which were discounts you'd already earned.

That's a different exercise than confirming something happened. It's slower, it takes more setup, and every so often it will tell you that a promotion everybody in the building loved didn't grow anything at all. Right now, with traffic where it is, that's the more valuable finding of the two.

AiOpti will be at the QSR Evolution Conference in Atlanta, September 8–10. If you're working through what your promotions are actually driving, come find us.