Restaurant win-back campaign ROI: how to tell if it actually made money

Almost every report you will be shown counts guests who were coming back anyway. Here is the one decision that separates the two, and the arithmetic that goes with it.

The number you get shown

A typical end-of-campaign report reads something like this. We contacted 1,200 lapsed guests. Ninety-four of them came back. At your average check, that is several thousand dollars of recovered revenue.

Every figure in that sentence can be true and the conclusion still be wrong, because of the question it does not ask: how many of those ninety-four were walking in anyway?

Lapsed does not mean gone. A restaurant's lapsed list is not a graveyard, it is a queue. People were travelling, or busy, or had a stretch of cooking at home, and some proportion of them drift back every month with no prompting at all. That drift is invisible unless you measure it. And when it is invisible, it gets billed to the campaign.

This is not a small correction. It is the difference between a campaign that pays for itself and one that quietly does not, and the report looks identical either way.

The fix is one decision, made before you send

Take the lapsed list and set aside a random slice before anything goes out. Ten to twenty percent is the usual range. Those people receive nothing. Then, over the same window, compare how many came back in each group.

The gap between the two rates is what the campaign caused. Everything below it was going to happen anyway.

Seventy-eight guests came back. Thirty-three of them came back because of the campaign. The other forty-five were the queue moving on its own, and a report without a holdout would have counted all seventy-eight.

Then finish the arithmetic honestly. From those thirty-three visits, subtract what the offer gave away and what the sending cost. What survives is the campaign's real contribution, and it is the only figure worth putting next to what you paid for it.

Two things that make a holdout honest

It has to be random

Not the guests without a phone number. Not the ones whose email bounced last time. Not the oldest slice of the list. Any of those choices makes the held-out group different from the contacted one in a way that has nothing to do with the campaign, and then the comparison measures that difference instead.

The window has to be fixed before the send

Both groups get counted over the same dates, and those dates get decided in advance. Choosing the window after seeing the numbers is the most common way a flat result turns into a good one, and it is rarely done dishonestly. It just feels reasonable to wait another two weeks when the figure is disappointing.

Why opens and clicks are not the answer

Open and click rates tell you whether a message was seen. They cannot tell you whether it changed a decision, and a campaign is only worth anything if it changed a decision.

They are worth watching for what they are good at: catching a technical failure. If opens collapse, something is broken, and that is useful. But a healthy open rate next to an unmeasured return rate is a comfortable number standing in for the uncomfortable one.

Five questions to ask whoever runs your campaigns

1. Was a group held out, and how many people were in it? The raw count matters more than the percentage.
2. How were they chosen? The answer should be some version of "at random". Anything else needs explaining.
3. What was the return rate in each group? Two numbers, not one.
4. Over what window, and was it the same for both? And was it set before the send.
5. Does the revenue figure subtract the offer and the cost of sending? Recovered revenue and contribution are not the same number.

A provider who runs campaigns properly will have all five answers ready, because they had to decide each one before pressing send. If the answers arrive as estimates after the fact, that tells you the measurement was not designed, it was reconstructed.

When you genuinely cannot hold a group out

Below roughly two hundred contacts, a holdout stops being informative. Split a small list and each group ends up so small that two or three extra walk-ins swing the rate by several points, which means the comparison is noise wearing the clothes of a result.

If that is your situation, the honest version is a before-and-after comparison over matched calendar periods, stated with its limitation attached: it cannot separate the campaign from the season, the weather, or anything else that happened at the same time. It is directional, not proof, and it should be described that way rather than dressed up.

And the more useful conclusion is usually the one nobody wants to hear. With a list that small, the bottleneck is not the campaign. It is that the restaurant is not capturing who walks in, and fixing capture is cheaper than any campaign you could run against the list you have.

Questions owners actually ask