Marketing Campaign Measurement Across a Franchise Network: Proof Instead of Argument

Marketing campaign measurement franchise leaders can defend rests on a comparison the network already contains: the locations that ran a campaign, set against matched locations held back from it, on transaction counts over identical weeks. The gap between the groups is the effect, and spend divided by it gives cost per incremental dollar.

Every network runs this argument. Headquarters presents reach and recall, an operator says the campaign did nothing for their Tuesday, neither can falsify the other, and the loudest person wins. It is the empirical half of the franchise national vs local marketing budget question, and your own locations can answer it.

What you are trying to measure, and why one location cannot

The quantity you want is a counterfactual: what these locations would have sold had the campaign not run. It is unobservable, so you approximate it with locations that did not.

That is the one thing a network has and a single owner does not. At one location the substitute is a till code against a written-down baseline — franchise local marketing ROI, good enough for a boosted post, but unable to separate the campaign from the weather.

Three decisions shape everything downstream.

• The unit of analysis is the location, not the customer. In a 40-location network your sample size is the locations, not the tickets inside them. This is where statistical power is won or lost, and where most decks cheat.
• The metric is transaction count, not sales. A discount raises transactions and lowers the ticket; sales blend the two, hiding a campaign that worked or flattering one that bought its volume. Track ticket separately.
• The window is fixed in advance: four comparable pre-weeks and the campaign weeks, comparable meaning no holiday, event or closure in either group.

Building the two groups is where marketing campaign measurement franchise networks succeed or fail

Build the groups before the campaign runs and write down who is in which. Everything else is recoverable; this is not.

1. Never let locations opt in. Volunteers are your strongest operators, which guarantees a positive result and teaches you nothing. Self-selection is how a network fools itself.
2. Match, then randomise within the pair. Pair each location with its nearest neighbour on volume band, service mix, tenure and prior-quarter trend, then toss a coin inside the pair. Matching removes the differences you see; randomising handles the ones you cannot.
3. Pair across markets, not inside one. Two locations in one metro share radio, streaming and customers, so a control next door to a test site is partly treated, and spillover makes a real campaign look weak.
4. Keep the controls genuinely dark. No crew brief, no window vinyl, no local post. A half-briefed control is not a control.
5. Verify execution, do not collect confirmations. Participation is normally a checkbox, and a checkbox measures honesty — the accuracy problem underneath self-reported compliance. Use the promotional SKU in the point of sale, or one unannounced visit.
6. Pre-register the decision. Metric, window and what result changes the budget, in writing before week one. A rule written afterwards is a rationalisation.

The arithmetic, on invented numbers

The numbers below are invented; the structure is the point.

Say 20 locations run a four-week campaign and 20 matched locations do not. In the four weeks before, both average 2,400 transactions a week. During the campaign the test group averages 2,568 — up 7% — and the control group 2,448, up 2%.

The naive answer is 7%. The defensible one is the difference between the two changes: 5%, or 120 transactions a location a week. What the control group gained is the season, and you would have had it anyway.

Now convert. Across four weeks that is 480 incremental transactions a location, 9,600 across the twenty. At a $13.50 average ticket, roughly $129,600 of incremental sales. Against $60,000 of media and production, the campaign cost about $0.46 per incremental dollar at the register.

Then say whose dollar it is, because a network has two ratios rather than one. The fund spent $0.46 to produce a dollar the operator keeps the margin on — and the operator paid into the fund. Present both and the room stops arguing about arithmetic, which is the argument worth having.

The power problem at forty units

Here is the part most writing on this leaves out.

Your sample is locations, and locations are noisy. One site's weekly transactions swing several percent on a road closure, a staffing gap or a wet fortnight. With twenty a side, a five-point difference is a real finding and a wide one — so compute a confidence interval from the spread across locations, publish it, and expect it wider than the room wants.

Most networks are in this position by construction. FRANdata's 2017 figures — the newest published distribution of US franchisors by brand size, covering roughly 3,800 of them, reported by Franchise Performance Group — put 82% of brands under 100 units. The textbook version assumes a sample size most systems do not have.

Four disciplines make a small network's version credible:

• Do not slice afterwards. By daypart, region or tenure, something will always look significant. Every cut buys another chance to be fooled.
• Use a switchback when you cannot field two groups. Each location runs the campaign in alternating periods with randomised start weeks, acting as its own control. It trades a clean match for confounding with time — below twenty a side, the better trade.
• Repeat rather than enlarge. The same design four times a year settles more than one heroic test: a consistent direction across repeats is evidence one interval never is.
• Say what would have falsified it. A method that could not have returned a negative measured nothing.

The one-pager, and what it should concede

The output is one page, sent a fortnight before the budget meeting rather than presented in it. Marketing campaign measurement franchise operators trust is measurement they read before anybody is defending a number.

Put on it: the question, the two groups and how they were assigned, the pre-period, the effect with its interval, the cost per incremental dollar, and what you would have to believe for it to be wrong. Then the concessions — spillover, the locations that did not execute, the heatwave in week three.

Publish the campaigns that lost, too: one honest negative buys more credibility than three positives.

Where several locations underperformed for the same reason — a promotion nobody could explain at the till, signage that arrived late — answer them together rather than one at a time: franchise group coaching applied to a debrief.

Franchisees are entitled to more of this than most brands volunteer. The advertising item in the disclosure document, set out in the FTC's Franchise Rule compliance guide, already commits you to an accounting of how the fund was spent — and a result carrying a confidence interval answers that far better than a pie chart. It is the only version an operator has ever thanked me for.

Run this once and you retire an old argument: national reach versus local spend.