1 in 20
false positive rate: periods with no change in which the method reported an effect
Measure whether a change works before you scale it.
Controlled Measurement for multi-location operators
We introduce your change at a selected group of locations and measure it against a control group matched on sales history, seasonality and market. Both groups trade through the same period, so the difference between them isolates the effect of the change from the season, the economy and other external factors.
01
What we can measure
Any change that can be introduced at some locations while the rest continue as before.
Price & menuPrice changes, new items, changes to the menu mix.
Assortment & inventoryProduct range, stock levels and placement.
LabourStaffing levels, shift structures, incentive schemes.
VendorsSupplier changes, contract terms, product specifications.
MarketingLocal promotions, new channels, changes in spend.
Systems & equipmentPOS features, new equipment, process changes.
02
How a test is designed
With tens of locations rather than thousands, random assignment rarely produces comparable groups. Most of our work goes into constructing a control group that closely matches the test locations. The design depends on the change being tested.
| Design | When it applies | Basis of comparison |
|---|---|---|
| Matched pairs | The change is limited to a subset of locations. | Each test location is paired with the control location whose historical sales most closely match its own. |
| Staggered rollout | Every location will receive the change eventually. | Locations receive the change in waves. Those still waiting serve as the control group. |
| Switchback | The change can be switched on and off quickly, such as a staffing rule or a promotion. | Each location alternates between on and off periods and is compared against itself. |
| Synthetic control | The change cannot be withheld, or has already been made. | A weighted combination of unchanged locations that closely tracked the changed ones before the change. |
This approach has been standard practice at the largest retailers and restaurant chains for decades. Causeway applies it to operators with 20–300 locations.
03
How we’ve developed the method
We measure the method on two things: how often it reports an effect that isn’t there, and how small a real effect it can reliably find. We tested both on real sales data from 39 multi-location businesses in public Texas records, using periods where nothing changed and periods where we had added an effect of a known size. The results are below.
1.8%
smallest effect on sales a three-month test can detect, with 20 of 65 restaurants making the change
5%
typical month-to-month variation in one restaurant’s sales when nothing has changed
Month-to-month variation at a single restaurant is typically around 5%. That is larger than most effects worth measuring, so comparing one location before and after cannot separate the effect from the noise. A matched control group can. Every result is held to a 95% confidence threshold. At Texas Roadhouse’s 65 Texas restaurants, a 3.1% increase in sales from their $5 offer is worth about $4.7 million a year.
Causeway is not affiliated with Texas Roadhouse, Inc. All figures are derived from monthly restaurant-level sales filed with the State of Texas, January 2018 to December 2025, and can be reproduced from our analysis. The full study →
04
How an engagement works
- 01AssessmentWe review the planned change and tell you whether it can be tested, and the smallest effect a test could detect. If it cannot be tested reliably, we say so before any work begins.
- 02PlanWe select the design and the test and control locations. Before launch we agree in writing the measure you will judge the result on, how long the test runs, and the result that means roll out or stop.
- 03RolloutThe change goes live at the test locations on your schedule. Control locations continue unchanged.
- 04ResultAt the close of the test you receive the overall effect and how it varied by region, format and location. The variation is often what determines the decision.
The plan is fixed once signed. We do not propose changes and have no stake in the outcome, so we can work alongside your existing advisers.