1.8%
the change in sales that a three-month test can read
the case
Texas Roadhouse: what a test would have told them.
Texas Roadhouse runs 65 restaurants in Texas that take about $560 million a year. In 2025 it put a $5 pint and a $5 margarita on the menu, first in some restaurants, then in all of them. The company says the offer is popular but it has not said what it is worth, and on the numbers it reports it cannot. A restaurant's sales move about 5% month to month on their own, so comparing after with before would have called the offer a success four times in five even if it had done nothing.
The offer is a real decision with real money on both sides. A discount gives away margin on every pint and margarita that would have sold at full price anyway, about $3.8 million a year at their menu prices, and the point of a $5 drink is the people it brings through the door. The line that answers whether it worked is total sales, and the straight read cannot answer it: run on their own history, in windows where nothing was changed, it declared an effect 81% of the time.
the test
Twenty restaurants, three months.
Twenty of the 65 restaurants run the offer for three months. The other 45 carry on exactly as they are. The twenty are chosen so that the two groups match on their own recent trading, which means everything that moves the whole chain cancels out and what is left is the offer. Weekly sales from the till are all it takes. The test confidently reads changes in total sales down to about 1.8%, and the group of 45 gets the offer the day the read says it pays.
20 of 65
restaurants that change anything
$1.1M
what the test itself is worth a year, on this one decision
81%
how often the straight read calls a win when nothing changed
the worth
What that is worth.
Suppose the read comes back at +3.1%.
The offer is worth about $4.7 million a year across the 65 restaurants, after paying for the discount, and nine times in ten between $2.3 million and $7.0 million. The chance it cleared its break-even is above 99%. The decision is to roll it out to the other 45.
The break-even is the operator's to place. Here it sits where the offer pays for its own discount. An operator building a habit that pays back over years rather than a quarter, moves the line to what they are willing to spend, and the read says whether they got it. A plan to lose $2 million is a different plan from one that loses $4 million, and only a read tells you which one you are running.
The gray band is where that decision would have started. Before any test, the only thing anyone knows about a new offer is how offers like it have done: about half lose money or do nothing, and the other half are worth one to three percent of sales. Rolled out blind, this offer had a 46% chance of losing money. Deciding with the test instead is worth about $1.1 million a year on this one decision, and every test adds to the operator's own record.
the record
Where the numbers come from.
Causeway is not affiliated with Texas Roadhouse, Inc., and nobody at the company was contacted. Every number on this page is computed by seeded, reproducible scripts from monthly restaurant-level sales filed with the State of Texas, January 2018 to December 2025. The promotion and the company's statements about it are taken from its annual report and earnings calls.
The full case study: download the PDF.
If you run 20 to 300 sites and want this on your own numbers, write to hello@causewaydata.com.