Test the change before you commit to it.
Run two versions of a dish — a price, a description, a position — against real guests at the same time, and keep the one that actually won. A version that takes more money while making less of it does not count as a winner here.
- Two live versions
- Measured against each other
- Checked against food cost
What it does
Know what sold, what almost sold, and what caused the difference.
- Both versions run at once
- Half your guests see the new version and half see the current one, during the same services, so the result is not really a story about the weather.
- Traffic is a real gate
- You choose what share of guests takes part, and a holdout can sit outside it entirely. The current menu is one of the versions being measured, not an unmeasured default.
- It reports in money
- The outcome is stated as the difference in what was actually taken, with the covers behind it, rather than as a percentage floating free of any total.
- And in the money you keep
- A discount comes off every plate, so a price cut can lift takings and still leave you poorer. Where the dish’s recipe is costed, the result carries gross profit per guest beside the takings — and a version that wins the first while losing the second is not offered as the winner at all.
- It waits for enough evidence
- A result that has not seen enough covers is reported as unfinished instead of being called early because it looks good.
See it
Example data
£24.00
- Guests shown
- 1,020
- Ordered it
- 102
- Converts at
- 10.0%
- Took
- £2,448
- Profit a plate
- £14.00
- Made
- £1,428
£21.00
- Guests shown
- 1,014
- Ordered it
- 144
- Converts at
- 14.2%
- Took
- £3,024
- Profit a plate
- £11.50
- Made
- £1,656
+£576taken
+£228kept, after £6.00 of food a plate
£3,024 against £2,448, and £1,656 against £1,428 — the figures either side of both differences, so you can check them. The £3 came off every plate, which is why the second number is so much smaller than the first. Had it gone the other way, this version would not have been offered as the winner at all.
How it runs
In order
- Day 0
Pick the dish and the change
A price, a description, a photograph or where it sits on the menu. One change at a time, so the result means something.
- Day 0
Choose how much of the room takes part
Set the share of guests entering the test and hold the rest back entirely. Everyone who enters sees one of the two versions.
- Days 1–13
Both versions run through real services
Nobody has to do anything. Guests are assigned as they scan, and the totals build against each version separately.
- Day 14
Keep the one that paid
The result is reported in money against covers, and in gross profit where the dish is costed. If it has not seen enough covers to be sure, it says so instead of calling it — and if the version that won took more while making less, it says that instead of handing you a trophy.
What changes
Tonight, versus tonight with it on
- Change a price on a hunch and never find out whether it helped.
- Change it for half the room and read the difference.
- A good month gets credited to the new menu and the sun.
- The new version is compared against guests who saw the old one.
- A price cut lifts takings, gets called a win, and nobody ever works out the margin.
- Takings and gross profit are both on the result, and the one that lost you money is not the winner.
Where it sits
One system, five pillars
Experiments is part of revenue & insights.
Know what sold, what almost sold, and what caused the difference. It runs on the same menu, the same tables and the same bill as everything else here — nothing to integrate, nothing to reconcile.
Take the order at the table, the counter or the kerb, and get it to the kitchen.
QR ordering · Digital menus · Live orders · Kitchen display · Guest payments · Dish recommendations · Sage AI waiter
Fill the diary — and give a forgotten table a chance to be sold again.
Bookings & events
Cost the menu on what you are actually paying for the food.
Stock & purchase orders
Compare what you ordered, what arrived, and what you were billed for.
Supplier control
Build the rota, publish it once, and know who did what on the floor.
Team & shifts