Restaurant Analytics: A Complete Guide
Revenue is easy to report and easy to misread. It tells you what came in, not what to do differently next week.
Revenue without context is a vanity number
Every till in the world can tell you what came in. Almost none of them can tell you what to do differently tomorrow, and those are completely different questions. A good week and a bad week can have the same revenue and entirely different reasons, and the number itself will never say which you had.
Useful restaurant analytics answers three things: what is actually making money, when you genuinely need more people, and which parts of the operation are quietly costing you time. None of those come out of a sales total.
Margin, not just sales
The single most valuable report in hospitality plots every dish on two axes: how much it sells, and how much it makes after the ingredients are subtracted. Four corners fall out of that, and each one is a different instruction.
Stars — sells well, good margin
Protect these. Do not reprice them, do not quietly shrink them, and make sure they are the dishes a guest sees first.
Plowhorses — sells well, thin margin
The ones that keep the room busy and the bank account flat. Usually a portion or a supplier problem rather than a pricing one.
Puzzles — good margin, nobody orders it
The cheapest fix on the menu. It already makes money; it just needs to be seen. This is what a screen on the wall is for.
Dogs — sells badly, makes little
The only honest answer is usually to take it off, and the reason it is still there is usually that nobody has looked at this chart.
One methodological point that matters more than it sounds: the crosshair between those corners should sit on the medians, not the averages. One runaway seller drags an average high enough to push half a healthy menu into "unpopular", and then the chart is telling you to remove things that were fine.
Peak hours and table turnover
"Fridays are busy" is not a staffing decision. "Fridays from seven, and Wednesdays are quieter than everyone assumes" is. An hour-by-weekday grid is the cheapest analytics you can look at and usually the first one to change a rota.
Turnover is the other half. Revenue per table and per seat, how many times a table turns in a service, and the median time from order to bill. The interesting result is almost always that the room is not uniform — one section fills first, one fills only when everything else is gone, and the layout has never been changed to reflect it.
The comparisons that quietly lie
Most restaurant reporting is a comparison, and most comparisons in a restaurant are unfair unless someone has thought about them.
Server against server
The gap between two servers' revenue is mostly which section they had. Comparing them without accounting for that measures the floor plan, not the people.
Table against table
A four-top and a two-top are not comparable on revenue. Per seat is the honest unit, and even then a window table is a different product from one by the kitchen door.
This week against last week
Weather, holidays and a football fixture move covers more than anything you did. Longer windows and same-weekday comparisons survive that; a week-on-week percentage does not.
An average with no sample behind it
Three responses averaging 4.7 is not a rating. Any number should arrive with the count it came from, and below some threshold it should decline to say rather than print something confident.
What analytics cannot tell you
It is worth being clear about the ceiling. A system can tell you what was ordered, when, at which table, by which server, and what it cost you. It cannot tell you why a guest did not come back, whether the room was too loud, or that the new starter is being served warm. Those come from asking people, and the honest move is to keep the two apart rather than to let a dashboard imply it knows things it does not.
It also cannot connect things that were never connected. A browsing session on a menu and an order placed an hour later are two events, and joining them because they happened near each other invents a fact. A report that refuses to make that leap is more useful than one that makes it silently.
What to actually look at weekly
The hour grid
Fifteen seconds, and it decides next week's rota.
The four corners
Once a week is plenty. Act on the puzzles first — they are free.
Waste
Not the total. What was thrown and why, because the reason is the fix.
No-shows
By party size and by how far ahead the booking was made.
One dish in depth
Rotate through the menu. Views, adds, orders, and what it gets ordered alongside is where a combination you had not thought of shows up.
The Reporting module covers what Gueridon measures and, more importantly, what it refuses to state when the sample is too small.
Common questions
How much data do we need before any of this means anything?
Hour grids and menu margin start being useful within a month. No-show rates by lead time need a season before you should act on them.
Do we need a separate BI tool?
Not if the reporting reads the same data the service runs on. The reason most venues need one is that their systems do not talk, so somebody has to reassemble the month by hand.
What is the single most useful report?
Menu margin against sales. It is the only one that routinely tells an experienced operator something they did not know.
Can reporting tell us why guests stopped coming?
No. It can tell you that they did, and which ones. Why is a question you have to ask them.