Every operator wants a higher average check, and almost everybody measures it in a way that makes it useless. The number moves when a party of eight books instead of two couples. It moves when somebody splits a bill. It moves when you add delivery. It moves for a hundred reasons that have nothing to do with anything you decided, and then people set targets against it and wonder why the targets do not respond. Every one of those movements is recorded separately by your restaurant POS, and almost nobody asks it for the breakdown before setting the target against the headline figure.
There is also a specific problem with the advice available on this subject. A large share of the menu psychology canon, the descriptive item names, the effect of item position, the plate and portion findings, traces back to one research lab whose director had eighteen papers retracted and was found by his university to have committed academic misconduct. So when you read that renaming a dish lifts its sales by a confident sounding percentage, the honest answer is that the field is much thinner than its literature suggests. What follows uses what survives. Some of what survives is not psychology at all but plumbing: an e-menu and mobile ordering flow that offers the modifier at the moment of ordering does more measurable work than any rename.
Which average are you looking at?
Average spend per check and average spend per cover are different numbers, and only one of them is about your operation.
Spend per check is a function of party size before it is a function of anything else. There is a neat demonstration of this in the one decent experiment on menu pricing formats, which we will come to: when the researchers controlled for the variables around the total bill, party size and time at the table were the strongest predictors by a wide margin, with the price format they were actually testing coming a distant last. Party size is not a lever. It is weather.
Spend per check is also distorted by how people pay. A table of six that pays on six cards produces six checks, and your average check falls by roughly a factor of six without a single euro changing hands. If your check count and your cover count have drifted apart, that is usually the reason, and we went into the economics of it in splitting the bill.
So measure per cover, and segment before you draw any conclusion. Lunch and dinner are different businesses. A Tuesday and a Saturday are different businesses. Dine in, takeaway and delivery are entirely different businesses and blending them produces a number that describes none of them. If you take one thing from this article, take this: a single site wide average check figure is not a management metric, and comparing yours to an industry benchmark is worse than not comparing it at all.
Average check is not the goal
Here is the part that separates operators who run a dining room from operators who read reports about one.
The relevant metric is revenue per available seat hour, RevPASH, coined by Sheryl Kimes at Cornell. It is total revenue divided by the number of seats multiplied by the hours those seats were available. Equivalently, it is your seat occupancy multiplied by average spend per cover, divided by the average dining duration in hours. The neat property of that second form is that in a theoretical best case, where every seat is full for the whole hour, RevPASH equals your average spend per cover. The gap between the two is the revenue the room did not capture.
Once you write it that way, the standard advice about upselling becomes visibly wrong half the time, because dining duration is in the denominator.
Work it through. Suppose you have 60 seats, average spend of €35 per cover and an average stay of ninety minutes. Your RevPASH at full occupancy is about €23. Now you introduce a successful dessert and coffee push at your Saturday peak. Spend per cover rises to €39, a gain of over 11%, which is the sort of result that gets celebrated in a meeting. But the extra course adds fifteen minutes to the table. Duration goes to 105 minutes and RevPASH falls to about €22. You sold more per guest and made less money, because at peak the seat, not the guest, is the scarce resource. You turned away the party at the door to sell a tiramisu to the party at the table.
The same push on a Tuesday at 2pm, when half the room is empty and nobody is waiting, is close to free money. The seat had no alternative use. Nothing was displaced.
That gives you a rule that most upselling training completely ignores. At peak, sell things that do not add time: a better bottle rather than a second course, an aperitif while people read the menu, the higher margin version of what they were going to order anyway. Off peak, sell time as well as value, because the extra fifteen minutes cost you nothing. Our guide to table turnover covers the pace side of the same equation, and restaurant KPIs covers where RevPASH sits among your other numbers.
The best evidence that selling works comes from a study about stealing
The strongest causal evidence I can find that servers can lift spend, and by how much, was produced by accident in a piece of research about theft.
In Management Science in 2015, Lamar Pierce, Daniel Snow and Andrew McAfee studied 392 restaurant locations across five casual dining chains that switched on transaction monitoring software, using locations that had not yet adopted as controls. The theft they recovered was worth $108 per restaurant per week. Weekly revenue rose by $2,982, about 7%, and the authors found this came from behaviour change among existing staff rather than from replacing them. The mechanism they examine is that servers who could no longer make money one way made it the other way: they sold. A dessert, a second beer, the better bottle. Their tips rose with the bill.
Seven percent from selling, from the same people, in the same room, with the same menu. That is a much larger number than anything the menu design literature can support, and it points at the actual lever, which is not the menu at all. It is whether the person at the table has a reason to sell.
Which is where European operators need to be careful about importing the result. That mechanism ran through tipping. In a dining room where staff are salaried and a service charge is either absent or pooled and distributed flat, the private incentive that drove the 7% is largely missing, and you cannot conjure it with a briefing. If you want the selling behaviour, you have to build the reason for it deliberately, whether through how your service charge is distributed, a category based incentive, or simply recognition that is specific rather than general. Our comparison of service charge and tips sets out the distribution options. The full theft study, including the parts about what monitoring does and does not do, is in our piece on employee theft.
What the menu research actually supports
Now the menu itself, with the evidence stated at its real strength rather than the strength it gets quoted at.
The one properly conducted field experiment on price presentation is Sybil Yang, Sheryl Kimes and Mauro Sessarego, published in the International Journal of Hospitality Management in 2009. They gave 201 diners at a lunch service one of three menus at random: prices as numerals with a currency sign, numerals alone, or written out in words. Guests given the numeral only menu spent an estimated $5.55 more per check than those given either of the other two, which the authors translate to about 8.15% more per person, from $23.00 to $24.87. There was no significant difference between the currency sign and the written out version, which was the opposite of what they had predicted.
Here is the honest reading. The overall effect of price format was not statistically significant once party size, duration and individual propensity to spend were controlled for. The $5.55 came from a specific contrast between formats, and the authors are careful to say the findings may apply only to lunch at that one restaurant. As one of them put it at the time, changing menu typography is picking the low hanging fruit: the yield may not be large, but it is easy and there is very little downside.
That is the best available study on the most repeated piece of menu advice in the industry, and it is a 201 person lunch. Which should calibrate everything else you read. Drop the currency symbols, because it is free and the one real experiment mildly supports it. Do not expect 8%.
The rest of what is worth doing to a menu is arithmetic rather than psychology, and it is genuinely reliable because it is just accounting: work out contribution margin and popularity for every item, then decide what to promote, reprice, rework or remove. That is menu engineering, and it is where the returns are. Menu pricing strategy and menu design go further into both halves.
Mix, not persuasion
The most useful reframing I can offer is to stop targeting average check and start targeting attach rates.
An attach rate is the share of covers that included a given thing. Covers with a starter, divided by covers. Covers with a dessert. Covers with coffee. Covers with a second drink. These are the numbers that respond to what you actually do, they are readable from your restaurant POS without any new system, and unlike average check they cannot be moved by a large table walking in.
They also make the size of the opportunity concrete in a way that percentages never do. Take a venue doing 400 covers a week. If coffee attaches to 18% of covers and you get it to 26%, that is 32 more coffees a week. At €3.20 with a cost of maybe 30 cents, it is roughly €93 of contribution a week, about €4,800 a year, from one category, without touching a price or adding a minute to a table. Run the same arithmetic on aperitifs before the menu arrives and on the second drink, and you will find that the whole of a respectable average check improvement lives in three or four attach rates that nobody currently measures.
Where to look first
Compare attach rates across your own staff for the same daypart. The spread between your best and worst server on dessert or coffee is your realistic ceiling, because it is being achieved in your room, with your menu, on your guests. It is also the only benchmark that is actually valid for you.
Then compare across dayparts. A category that attaches at 30% at dinner and 8% at lunch is usually a service sequence problem, not a demand problem. Somebody is not asking.
Two sentences beat a briefing
The training that changes attach rates is not "upsell more". It is deciding the exact moment and the exact words for one category, and then checking whether it happened. "Still or sparkling, or tap water for the table?" placed before the menu is a different business from asking at the end. A dessert menu put on the table without a question attaches worse than a named recommendation from someone who has tasted it. Pick one category, script one sentence, place it at one moment, measure for two weeks. Our staff training guide covers the drill.
Digital ordering deserves a note here because it is genuinely different: a screen asks the question every single time, without judgement about whether the guest looks like they want a dessert. That consistency is the mechanism, and it is why mobile ordering and self ordering kiosks tend to report higher attach rates. Treat vendor uplift figures, including any of ours, the way you should treat the 4% theft statistic: ask what the sample was.
Four European rules that break the standard playbook
Nearly all of the advice on raising spend per head is written for the United States, and four common tactics in it are restricted or illegal somewhere in Europe. These are the ones that catch people out.
Germany: your cheapest drink is not yours to choose
Section 6 of the German Gaststättengesetz, known colloquially as the apple juice paragraph and in force in this form since the end of 2001, requires any premises permitted to serve alcohol to also serve non alcoholic drinks on request, and at least one of those must not be more expensive than the cheapest alcoholic drink. Crucially the comparison can be made on the basis of the extrapolated price per litre, which closes the obvious loophole of selling a small glass of juice against a large beer.
The practical consequence for pricing is direct. If you position a cheap beer as your entry point drink, you have just capped the price of a soft drink, per litre, at that level. Nor can you satisfy the rule with something nobody wants: the drink is supposed to be an attractive one matching normal demand, which is generally taken to exclude milk, coffee and hot tea. Austria tightened its equivalent to require two such drinks after operators responded by raising mineral water prices to the limit, which tells you how closely this gets watched.
France: the happy hour has to include the people not drinking
Two rules matter. Under article L3322-9 of the public health code, introduced in 2009, offering alcoholic drinks free and unlimited for commercial purposes, or selling them principally for a flat sum, is prohibited. The open bar, the bottomless brunch and the fixed price drinks package are all in scope, with fines reaching €7,500 and five times that for a company.
Second, if you offer alcoholic drinks at reduced prices during a limited period, you must offer non alcoholic drinks at reduced prices on equivalent terms at the same time. A happy hour on beer with nothing on the soft drinks is an offence. And separately, licensed premises are required to display at least ten bottles or containers of the non alcoholic drinks they sell, kept apart from the other drinks and placed prominently where customers are served. Our note on happy hour has the general mechanics.
Spain: water is not a revenue line
Article 18.3 of Spain's Law 7/2022 on waste and a circular economy, in force since 10 April 2022, requires hospitality and restaurant businesses to always offer customers the option of unpackaged water, free of charge, as a complement to what the establishment sells. It is a consumer right rather than a courtesy, and it cannot be charged for.
Any plan that leans on bottled water attachment does not work in Spain. A filtered or branded water served in a marked bottle at a stated price on the menu is a different product and can be sold, but only alongside a genuine free option that the guest knows about.
Italy: the coperto, and your prices in the window
There is no national Italian rule banning the coperto, and there is no national cap on it. What there is, from a 1940 royal decree that still applies, is a transparency duty: the price list must be available and clear before the customer orders.
Regions then go further, unevenly. Lazio's regional law 21 of 29 November 2006 states at article 16 that where service is at the table the price list must be provided before ordering and must show any service component clearly, and then adds that charging additional costs for the coperto is prohibited. The same article bans extra charges for service, coperto, drinks or anything else not included in a publicised fixed price offer, and requires restaurants to display their price list outside the premises, or so that it is legible from outside, with the wine list excepted. So in Rome the coperto is not available to you as a line, and neither is keeping your prices out of sight until people are seated. Elsewhere in Italy both are fine. Check the region you actually trade in rather than the country.
Measuring it so the number means something
Set this up once and the reporting looks after itself.
Report spend per cover rather than per check, and keep the per check figure only for spotting payment behaviour changes. Segment by daypart, by party size band, and by channel, with delivery and takeaway held completely separate from dine in. Add RevPASH for your two busiest and two quietest trading hours, because those are the four hours where the peak and off peak logic above actually changes your decisions.
Then set targets on attach rates for named categories rather than on the average check itself, because attach rates are the things your staff can influence on a shift and the average is a consequence rather than an action.
What to do this month
Pull attach rates for four categories, aperitif, starter, dessert and coffee, split by daypart and by server. Do nothing else until you have looked at the spread between your best and worst server.
Calculate RevPASH for your busiest hour and your quietest hour. If the two numbers are close, your problem is spend. If the busy hour is much higher, your problem is filling the quiet one, and pushing extra courses at peak is actively costing you money.
Pick one category and one sentence, at one moment in the service sequence, and run it for two weeks. Measure the attach rate before and after.
Take the currency symbols off your menu the next time you reprint. It is free, the one real experiment supports it, and there is no downside.
And if you trade in more than one of these countries, check the four rules above against your drinks list before your next promotion, because the cheapest way to lose money on a higher average check is a fine.
Read next: menu engineering, table turnover, and employee theft.
