Methodology

Why this page exists

A benchmark is only worth what its method is worth. Anyone can publish a median; the question a buyer should ask is what was counted, what was left out, and what happens when the data runs thin. This page answers all three before you spend anything.

A longer version is published as a free PDF, and every report restates these rules in its own methodology note, so a figure can still be defended after the report has been passed on to someone who was not in the room.

1. Open-source data only

Every figure traces back to a menu a restaurant published on its own website: the same document a guest reads before booking. Nothing is licensed, purchased or obtained privately. There is no sales data, no transaction data, no reservation data and no confidential submission anywhere in the dataset.

This is a real constraint and it costs us coverage. A restaurant that publishes no menu is simply absent, and we would rather be absent than estimate.

2. No aggregators

Delivery platforms, reservation sites, review portals and listing directories are excluded as sources by design, not by oversight. A menu published through those channels has been edited for the channel and priced for the channel, so it cannot be compared with the menu a guest reads at the table.

This is the single largest difference between MenuIndex and the other datasets available for this region, most of which are built on delivery platforms. It is explained at length in delivery against dine-in prices.

3. One defined segment

The frame is fixed and does not move between reports: Dubai restaurants, cafés, bars and lounges in the $$–$$$$ price segment, from upper-casual through fine dining. Most places covered are more than one of these at once, which is how this market works: a restaurant that runs its own bar, a lounge with a full kitchen. Budget and fast-food outlets are out of scope.

A benchmark is only meaningful if the frame stays still. If the frame moved between reports, a difference between two readings would describe our own coverage rather than the market.

4. Priced menus only

We index the menus a restaurant publishes with prices. Where a line carries no number of its own, because the price belongs to a set menu or a brunch, or because the kitchen prints market price, the line is recorded as part of the menu and never enters a calculation. No price is ever inferred, modelled or carried across from a similar dish.

The consequence is worth stating plainly: the dataset is weighted towards restaurants that publish their prices. Every median you read is built from prices a restaurant actually printed.

5. Aggregation and anonymity

Reports describe the market, not individual businesses. No restaurant is named, quoted, placed in an order of merit or made identifiable, in the text, in a table, in an example or in an image caption.

Some breakdowns we hold rather than publish, and the withheld row stays visible in the table, marked as withheld, so you can see exactly where the data stops rather than being left to assume it continued. We withhold any breakdown drawn narrowly enough to point at one business, however it is drawn.

This rule is not negotiable and it is not a limitation we apologise for. It is the reason a report can be shared with a landlord, a lender or an investor without exposing anyone.

6. Dated snapshots

Every collection cycle is a dated snapshot. Items that disappear from a menu are marked as gone rather than deleted, and every price is held against the date it was published, so the dataset accumulates a price history rather than a current-state view.

Every report produced to order carries its snapshot date. The free documents published on this site are deliberately undated, because they describe how the market is structured rather than what it cost on one particular day.

The six figures we publish

Every figure MenuIndex publishes is one of six. Each has one name and one definition that does not change between reports, because a figure without a name cannot be quoted, checked or argued with.

• Course Ratio. How much more expensive one course is than another on the same menu. On a menu where the starter costs 60 and the main course costs 120, the main course is twice the starter: that is a Course Ratio of 2.00.
• Menu Position. Where a dish stands by price on its own menu, on a scale from 0 to 100. Line up every dish on one menu from the cheapest to the most expensive, with the cheapest at 0 and the most expensive at 100: a lobster dish sitting 83 of the way along has a Menu Position of 83.
• Item Premium. How much more a dish costs than the dishes printed beside it, where it names one of the ingredients we track. The other pasta dishes on a menu run to 100 and a truffle pasta costs 144: it costs 1.44 times as much as its neighbours, so the Item Premium is 1.44.
• Price Tier. The price group a restaurant belongs to, worked out from its own menu. A restaurant where a starter, a main course and a dessert come to 300 AED a head belongs to the 250 to 350 group.
• Tier Band. The usual price range for a course within one price group. In one group, half the restaurants charge between 235 and 285 AED for a main course; a quarter charge less and a quarter charge more. The Tier Band for that group is 235 to 285.
• Anchor Price. The price of an item that barely changes with the price of the restaurant. An espresso costs about 26 AED in a modest restaurant and about 28 in an expensive one, and that gap is small enough to make it an anchor. A glass of wine is not one, because it costs roughly twice as much at the top of the market as at the bottom.

One item, and one menu

Four of the six figures are worked out inside one menu, and neither word is obvious. A dish printed on both the à la carte and a set menu is one priced item, at its à la carte price. A wine sold by the glass and by the bottle is two items, told apart by the serving unit the restaurant printed. The same dish at the same price printed in two files of one menu is one item, counted once.

A file is not a menu. A restaurant publishes several files, and a menu is everything it publishes with prices for its main service, taken together: its all-day and dinner listings, or its lunch listing where it publishes neither. A bar list, a brunch and a children's menu are each a menu of their own. Without that definition, a figure comparing a main course with a starter could be comparing a dinner list with a brunch.

The examples explain the figures. They are not readings of the market: real values carry the date they were read on and are published in dated reports.

Price Tier

An average price for a whole city answers a question nobody asks. A restaurant does not compete with every place in Dubai. It competes with places at its own price level, and that is the comparison our figures are built on.

This industry has no single standard for price groups. The MICHELIN Guide comes closest to a reference point: it publishes thresholds in dirhams for the UAE, counted per person for a full meal without drinks, at 150, 250 and 350. We treat those thresholds as a guide and not as a rule. We tested them against the prices we read, and we publish three groups where the Guide draws four. Our lowest group runs from 100 to 250 AED a head; above it we keep the Guide's own boundaries, 250 to 350 and 350 and above. Below 100 a head a restaurant falls outside the segment this index covers, and we place it in no group at all.

We place a restaurant using its own menu: the median starter, plus the median main course, plus the median dessert, all taken from that one menu, and only where the menu carries at least three priced items in each of the three courses. We use nothing but the menu. Search engines, review sites and map services publish price ratings of their own, but those ratings are estimates of the very thing we set out to measure, and they are built from what customers report rather than from what a restaurant printed. They play no part in placing a restaurant.

The boundaries are fixed numbers rather than the quarter points of our own coverage, because a boundary drawn from the sample would shift whenever the sample changed, and that would let our own coverage back into the figures through the side door.

The test a figure has to pass

Coverage of a market grows over time, and a figure that moves every time coverage grows is describing us rather than the market. Every item carries the date it entered the data, so the data can be rebuilt as it stood on any earlier date. Before anything is printed we rebuild it, run the same query against each rebuild, and look at the spread of answers.

A figure that moves by 5% or less is published. One that moves by more than 12% is not published at all. Between those two it is published only where the report prints how far it moved. The same query is also run on the restaurants present in every rebuild compared: movement on that fixed set is the market, and the difference between the two is us. Nothing is exempt from the test, and it is the reason some figures a buyer might expect to find here are not on offer.

Where a price level is published it is published as a range rather than a single figure, because a single figure claims a precision this data does not have. A range is honest at any size and narrows on its own as coverage grows.

How a figure is written

• Medians, not averages. One very expensive dish moves an average and does not move a median. See why the median, not the average.
• Every figure names what it is a share of. A share of items and a share of restaurants are different numbers and are never mixed in one column. A share of everything we cover is never used, because a reader cannot check it.
• Comparisons are made inside a menu section. A dish priced as a starter and the same dish priced as a main course are two different products.
• Percentages print the numbers behind them. Where a report states that something adds 40%, the price with and the price without are both printed beside it.
• The matching rule is printed beside the table. A strict rule and a broad rule give different answers on the same data, so the rule that produced the figure is stated where the figure is.

Alignment to the restaurant accounting standard

Menu items are classified to the revenue categories of the Uniform System of Accounts for Restaurants (USAR) wherever that is possible, so that our categories line up with the ones an operator already keeps its own books in. Every priced menu item falls into exactly one category: none is left outside and none is counted twice.

The limit of that alignment is worth stating plainly, and we would rather state it than have it found. USAR is a standard about money received. We measure the price printed on a menu, before any discount and before anything is sold, so the alignment holds for how items are grouped and nowhere else. USAR accounts that need a restaurant's internal records are not used.

What we will not sell

Some requests cannot be met at any price, and it saves everyone time to say so here: the menu record of a restaurant supplied to anyone other than that restaurant, an order of merit of named businesses, or a breakdown narrow enough to amount to either. If a question can only be answered by identifying someone else, the answer is no.

One report is the exception, and it is a narrow one. The Menu Position Report describes the menu of the operator who commissioned it and names that one business, its own. Everything it is measured against stays aggregated, and no other business is named or left identifiable inside it. A machine builds it from the rules on this page, and it takes the same form for everyone who buys it.