The same dish, two prices
A restaurant that sells through a delivery platform is paying a commission that can reach roughly a third of the order. It does not absorb that. It prices the delivery menu higher, and the difference is well documented across markets: commonly in the region of fifteen to thirty-five per cent, with individual dishes far outside that band in both directions.
So a benchmark built on delivery listings answers a real question, but not the one an operator usually has. It tells you what the market charges a delivery customer. It does not tell you what the market charges the guest sitting in your dining room.
The drinks list does not exist there at all
This is the larger problem and it is structural rather than a matter of degree. Alcohol is not sold through delivery platforms in this market. A dataset built on them therefore contains no wine list, no spirits shelf and no cocktail list.
In our own dataset, drinks are the largest part of the menu after food. Wine, spirits and the bar list together account for a very substantial share of every priced item we hold. A menu dataset with that part missing is not a smaller version of the same thing; it is silent on the part of the menu where much of the margin lives.
The top of the market is barely represented
Premium restaurants either do not deliver, or deliver on their own terms rather than through a platform. What appears on delivery platforms from the fine-dining end is a fragment: a few dishes, often a simplified version of the menu, sometimes only during limited hours.
The effect on a benchmark is a quiet downward bias exactly where the interesting prices are. You would not notice it in the data. You would only notice it when the numbers failed to match what you see walking down the street.
Formats disappear
Tasting menus, brunch packages, set lunches and full à la carte lists are not published to delivery platforms, because they are experiences sold in a room rather than products sold in a box. For any question about how a format is packaged and priced, a delivery-based dataset simply has nothing.
What a platform dataset is genuinely good for
It is worth being fair about this. Delivery data is the right source for questions about delivery: what the market charges for delivered food, how promotions are used, how restaurants rank inside an app, how a virtual brand is priced. It is broader in raw count than anything built from restaurant websites, because it captures the everyday segment that often has no website at all.
It is the wrong source for a dine-in benchmark, and the two should not be mixed in one table.
The published menu as a source
MenuIndex reads the menu the restaurant publishes on its own website: the price a guest pays in the room, the full list including drinks, and the formats the restaurant actually sells. That constraint costs coverage, because a restaurant with no published menu is absent from the dataset entirely.
It buys three things a platform cannot: the dine-in price, the whole drinks list, and the top of the market.
If the question is about delivery, use delivery data. If the question is about what to charge in your dining room, about a wine list, or about anything at the premium end, a delivery dataset is answering a different question and will do so confidently.