Optimize the menu for delivery: what gets repeated, what actually moves margin

How many dishes should a small independent's digital menu carry?

Between 12 and 18 dishes, and that number comes from the channel's arithmetic rather than from a trend:

at the home-style restaurant in Bogotá billing 41 million pesos a month through Rappi, the digital menu carried 63 items inherited from the dining room, half of them without a single sale in forty days, while average assembly time climbed to 11 minutes. If you run fewer than 15 tables and delivery accounts for 20% to 45% of sales, every dead item costs you tied-up inventory, one more mise en place line and seconds of customer decision time on screen. Cut first, negotiate packaging afterwards. Healthy industry food cost sits between 28% and 35% according to the National Restaurant Association, and with 63 references you cannot hold that range even counting waste.

NET costing: why a 30% dish in the dining room lands at 44% in the channel

The price reaching your till is not the price on the storefront, and that is where nearly every delivery menu I review falls apart. Take a dish with 30% food cost against selling price: subtract the aggregator commission, the packaging —which in operations with sauce on the side usually runs three pieces per order— and the promotional discount that got you into the week's campaign, and the same dish sits near 44% effective. Against the National Restaurant Association benchmark of 28-35% food cost, an item at 44% no longer works for you: it works for the platform. Every line of the digital menu gets costed with the channel formula —price minus commission minus packaging minus promotion—, never with the dining room one, and anything above 35% effective leaves the menu or raises its price this week.

Product physics: what the photo promises and 18 minutes destroy

Photography closes the order; the trip decides the review, and that asymmetry is the channel's most expensive trap. A breaded milanesa loses its crunch predictably inside a thermal bag, pasta with an emulsified sauce breaks, and a delicate fried item arrives soft even if you change packaging three times. Better for operations with a small hot line: keep stews, rice dishes, bowls, braises and anything that gains from twenty minutes of rest, and pull from the digital menu whatever only exists freshly plated. With Deliveroo reporting a record frequency of 3.5 orders per month per consumer in the UK and Ireland in 2024 (Deliveroo plc), repurchase weighs more than the single ticket, and repurchase dies with a dish that arrives badly twice in a row. Test each candidate item by leaving it sealed 20 minutes and eating it yourself.

The third layer: what the aggregator measures before showing your storefront

The algorithm does not reward your long menu, it rewards your operational predictability, and that is the point almost no owner costs out. Declared preparation time against real time, cancellation rate, items flagged as sold out and order rating: those four indicators move your position on the first screen of the customer eight blocks away. In the Bogotá case, 11 minutes of assembly pushed preparation time past the threshold Rappi uses to rank its storefront, and no promotion offsets showing up in row three. Diego F. Parra frames it at Masterestaurant from the first session with a distinction that saves months: the delivery menu is not the dining room card with fewer dishes, it is a different product, with its own costing, its own packaging and its own time promise. Confusing them explains why so many independents swear the channel does not pay.

When cutting to 12-18 dishes is the wrong move?

Three scenarios exist where the popular recommendation hurts you, and I would rather say it upfront. One: if you run a dark kitchen with several virtual brands over a single line, your unit of analysis is the brand, not the total catalogue;

with roughly 7,606 ghost kitchens active in the United States according to OysterLink (2025), you win with sharp brands of 8 to 12 items each, adding up to more than 18 across all of them. Two: if delivery accounts for less than 15% of sales, running two separate menus costs more than it saves and a short selection inside the single card serves you better. Three: if you sell combos or a rotating daily menu, visible SKU counts mislead you, because your real limit is how many components the line prepares, not how many photos you publish.

Four warning signs when comparing platforms and packaging suppliers

Before signing with anyone, check four concrete things that show up in the numbers and not in the sales deck. First: tiered commissions tied to a volume you do not reach, drafted as if you already did. Second: co-funded promotions where the discount comes entirely out of your margin and the aggregator only supplies the storefront; with a group like Delivery Hero moving 48.8 billion euros of GMV in 2024 (+8%), bargaining power sits clearly on one side and that discount must be costed as your own expense. Third: packaging quoted per unit without telling you how many units your average order carries —three pieces per order turn a cent into a monthly problem of thousands—. Fourth: a contract that withholds your own cancellation and timing data, because without that file you cannot audit why your storefront ranking dropped.

What would happen if you cut the price instead of the menu?

Cutting price to defend delivery volume ends up funding everyone except you, and the exercise can be followed step by step. Suppose you shave 10% off your flagship dish to enter a promotion:

effective food cost jumps several points over that smaller base, packaging does not drop a cent because it costs the same, and commission gets charged on a lower ticket, so the platform loses less than you in relative terms. If volume rises 20%, your kitchen assembles more orders with the same staff and preparation time stretches, which penalises your storefront position exactly when you need orders most. The circle closes with more sales and less profit. The way out is not price, it is a short menu with clean margin.

The right order: cut, recost, renegotiate and measure at 30 days

Cut first, renegotiate afterwards, never the other way round, because negotiating packaging for 63 references means negotiating the wrong problem. Week one: pull every item without a sale in 30 days and every dish with effective food cost above 35%. Week two: unify packaging into two formats at most and quote with volume already concentrated, which is when a supplier actually moves the price. Week three: bring assembly time below 6 minutes with mise en place dedicated to the channel. Week four: measure average ticket, net margin per order and storefront position, then compare against the previous month. And do not expect technology to fix your menu: only 6% of United States restaurants use AI to take customer orders in 2026 according to the National Restaurant Association, so this is still won with costing and judgement.

Five differences between a profitable delivery card and an amputated one

Net costing versus gross costing. In the dining room you look at food cost against selling price, and 30% lets you sleep. In delivery the effective price hitting your till is the price minus commission, minus packaging, minus whatever discount got you into this week's promotion; a dish at 30% food cost in the dining room can land at 44% effective in the digital channel, which makes it a dish working for the aggregator instead of you. Cost every digital item with the channel formula, never with the dining-room one. Product physics versus product photography. The photo sells the order; the 18 minutes in transit decide the review. Breaded cutlets, delicate fried items and emulsified pasta sauces degrade predictably, and changing the package does not always fix it. Some dishes simply do not belong in the channel, and pulling them is not a defeat: it protects the rating of everything else.

Five differences between a profitable delivery card and an amputated one — in practice

Depth versus breadth. A dining-room card can afford a slow-moving item for narrative or affection; the digital card cannot, because every sleeping SKU takes fridge space, lengthens the count and multiplies the sold-out states the algorithm reads as unreliability. Digital rewards depth: fewer dishes, executed identically a thousand times. Built ticket versus expected ticket. In the dining room a server raises the ticket with judgement and a read of the table; in delivery that job belongs to menu architecture, with bundles, well-named add-ons and a deliberate storefront order. Design the digital upsell or it will not happen. A channel versus an identity. Dark kitchens and virtual brands are formats, not strategies. They work when demand is identified, the radius is measured and the kitchen has genuine idle capacity; opening brands to fill a storefront splits the same capacity across more fronts and degrades prep time for all of them.