UK Restaurants & Hospitality

One number governs this sector. UK full-service restaurants run net margins of 3 to 6%, and delivery aggregators take 25 to 30% of order value on platform-delivered orders from most independents. The largest available growth channel takes five to nine times the entire net margin of the business.

What the sector publishes

• UK licensed outlets, March 202698,609Observed
• Against 2020−14.2%Observed
• Hospitality insolvencies in 20253,353Observed
• Uber Eats published UK rate: platform-delivered / self-delivery30% / 13%Observed
• Labour as a share of revenue25–35%Observed
• Modelled contribution: on-premise against platform-delivered47% / 17%Modelled
• Sector covers, occupancy and spend per head—Not established

Set two published figures side by side and the problem states itself. Full-service restaurants run net margins of 3 to 6%; Uber Eats publishes a UK rate of 30% when it delivers and 13% when the restaurant does. The commission is not a marketing cost to weigh against alternatives — it is five to nine times the entire net margin, taken off the top of every order. That does not make delivery worthless. It makes it conditional: profitable when the covers are incremental, destructive when they substitute.

Wet-led pubs are excluded on economics rather than on category: drink carries a different gross margin, a different labour model and frequently tied or tenanted property. A food-led pub sits inside the boundary; a drink-led one does not. The takeaway-led archetype is the interesting one, carrying 35 to 60% of revenue through delivery, which makes commission a modelled 12% of total revenue — comparable to rent, rates and utilities combined. Operators negotiate rent once a decade and accept commission as marketing. Only one of the two is changeable by a decision.

Hospitality is the first sector in this series with no register of any kind. There is no regulator, no professional body count and no survey. Operator numbers come from commercial databases, and every operational figure in this model is modelled.

Switching cost is effectively zero — no contract, no registration, no relationship, no friction of any kind — which is why the public review sits inside the value chain rather than after it. It is not feedback; it is how the next customer finds the site at all. It also explains the discount trap: a promotion reliably fills seats, and just as reliably fills them with customers who came for the promotion and do not return at full price. In a market losing a seventh of its outlet base, that is how a busy restaurant closes.

Revenue begins with a fixed quantity of seat-services and is decided by two things the operator controls: how many are filled, and through which channel. The second decision is worth roughly thirty percentage points of contribution and is rarely made explicitly.

Commission rates are negotiable and vary by agreement, volume and location. Uber Eats’ published UK rate card is the firmest anchor; Deliveroo and Just Eat ranges are reported convention. Covers, occupancy and spend per head are modelled: no source publishes them at sector level.

The gap between 17% and 47% is the sector’s largest controllable variable and the hardest to act on, because the decision is not really about delivery. It is about whether a given delivered cover is incremental or substituted, and those are completely different transactions that look identical in every system a restaurant operates. A Tuesday cover filling an idle kitchen is clearly worth taking. A Saturday cover that would have booked a table has converted 47 pence in the pound into 17, and nothing in the till distinguishes them.

Owner labour is normalised to a market replacement rate before economic EBITDA, which matters here because the owner is frequently the head chef. Aggregator commission is shown as its own line rather than inside marketing, because at the takeaway-led archetype it is comparable to rent.

Base central model: 60 covers · 2 services · 6 days · 48% modelled occupancy · £34 spend per head — roughly 17,970 covers a year

Sixty covers over two services and six days gives 37,440 seat-services a year; at 48% occupancy and £34 spend per head that is roughly £611,000, and four points of occupancy is worth about £51,000 at close to full contribution. The third break-even is the one that decides survival. On a 3 to 6% net margin, a two-point rise in labour share removes between a third and two thirds of the entire margin, and no cost programme available to a single site offsets that.

Seat-services bind absolutely and cannot be increased without capital expenditure or longer trading hours. An empty seat at a laid table in a staffed dining room costs almost as much as a full one, which is why occupancy is the widest driver in the sector.

The practical ceiling sits at 72% for structural rather than operational reasons: demand concentrates into Friday and Saturday evenings, and a site sized for peak is necessarily oversized for the rest of the week. A restaurant can turn away bookings on Saturday and sit half empty on Tuesday against the same fixed cost base — which is also where the delivery question resolves. Tuesday delivery covers are genuinely incremental and worth their 17%; Saturday ones are almost certainly substituting for a 47% cover.

Hospitality does not usually have a food problem. It has an occupancy and channel problem, and both are decided before the customer arrives — one by discovery and booking, the other by a supplier contract most operators never renegotiate.

Ask a prospect for these twelve figures. Most can produce revenue, covers and food cost. Contribution by channel, occupancy by service and waste against theoretical are the ones that are missing, and they bracket both of the sector’s widest drivers.

Two decisions determine whether a site works and neither happens during service. Occupancy by service is decided by discovery, booking and how the quiet periods are handled; channel mix is decided by a supplier contract. Both are invisible in the way most sites report — aggregate weekly covers conceal a full Saturday and an empty Tuesday, aggregate revenue conceals a channel taking 28% off the top. The available figures are the ones needed to file accounts; the missing ones are the ones needed to decide which covers are worth having.

Hospitality has the lowest transferable value of any sector in this series and the highest failure rate. Roughly eight restaurants closed per day in the final quarter of 2024 and 3,353 hospitality businesses became insolvent in 2025, with losses concentrated among independents. The ladder below is ordered by survival, not by growth.

Competence is not sufficient, and the closure statistics say so. A site running a 3 to 6% net margin has almost no tolerance for error: two statutory labour increases on a base already at 25 to 35% of revenue, plus a channel taking a quarter of order value, can remove the entire margin without any operational failure. The honest position on rung five is that an independent restaurant is rarely worth much even at the top of the ladder. The rungs are worth taking for the operator’s own income, not for a saleable asset.

Modelled effect on normalised EBITDA of a one-standard-step move in each driver, independent full-service archetype, widest first. In a single-digit-margin business every driver on this list is capable of removing the entire margin on its own.

This sensitivity chart looks different from every other in the series. Elsewhere two drivers dominate and the rest trail; here five of six sit close together, and the clustering is itself the finding. A two-point move in labour share, a three-point move in food cost, a five-point move in occupancy or a shift of covers from dining room to platform can each remove the entire margin independently. In most sectors an operational error costs margin. In this one it can cost the business, and 2025 delivered several at once.

Technology creates value in hospitality only when it moves one of the drivers on the previous page. This is not a software catalogue and this sector is unusually well supplied with systems that add cost without adding contribution. Each intervention below names the driver and the test.

Almost every restaurant can be sold a booking system, a marketing package or a loyalty scheme, and for most none of those is the right first move: they all add covers, and adding covers is only valuable if the covers contribute. A site that does not know its contribution by channel cannot tell whether its last hundred delivered covers made money or moved money from a 47% channel to a 17% one. Measuring it requires no software purchase — the menu and a settlement report produce the number in an afternoon.

The sector model becomes commercially useful at the moment a named site is scored against it. The eight dimensions below convert everything in this model into a diagnostic that can be completed in a single working session with the operator’s own till, booking and settlement data.

Channel economics scores one almost universally among independents, and not because operators are careless: aggregator revenue arrives net in a settlement report and till revenue arrives gross, so the two are never naturally compared. The output is a profile rather than a total, and in a single-digit-margin business the profile decides which intervention is worth the operator’s limited attention. Sequencing follows the sensitivity ordering, which here means measurement before demand generation without exception — adding covers that do not contribute is actively harmful.