Fifteen pounds, a million people, and nobody eating badly

One principal is responsible for feeding a million people every day. Three meals each, fifteen pounds a head. The money exists, it is budgeted, and it is spent in full.

The current arrangement transfers the fifteen pounds to each person and lets them feed themselves. It is simple, it respects their choices, and it does not work, because fifteen pounds buying for one, shopping for one and cooking for one does not put three adequate meals on a table. So the budget is fully spent and some people are still hungry.

That is the cleanest definition of waste available anywhere in this study. Complete expenditure, partial outcome, and nothing on any ledger showing that anything went wrong. No line item says "purchasing power destroyed by fragmentation", so nobody is accountable for the largest number in the system.

The alternative is to engage an agent. Feed as many people as possible, adequately. There is no extra budget. Keep the balance. Everything below is what that instruction actually requires, and what it is worth.

The first result is the one that restructures everything, and it comes out of a single input nobody thinks to ask about: how long a person occupies a place while being fed.

Ninety minutes of dwell against a two hour service window means the seats required are roughly three quarters of the people. Building that is not a difficult capital project, it is an impossible one, and if it were built the facility cost would consume the budget before anybody ate.

So the ninety minutes has to happen somewhere the agent does not pay for, which turns out to satisfy the other constraint as well, because the place people most want to be is where they already are. That forces production, collection and consumption apart. They are assumed to sit in one building only because that is what a restaurant is.

Production then goes where land and labour are cheap, since nobody visits it. Collection goes everywhere people already are, and a collection point is not a kitchen, it is a counter or a locker borrowed from somewhere with existing footfall. Consumption happens at a desk, at home, on a bench, on the participant’s own time and at the agent’s zero cost.

Scale economics want few sites, many covers each, and a narrow menu. All three constraints attack exactly that, which is why the naive answer fails and why the real one is interesting.

Dwell time forbids seating, as above. It is the constraint that does the most structural work and the one least likely to appear in a specification.

Proximity forbids consolidation. People will not travel far, so collection points are placed for a short walk rather than for utilisation, and average occupancy of those points sits well below capacity. That is not inefficiency, it is the price of the constraint, and it should be costed as such rather than optimised away.

Choice forbids the narrow menu, and this is the one that looks fatal and is not. People want variety in dishes, not in ingredients. A narrow set of components combining into a wide apparent menu keeps the purchasing power intact while giving real choice, which is the arithmetic every assembly line food operation runs on.

And pre ordering makes choice nearly free. If people choose ahead, production matches demand exactly, waste collapses, and forecasting stops being guesswork. The price is that they decide in advance, which most people will accept for food they actually want. That single mechanism resolves the tension between choice and scale, and it does something more important that the next chapter but one depends on.

Two rules turn this from a contract into an annuity. The agreement never renews, continuing as long as people want to be fed, and the base is replenished against death but not against departure. So the only thing that can shrink the business is a participant deciding they are not being fed properly. Death is free for the revenue, and as the table shows it is not free for everything.

Revenue is therefore a direct function of service quality, forever, and nothing else touches it. There is no acquisition, no growth and no renegotiation. There is only not losing people.

Which makes the value of one participant the number that should run the entire operation, and it is not the fifteen pounds. It is the whole remaining perpetuity, and it dwarfs the cost of almost any effort to keep somebody. No annual budgeting process will ever produce that conclusion, because it weighs a permanent loss against a monthly cost line and the two are not on the same scale.

It also puts the risk at the worst possible moment. Because losses are permanent, an early failure costs a participant’s entire remaining life, and that failure is most likely in the first weeks, when the operation is least settled and the agent knows least about anybody. The rational response is to deliberately over serve at the start, at a cost per head well above steady state. It looks wasteful under a cost per meal lens and it is cheap under a lifetime one.

The obvious safeguard is for the principal to specify what adequate means, and it is wrong. A specified standard becomes the thing the agent optimises, and it drifts from the thing the participant wanted. That is the funded measure problem, which is the subject of the rest of this study, and writing it into the contract as a protection would reinstate the very failure the arrangement was built to avoid.

Adequacy is person dependent. What keeps one participant will not keep another, so there is no single threshold for an agent to sit just above. There are a million of them, and the only way to stay above all of them is to actually serve each person.

Which is why pre ordering matters far beyond waste control. Every order is a statement of what that person considers adequate, collected continuously, at no cost, without a survey. Adequacy gets learned from revealed preference rather than declared in a schedule.

And the exit is not the only signal, it is the last one. Before somebody leaves they order less, they skip meals, they retreat to the same safe thing, they stop collecting. Declining consumption is the leading indicator of departure, measurable per person, daily. The agent watches consumption rather than satisfaction, which is a better instrument than any specification would have produced.

One residual risk remains and the framing sharpens it rather than removing it. With nothing specified, nothing obliges the agent to keep trying with the people whose adequacy is most expensive to discover. The protection therefore belongs on coverage rather than on the food: not what adequate means, but whose adequacy you are obliged to keep discovering.

A participant who is badly fed does not fall out of the system. There are controls, and a mirror of what has been discovered about them, so another provider takes them on without the usual constraints. Their record travels, so they do not have to teach somebody new what they eat, and there is no gap, no reassessment and no penalty.

One principle sits underneath that and it is worth stating on its own, because it is the opposite of how schemes like this are usually built. THE PARTICIPANT IS GOOD. They are never the instrument of discipline, never the party whose behaviour is being corrected, and never the one who pays for a failure. The entire consequence lands on the provider.

That changes the character of the exit completely. Leaving is now cheap for the participant, because they keep being fed, so they will do it sooner and on smaller provocation. The signal gets faster and more sensitive, which is harder for the provider and better for everybody the scheme exists to serve.

And it makes the loss zero sum rather than merely absolute. A departure no longer evaporates into a cash payment, it arrives at a competitor as scale. So the league table is not a ranking anybody has to construct: it is the net flow between providers, observable monthly, unambiguous, and impossible to present as anything other than what it is.

Two clauses have to hold or the whole mechanism inverts, and they are symmetric. The receiving provider must not be able to refuse. Give it discretion on intake and it will decline the expensive, and the worst served participants become unplaceable, which is the parking problem re entering through the door marked exit.

The other is harder and less obvious. If every participant carries the same fifteen pounds and any rival must accept them, then a provider can deliberately under serve its most expensive people until they leave, and offload its costly tail onto somebody obliged to take it. Free movement plus a flat rate is an arbitrage against the safeguard, and it is profitable.

It is also detectable, which is the useful part. Dumping shows up as the composition of the leavers rather than the number: a provider losing its expensive tail while retaining its cheap core is doing something quite different from one losing people evenly. The remedy is that the money following a participant should reflect what it costs to serve them rather than being flat, so that taking on a difficult person is a transfer of funding rather than a transfer of loss.

The exit right is the cleverest feature of the arrangement. It makes quality and margin point the same direction by construction rather than by supervision: it reports itself, it needs no inspector, it arrives without a survey, and every departure removes a cover from the scale the agent’s economics depend on.

It only works while the agent has margin to protect. An agent operating on a thin margin cannot afford to discover anybody’s adequacy, cannot afford to keep trying with an expensive participant, and cannot afford the choice that makes discovery possible.

And because a departure now arrives at a competitor rather than evaporating, an agent without margin is not merely declining, it is feeding the provider that will replace it. So there are two ways to destroy it and they are the same move in different clothes. Compete the price per head to the floor, or auction the concession for a large capital sum. Either converts the quality margin into a payment, and a payment behaves exactly like fixed cost: it does not shrink when participants leave, so it rises per head precisely as the base erodes.

The tempting valuation at a low price or a high bid is therefore not conservative, it is arithmetically false. It discounts a perpetuity while assuming a retention rate that the price itself has made unfundable. You cannot put the quality spending in the denominator and the base stability in the numerator.

Which leaves the principal with the genuinely hard question, and this page exists to make it precise rather than to answer it. What is the lowest price at which the agent can still afford to care, and how much of the aggregation gain is the principal willing to leave in the food rather than take as a cheque?

None of this was about food.

A resource that is adequate in aggregate and useless when fragmented. A thousand individual preparations of the same thing, each one reasonable, together destroying most of the purchasing power. A constraint on how long any one person can occupy the resource, which turns out to forbid the obvious design. Users who will not wait and will not accept a narrow menu. And a quality signal that exists only as people quietly leaving, one at a time, permanently, without telling anybody.

That is a reporting model. It is the same problem with different nouns, and the table above is the translation.

The row that matters most is the one about people never coming back, because it is already documented elsewhere in this study from the other direction. At least one unofficial spreadsheet per region, maintained by somebody whose job is something else. That is not a governance failure or a discipline problem. It is a participant taking the cash, because what they were offered was not adequate for them, and they will not be back.

And here is the part that should be uncomfortable rather than reassuring. This is what a badly built model looks like from the inside, and a good one is not far removed. A good model is still an aggregation play with the same structure; the difference is unit economics, not kind. Which means its failure mode is identical and merely slower. The users drift off one at a time, permanently, quietly, and nothing in any dashboard records a departure as a departure.

That is the finding this study keeps arriving at from different directions. A model that was properly fixed stays fixed. A different one grows into the same condition unobserved, because nothing watches the curve and the losses do not present themselves as losses. The thing worth building may not be a faster model at all.