Reorder Point for a Store Chain: Why "7 Days of Cover, Top Up to 14" Loses Money

Ask how a store chain decides when to reorder and the answer is often a rule of thumb in days of cover: when a product is down to about a week of sales, order enough to bring it back to two weeks. It is easy to explain, easy to put in a spreadsheet and wrong in a way that depends on the supplier. This article sets the rule against the textbook alternative, shows what the difference was worth in a simulated year across four chains, and works through the formula for two suppliers.

1. Two ways to decide when and how much

Write μ for a product's average daily sales in a store and IP for its inventory position, stock on hand plus stock already on order. The rule of thumb, on a supplier's order day: if IP is below 7μ, order up to 14μ².

The periodic-review formula, known as (R, S), asks how long this order has to last. An order placed today arrives after the lead time L and must last until the next order can arrive, R days later, where R is the wait until the supplier's next order day. Over those L + R days the store expects to sell μ(L + R), and it adds a buffer for the spread of demand σ and of the lead time σ_L: S = μ(L + R) + z·√((L + R)σ² + μ²σ_L²). On each order day it orders up to S. The factor z sets the service level; Retail-Bench's tuned version uses 2.05, 1.64 and 1.28 for products in the top, middle and bottom revenue classes, which correspond to about 98%, 95% and 90%, and caps short-life products at their shelf life of sales².

2. What a simulated year shows

Retail-Bench, the open benchmark marql built, runs four simulated chains for a year after a three-month warm-up: convenience stores in London, fresh grocery in Manchester, home and garden in Bristol with suppliers four to six weeks away by sea, and a drugstore chain in Leeds kept as a hold-out. Every contender meets the same customers and the same late trucks, over twenty runs per chain, so the differences come from decisions alone. Profit is counted after the cost of holding stock, write-offs and ordering^{1, 2}. Against the tuned formula, the rule of thumb earned:

The pattern follows the lead time. Where suppliers deliver in a day or two the rule costs a few per cent of profit; where they are weeks away it costs half. On the benchmark's own scale, which runs from the rule of thumb at zero to a contender that knows the future at one hundred, the tuned formula captures 60% of everything better ordering could win¹.

3. Why the rule loses: a worked example

Take one product that sells four units a day on average, with a daily spread of three, at a 95% service level. Buy it from two suppliers: a nearby one that delivers in three days and takes orders once a week, and an importer that delivers in 38 days, give or take four, also ordered weekly.

For the nearby supplier the rule's target is about right, but its trigger is not. It looks only at stock against a week of sales, and forgets that the next chance to order is a week away and the truck after that takes three more days. Thirty units look safe and last seven and a half days. For the importer the rule is simply too short: two weeks of stock for an order that has to last six and a half. Both errors come from the same omission: the rule counts days of cover but not the days the cover has to last.

The rule also has no way to see an erratic supplier. Late deliveries widen σ_L, and the formula widens the buffer with them; the rule keeps the same seven and fourteen days for a punctual supplier and an unreliable one.

4. What the formula needs from your data

The first input matters more than it looks. A day that ended with an empty shelf records low sales, not low demand; averaging it in pulls μ down and the next order with it. Retail-Bench's formula reads those days at the level of normal ones², and the out-of-stocks guide covers how to spot them. Round the result up to the case pack and the supplier's minimum, and check the other end too: the same arithmetic run against stock that is not selling is how the dead-stock guide finds excess.

5. Where the formula is not enough

The formula is a strong baseline for a reason. When a lost sale costs far more than holding a unit for a while, order-up-to rules are close to the best possible; in Retail-Bench a lost unit costs as much margin as holding that unit for 506 to 1,128 days². It is weakest with perishables, where every extra unit risks a write-off: in the fresh-grocery chain the tuned formula captured only a small part of what perfect foresight could¹.

Whether AI adds to it is less settled than vendors suggest. In InventoryBench, a 2026 academic benchmark of single-product inventory decisions, a language model reviewing a capped base-stock policy scored 0.538 against 0.445 for the policy alone, and the authors note that models "sometimes falsely identify demand shifts when it is pure noise"³. The guide to AI in store replenishment covers the field evidence on people and algorithms working together.

6. Limitations

The figures in section 2 come from a simulation, calibrated against real chains but not measured in them, and from the benchmark's version of 19 September 2026. The simulation does not yet model substitution, shelf capacity, markdowns or a distribution centre. marql builds the benchmark and competes in it; the baselines discussed here are not marql's, and the full results, including the rounds where marql's own engine lost, are published.

How marql uses this formula

In Retail-Bench the tuned formula is the bar every contender is measured against, marql's own replenishment engine included. The first round put that engine below the formula in three chains of four, and the replays showed why: a buffer blind to how erratic a supplier was and to how far away the next order day was. Any ordering system can run through the same protocol, and every result is published.