Ecommerce inventory forecasting between stockouts and dead cash

Aug 27
3 min read

Updated: 2 days ago

Introduction

Every online seller sits between two failures. Order too little and the best-selling lines run out during the period they would have earned most. Order too much and the cash is locked in boxes.

Forecasting is the only lever between them, and in most small businesses it is done by looking at what happened last month and feeling optimistic.

1. Ecommerce inventory forecasting has to be done per product

A catalogue-level forecast is useless because products behave differently.

One line grows steadily, another is seasonal, a third sold well once because of a single promotion. Averaging them produces a number that is wrong for every individual product, and the ordering decision is always per product.

2. Forecast from demand, not from what you sold

These are different numbers and the difference matters.

Sales figures are capped by availability. If a product was out of stock for a fortnight, the sales record understates real demand and will cause you to under-order again. Record stockout periods so you can adjust for them.

3. Build the lead time into everything

The question is never how much you will sell but how much you will sell before more arrives.

Supplier production time, shipping, customs, and your own receiving and putaway. A four-week lead time means you are forecasting a month ahead, and every day of unrecorded lead time is a day of potential stockout.

4. Set a reorder point rather than checking stock by eye

A reorder point converts forecasting into a rule that does not depend on somebody noticing.

Expected daily sales multiplied by lead-time days, plus a buffer. When stock reaches that level, you order. This single mechanism prevents most stockouts and takes the judgement out of a decision that is otherwise made when it is too late.

5. Hold safety stock in proportion to the risk

Buffers should not be uniform across the catalogue.

Bigger buffers for unreliable suppliers, long lead times, volatile demand and your highest-margin lines. Minimal buffers for slow, predictable items where the cash is better used elsewhere. A flat percentage across everything both overstocks and understocks simultaneously.

6. Recognise seasonality with more than one year of data

A single year cannot distinguish a season from a trend.

Where you have history, look at the same period in previous years. Where you do not, use whatever proxy exists — category patterns, search interest, your own promotional calendar — and record this year's figures so that next year's forecast has something to work from.

7. Classify your products and spend your attention accordingly

A small proportion of lines will account for most of your revenue.

Forecast those carefully and frequently. Manage the long tail with simple rules and larger buffers, or consider whether they should exist at all. Applying equal effort across hundreds of products means the important ones get insufficient attention.

8. Watch the cash consequence, not only the units

A forecast that is operationally correct can still be financially impossible.

Stock is cash converted into boxes, and a large order placed at the wrong moment causes a shortage that no amount of future sales will fix in time. Forecast the payment schedule alongside the units, particularly before a peak.

9. Compare forecast against actual, every month

This is the step that makes the next forecast better.

Record what you predicted and what happened, per product, and look at the direction of the error. Most sellers have a consistent bias — usually optimism on new lines and pessimism on established ones — and simply knowing your own bias improves accuracy more than any technique.

Conclusion

Forecast product by product, because the ordering decision is made per product and catalogue averages describe nothing.

Adjust for demand suppressed by past stockouts, build full lead times into the horizon, use reorder points rather than visual checks, size safety stock according to risk rather than uniformly, use multiple years to separate seasonality from trend, concentrate effort on the lines that carry the revenue, forecast the cash as well as the units, and review forecast against actual every month to find your own bias.