How much stock each replenishment rule needs to deliver the same service
We compared three common replenishment rules with exactly the same demand: a fixed minimum, fixed weeks of cover and forecast-based replenishment. For each one we measured the stock needed to serve 95 % and 98 % of what customers ordered. The data and the code are public.
Book a 30-minute demoHow it works
Read-only connection · Tested on your own data first · Data hosted in the EU · Demos in English or Spanish
The short answer
In a simulation with 1,200 sales series and two lead times, forecast-based replenishment needed 25 % less stock than fixed cover and 42 % less than a fixed minimum to serve 95 % of demand for seasonal products with a four-week lead time. A fixed minimum needed more stock than forecasting in every comparable case. With steady demand and a short lead time, well-tuned fixed cover did just as well.
By Manuel Gómez González, founder of Arume · Reviewed on
What it does
1,200 synthetic sales series
400 product-location series for each demand type: steady, seasonal and intermittent, over two years, week by week.
Two lead times
Short (one week, sometimes two) and long (four weeks, sometimes three or five), with lost sales when stock runs out.
Three rules, many settings
Each rule is run from its most cautious to its most generous setting, so it draws a full curve of service against stock.
A fair comparison
Instead of comparing one setting with another, we read on each curve how much stock a rule needs to reach the same service level.
Open data and code
Results in CSV and JSON and the Python script with a fixed seed: anyone can repeat the calculation and get the same numbers.
What are the three rules?
Fixed minimum: when stock falls below a minimum, it is topped up to twice that minimum. The minimum is the average weekly sales of the past year times a factor. Fixed cover: every week, stock is topped up to a number of weeks of the average sales of the last four weeks. Forecast: every week, stock is topped up to the forecast for the lead time plus a buffer calculated from the measured error of that forecast, with the seasonality of the previous year.
All three rules see sales, not demand, as happens in real life: when an item runs out, what was not sold does not appear in the history.
What did the study find?
- Seasonal demand, four-week lead time (95 % service): fixed minimum 4.7 weeks of stock, fixed cover 3.6, forecast 2.7. Forecasting needs 25 % less stock than fixed cover and 42 % less than a fixed minimum.
- Seasonal demand, one-week lead time: fixed minimum 1.7 weeks, fixed cover 1.2, forecast 1.3. With a short lead time, fixed cover catches up.
- Steady demand, one-week lead time: fixed minimum 0.9, fixed cover 0.6, forecast 0.8 (the forecast already serves over 97 % with its most cautious setting). Here a forecasting system saves no stock: its value is the alert and the reason behind each shipment.
- Steady demand, four-week lead time: fixed minimum 3.1, fixed cover 1.8, forecast 1.7.
- Intermittent demand, one-week lead time: fixed minimum 7.3, fixed cover 11.3, forecast 6.9.
- Intermittent demand, four-week lead time: none of the three rules reaches 95 % within the settings tested.
Why does the fixed minimum need so much stock?
Because it is calculated with the average of the year and reacts only when stock has already fallen. In the weeks before a peak it orders too little, and after the peak it keeps too much. The longer the lead time, the more those weeks cost, because what was ordered late arrives late.
Why is intermittent demand with long lead times so hard?
When an item runs out it stops selling, so its sales history drops and every rule that reads sales orders less next time. This is censored demand: the history shows what was sold, not what was asked for. Correcting it is one of the things a serious replenishment system has to do, and the simple forecast in this study does not.
What this study is not
It is not the result of any customer and not a promise of what will happen in your company. The forecast used is deliberately simple: the level of the last eight weeks, last year's seasonality and a buffer from the measured error. To know what would happen with your products, what counts is a simulation with your own data.
Download the data and the code
Results in CSV (every setting of every rule in every scenario), results in JSON with the summaries, and the Python script (NumPy, fixed seed). Licence CC BY 4.0: you may use them citing «Arume, Replenishment study 2026» with a link to this page. The full study with the charts is in Spanish, linked below.
Frequently asked questions
Is this real customer data?
No. They are synthetic sales series generated with a fixed seed, so anyone can repeat the calculation and get the same numbers. That way the three rules face exactly the same demand, which with real data is only possible by replaying the past.
Why is it measured in weeks of stock rather than money?
Because the result does not depend on the price of each product. One week of stock is what is sold on average in a week; to turn it into money, multiply by the cost of your weekly sales.
Does forecasting always win?
No. With steady demand and a short lead time, well-tuned fixed cover needs slightly less stock than the forecast in this study. Forecasting wins with long lead times, seasonality or irregular sales, which is where fixed rules react late.
Can I reuse the data?
Yes, under CC BY 4.0. Cite «Arume, Replenishment study 2026» and link this page.
How do I know what would happen with my own products?
With a simulation on your historical data: we replay past dates with what was known each day and compare what the system would have sent with what was actually sent. It is the first step before using the system.