Use case

Inventory forecasting for a single SKU, without a planning suite

Forecast daily SKU demand from a CSV, read the 80% range around the median, and set a reorder point that accounts for lead time.

Run the sample Put this in production

Inventory mistakes are usually a range problem, not a point problem. A single “expected demand” number hides the week when demand lands 40% above plan, which is exactly the week that creates a stockout. Timestral returns a median and an 80% range, so the reorder decision can use the upper band instead of a false-precision point.

This page is for one SKU at a time: daily units sold, a date column, and enough history to show a weekly pattern. If you sell in several locations, run one file per location or use up to eight numeric columns for related series and read the output per column.

The sample series

The studio below is preloaded with this daily units sold series and a 14-step horizon. It is an illustrative example, not a real company’s data. Replace it with your own CSV to run the same workflow.

Show the CSV

date,units
2026-01-01,42
2026-01-02,38
2026-01-03,35
2026-01-04,31
2026-01-05,44
2026-01-06,47
2026-01-07,52
2026-01-08,49
2026-01-09,45
2026-01-10,40
2026-01-11,36
2026-01-12,46
2026-01-13,50
2026-01-14,55
2026-01-15,53
2026-01-16,48
2026-01-17,43
2026-01-18,39
2026-01-19,51
2026-01-20,54
2026-01-21,58
2026-01-22,56
2026-01-23,52
2026-01-24,47
2026-01-25,42
2026-01-26,55
2026-01-27,59
2026-01-28,62
2026-01-29,60
2026-01-30,56
2026-01-31,50
2026-02-01,46
2026-02-02,58
2026-02-03,63
2026-02-04,66
2026-02-05,64
2026-02-06,60
2026-02-07,54
2026-02-08,49

CSV forecast studio

Forecast your inventory forecasting series.

This studio is preloaded with the daily units sold example above. Replace it with a CSV of your own history, choose a horizon, and run the forecast. The public demo is free while the Hugging Face Space is available; the API is the paid product.

1 · Paste or upload a CSV2 · Pick a horizon3 · Download the 21 quantiles

Steps

1 credit per API call · the web demo is free while the public Space is available

What to watch in the output

Use the upper band for the reorder point

If the cost of a stockout is higher than the cost of carrying extra units, plan against the 80% high band, not the median. The median is the most likely path; the high band is the path you need cash and shelf space for.

Keep lead time in the same unit as the forecast

A 14-day forecast and a 21-day supplier lead time are not comparable. Forecast the lead-time window plus a review period, or convert the supplier promise into days and add a safety buffer explicitly.

Separate a stockout from a demand drop

A zero-sales day caused by an empty shelf is not a demand signal. Replace stockout days with an empty cell or a NaN so the model treats them as missing, not as “nobody wanted the product”.

Common mistakes

—Treating every zero as real demand. Stockout days and closed days distort the pattern; mark them missing.
—Mixing wholesale and retail units in one series. The model will learn an average that describes neither channel.
—Using a forecast beyond the history length as if it were equally reliable. The band should widen; if it does not, treat the result as a warning.

How to put it to work

1. Start with the sample. Run the preloaded series once so you see the output shape before you debug your own CSV.
2. Replace it with your history. Keep one timestamp column and one or more numeric columns. Mark outages and stockouts as empty cells, not zeros.
3. Backtest before you plan. Hold out the last few periods, compare the forecast with what happened, and check whether the actuals land inside the 80% band.
4. Move it into production. Use the REST API for a nightly job or an internal tool. One credit per successful call; failed calls are refunded.

FAQ

Can I forecast several SKUs at once?

Yes, up to eight numeric columns in one file, but only if the series are related at each timestamp. For independent SKUs, separate files produce cleaner results and a clearer error trail.

Does the model know about my promotions?

Timestral does not receive a promotion calendar as a causal input. If promotions are a large part of your demand, add a numeric promotion flag or a related covariate column and check whether the forecast band covers the promoted periods in backtests.

How many days of history do I need?

The hard minimum is eight rows. In practice, at least four to eight weeks of daily data is needed to see a weekly pattern. For monthly inventory, bring at least two full seasonal cycles.

Need this as an API?

The same forecast is available as a metered REST endpoint with JSON and a downloadable quantile CSV.

See the API