Probabilistic supply-chain forecasting
Lower inventory cost, without lower service.
Wertix forecasts the full distribution of demand for every SKU — not a single guess — then sets the order that meets your target service level at the lowest total cost. Prove the savings on your own sales history before you change a thing.
Upload a CSV. No integration required.
The problem
Spreadsheets and ERP rules quietly overpay.
Most inventory still runs on a point forecast and a fixed cover rule. Both hide the one thing that actually costs money: uncertainty.
- 01
A single number hides the risk
A point forecast can't tell you how likely a stockout is. When demand is uncertain, the average is the one outcome that rarely happens.
- 02
Fixed rules over- and under-stock
A flat “weeks of cover” rule carries dead capital on stable items and starves the volatile ones — the same setting, wrong for almost every SKU.
- 03
The real cost stays invisible
Holding, shortage, and ordering cost are never added up, so no one sees the euros leaking out of the policy each year.
How it works
From demand history to the order you place.
Three steps, one loop. Nothing to re-platform.
- 01
Probabilistic forecast
Every SKU gets a full demand distribution — a fan of quantiles, not a point. Built on the Nixtla ecosystem (StatsForecast, MLForecast, NeuralForecast), Polars-native. No foundation models.
- 02
Cost-optimal policy
The engine reads the whole distribution and picks the order that hits your target service level at the lowest total cost — holding, shortage, and ordering, added up.
- 03
Backtest proof
Replay your own history to see the euros the policy would have saved — the same closed-loop simulation used to score public benchmarks, run on your data.
What you see
The whole fan — and what it saves.
Because Wertix sees the entire distribution, it can hold less safety stock where demand is calm and more where it's spiky — the same service level, lower total cost.
- median forecast
- likely range · 50 / 80 / 95%
- actual demand
Your current policy vs. the Wertix policy at the same service level. Example figures — your numbers come from your data.
Why trust it
Proof, not promises.
No black box, no borrowed logos. The method is the pitch.
-
Probabilistic-first
Every output is a set of quantiles, validated at every data boundary. Decisions consume the distribution, not a rounded average.
-
Decisions priced in euros
The policy is chosen by total cost — holding plus shortage plus ordering — not by forecast error alone.
-
Backtest-verified
A closed-loop replay shows the cost the policy would have incurred on real history, before you commit to it.
-
Open-source foundations
Built on the Nixtla forecasting stack and Polars. Reproducible runs, no foundation-model magic.
Start free
See what your own history says.
Upload a CSV of your sales. Get a forecastability report — how predictable each SKU is — and a labeled estimate of the euros your current policy is wasting each year. No integration, no sales call.
The estimate is exactly that — an estimate, with its assumptions shown.