08.1 Predictions: sales forecast

xeplr AutoML learns from your past sales and predicts the coming days, weeks or months. The predictions are written to a table, so you can report on them like any other data.

Today it offers one kind of prediction: Sales Forecast.

Where to find it

On the workspace Home page, click the xeplr AutoML tile. It’s not in the left menu.

Build a forecast

1. Click Build a prediction (or + New).

2. Predict: choose Sales Forecast and give it a Name.

3. Map features: point Analytics at your data. For each input pick the connection, database, table and column:

   • Date (required)
   • Sales amount (required)
   • Forecast by (store, SKU…): helpful, gives a separate forecast per store or product
   • Holiday calendar / country: helpful, improves accuracy around holidays

   A sample of your data is shown so you can check you picked the right columns.

4. Target: choose the Table the predictions are written to (it’s created if it doesn’t exist) and the Forecast horizon, e.g. 12 weeks.

5. Train: check the summary, then Save and train.

After training

• Previous runs lists each training run with its status, rows predicted, time taken and MAPE. MAPE is the average error as a percentage: lower is better.
• Train again whenever new sales have come in.
• Build a dataset on the target table to put the forecast on a dashboard next to actual sales.

Next: 09.1 The Problems page

Last updated 12 October 2026