Forecasting

Forecasting looks at how a number has moved over time and predicts where it’s going, with a low and high estimate around each prediction.

Who can see this: anyone who can open the project. Asking for a forecast adds steps, so you need to be able to edit the project.

1. The Forecasting step: the table of observed and predicted values.
2. The chart: history and forecast on one line chart.

When to use it

Use Forecasting for questions about what comes next:

• revenue or orders over the coming weeks
• demand for each product
• stock levels by warehouse

It needs a date and the number you want to forecast. It can forecast each group separately, such as each product type, in a single request.

Ask for it

Ask in project chat, and say what to forecast, how to group it, and how far ahead. For example:

Forecast weekly Shopify revenue by product type for 12 weeks.

If you don’t say, Querri groups the data by week and forecasts about 16 weeks ahead.

What runs

A forecast is usually three steps in the Data Flow panel:

1. Forecasting, which builds the forecast. It picks a forecasting method (ARIMA, ARIMAX or Prophet) to suit your data.
2. A step that sorts the result by group and date.
3. A chart of up to ten groups, with the forecast drawn as a dashed line and a shaded range.

Querri may add a cleaning step first if dates or numbers need fixing.

What you get

A table with your history and the forecast together:

Column	What it holds
The value you forecast	Actual values, then predicted ones
value_type	observed for history, predicted for the forecast
{value}_lower_estimate, {value}_upper_estimate	The low and high end of the range for each prediction

The step also explains the results in plain language, and says so if a group doesn’t have enough history to forecast.

Limits

• The model isn’t kept. Each run builds the forecast again from the data.
• Short or patchy history gives wide ranges, or no forecast for that group.
• The step runs generated code with a 10-minute limit.

Credits

Asking uses a credit for the question, as usual. Forecasting doesn’t add a separate charge. Your plan may differ; see How credits are counted on Monitor usage.

Tips

• Say the period. “Weekly” or “monthly” changes the forecast a lot.
• More history, especially whole seasons, gives steadier forecasts.
• If the data is messy, ask Querri to fix dates and fill gaps before forecasting.
• Read the range, not just the line. A wide range means the forecast is uncertain.

Related

• Statistics: test what’s driving a number
• Chat
• Projects