All guides
Guide · Analytics – BI

Predicting from your own history

Filling in a column you don't know yet — which orders will ship late, say — by learning from the rows where you do, with an accuracy check that says how far to trust the answers.

2 min read

Watch it done: Which orders will ship late?

Predictions takes a table of your own history and fills in the gaps. Give it a list of orders, some shipped late and some on time, and some still open: it learns from the finished ones and predicts the open ones, with an honest measure of how often it would have been right.

The table

Open Analytics – BI › Predictions and click New analysis. Upload a CSV or Excel file of up to 10 MB, with one row per thing — an order, a customer — and one column per fact about it.

One column needs empty cells: the ones you want filled in. The rows where that column is filled are the examples it learns from.

You don't need a file of your own making. A saved inquiry can be the table: open it under Analytics – BI › Inquiries and click Use for predictions.

Setting it up

Under Predict a column, choose the column to predict; the list shows how many cells in each are empty.

Leave Kind of prediction on Auto-detect — it works out whether you are predicting a category (late or on time) or a number (days late) — and the model on TabPFN (recommended).

Under Input columns, untick anything that can't help, such as an order number.

Predicting

Click Predict. Every empty cell gets an answer and, for categories, a confidence.

How far to trust it

Read the accuracy check before you act on the answers. BeanDish holds some of the filled rows back, predicts them as if they were empty, and compares: you see how often it was right and a table of where it went wrong. Download the results as CSV or Excel.

A prediction is an estimate from your past, not a promise. Use it to decide where to look first.

See every step

The video above shows an order history used to find the orders likely to ship late, in about a minute and a half. The tutorial below goes slower, through every choice. The same tutorial is in BeanDish under Learn › Tutorials.

Predict a column from your own history, step by step

A table uploaded to Predictions, the column to predict and the facts it learns from, Predict, and the accuracy check that says how far to trust the answers — plus an inquiry’s rows as the table. (2:16)