Complete AI Training

AI agent for business analysts

Forecast Backtest and Refresh Agent

Forecasts that improve over time because every method is tested on the same history

Forecast Backtest and Refresh Agent: what goes in, what the agent does and what you get

What it does

Forecasts are rarely tested against what actually happened. Each month this agent compares past forecasts with the actual results and computes the error by segment. It tries alternative methods on the same history and keeps the better one for each segment. It then checks the new forecast for gaps and outliers, such as a month with a sudden jump. The analyst approves any change of method. It also shows the result as a short chart for each segment so you can see where forecasts are drifting. Edge case: one product line has too little history for a complex method, so the agent keeps a simple average for it.

How it works

Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.

Start and resultWhat it doesA check on its own workWaits for your OKGoes back and retries
Yes, continueYes, continueApprovedNoNo 1 STARTS WHEN Monthly actuals close 2 USES A TOOL Load past forecasts and actuals 3 DOES Compute error by segment 4 DOES Test alternative methods on the same history 5 CHECKS THE RESULT Does an alternative beat the current method by theset margin? If not: keep the current method and note the result.Back to step 3. 6 DOES Build the new forecast with the chosen method 7 CHECKS THE RESULT Does the forecast have no gaps, jumps or impossiblevalues? If not: inspect and fix the segment inputs, thenrebuild. Back to step 6. 8 YOU APPROVE Analyst approves any change of method 9 RESULT Forecast and accuracy report
Read the steps as a list
  1. Monthly actuals close
  2. Load past forecasts and actuals
  3. Compute error by segment
  4. Test alternative methods on the same history
  5. Does an alternative beat the current method by the set margin?If not: keep the current method and note the result. Back to step 3.
  6. Build the new forecast with the chosen method
  7. Does the forecast have no gaps, jumps or impossible values?If not: inspect and fix the segment inputs, then rebuild. Back to step 6.
  8. Analyst approves any change of methodThe agent waits here for your OK.
  9. Forecast and accuracy report

How it decides

A method change is proposed when an alternative beats the current one on error for the same history, by a set margin.

  • Compare methods on the same history
  • Require a margin of improvement before changing
  • Keep a simple method for segments with little data
  • Report error by segment

Make it yours

Every agent is a starting point. You choose these settings for your own situation.

  • Margin to change a method (default: 3 points)
  • Error measure
  • Segments
  • History length

What keeps you in control

It always asks you first

  • Analyst approves any change of method

Hard limits

  • Never replaces a method without approval
  • Never edits actuals

It stops when

  • Done: the forecast is built and accuracy is reported
  • Stop: actuals are not final

Set it up

We guide you through the set-up, step by step

Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.

10 minto set it up in your AI
5 AIsChatGPT, Claude, Copilot, Gemini, Grok
  • One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
  • The agent then walks you through connecting your own data, one source at a time
  • A downloadable copy with the flow chart, the rules and the full guide
Get access to this agent

An example run

What happensLast quarter's forecast missed by 9% for the retail segment. The agent tested three alternatives on 24 months. A seasonal method cut error to 5%, beating the 3 point margin. The new forecast showed a jump in March, so the check failed. The agent found a promotion month with a duplicate entry and corrected it. The analyst approved the change.

More agents for business analysts