Complete AI Training

AI agent for finance and accounting specialists

Forecast Accuracy Variance Agent

A weekly accuracy report with traced causes and approved driver changes that reduce error

Forecast Accuracy Variance Agent: what goes in, what the agent does and what you get

What it does

A cash forecast that is wrong the same way every week loses the trust of the CFO. Each week this agent compares the forecast with actual cash by category, such as customer receipts, payroll, tax and suppliers. It finds the largest misses and traces each to its cause: timing, amount or a missing item. For repeated timing misses, it tests an adjustment to the driver, for example the average days customers take to pay, and backtests the new driver on recent weeks. It only recommends the change if it would have cut the error. If it did not, it tries another driver or marks the miss as one-off. The analyst approves which drivers change in the model. Edge case: a large one-time tax payment that moved by a week is classified as timing and not allowed to change the driver.

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 Monday after actuals post 2 USES A TOOL Load last week's forecast and actual cash 3 DOES Compute variance by category and rank the largestmisses 4 USES A TOOL Trace each miss to invoices, payments or missingitems 5 CHECKS THE RESULT Does the traced cause explain most of the gap? If not: look at the next data source and retrace themiss. Back to step 3. 6 DOES Classify each miss as timing, amount, missing itemor one-off 7 DOES Test an adjusted driver on recent weeks 8 CHECKS THE RESULT Does the adjusted driver cut the average error? If not: try another driver or mark the miss as one-off.Back to step 7. 9 YOU APPROVE Analyst approves which drivers change 10 RESULT Accuracy report and updated driver list
Read the steps as a list
  1. Monday after actuals post
  2. Load last week's forecast and actual cash
  3. Compute variance by category and rank the largest misses
  4. Trace each miss to invoices, payments or missing items
  5. Does the traced cause explain most of the gap?If not: look at the next data source and retrace the miss. Back to step 3.
  6. Classify each miss as timing, amount, missing item or one-off
  7. Test an adjusted driver on recent weeks
  8. Does the adjusted driver cut the average error?If not: try another driver or mark the miss as one-off. Back to step 7.
  9. Analyst approves which drivers changeThe agent waits here for your OK.
  10. Accuracy report and updated driver list

How it decides

A miss is traced when the cause explains at least 80% of the gap. A driver change is proposed only if backtesting it on the last 8 weeks reduces average error.

  • Investigate any category miss above 5% of the weekly total
  • Call a miss timing when the cash arrived within 2 weeks
  • Change a driver only when the backtest improves error on at least 6 of 8 weeks
  • Do not change drivers for one-off items over a set amount

Make it yours

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

  • Miss threshold that triggers tracing (default 5%)
  • Backtest window (default 8 weeks)
  • Categories to track
  • Day the report is produced

What keeps you in control

It always asks you first

  • Changes to forecast drivers

Hard limits

  • Never edits the live forecast model
  • Does not hide large misses by averaging

It stops when

  • Done: misses are traced and approved drivers updated
  • Stop: actuals are incomplete

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 week receipts were $1.9 million against a $2.6 million forecast. The agent traced $520,000 to three large customers paying 11 days later than the 38-day driver. Testing 45 days over 8 weeks cut error by 31%. A $90,000 refund was labeled one-off. A first test on 42 days only helped 4 weeks, so it failed and the agent chose 45. The analyst approved the change.

More agents for finance and accounting specialists