Decide which decision the forecast supports, then show the timing, source and owner behind each material number. A clear scenario is more useful than unexplained precision.
Can’t AI just predict the cash position?
A prediction you can’t trace is hard to act on. Keep the model checkable.
1. Define the question.
Scope. Choose the horizon, update frequency, bank accounts, entity and currency.
Build from cash. Use opening cash, expected receipts and expected payments. Don’t substitute accounting profit for cash movement.
Practical check: Choose an interval that reveals the decisions your team needs to make.
2. Due date and expected date aren’t always the same.
For each material flow. Record the amount, expected date, supporting source and last update.
Name an owner. Who confirmed the assumption, and what would change it?
Avoid double counting. Check that the same order, invoice or settlement isn’t included through two sources.
Practical check: Label uncertain receipts instead of making the forecast look certain.
A fictional example: the final balance can hide the gap.
A CA$20,000 receipt moves from week 1 to week 2. Amounts are in CA$000.
- Week 1, base
- 40 opening + 30 receipts − 60 payments = 10.
- Week 1, delayed
- 40 opening + 10 receipts − 60 payments = −10.
- Week 2, base
- 10 opening + 40 receipts − 35 payments = 15.
- Week 2, delayed
- −10 opening + 60 receipts − 35 payments = 15.
- Result
- Both end at 15. The delayed case still passes through a shortfall in week 1.
3. Where AI could help.
Prepare the inputs. Organize approved updates, identify conflicting dates and draft a summary of what changed.
Keep it checkable. Use explicit calculations and documented assumptions. A prediction is not a payment commitment.
Keep authority separate. The forecast doesn’t authorize borrowing, changing terms or releasing payments.
Practical check: Select the simplest model that supports the decision.
4. Compare the forecast with what happened.
Reconcile. Opening cash and actual receipts and payments, to the appropriate records.
Explain. Separate amount differences, timing changes and missing or duplicated flows.
Update. Correct the assumptions with their owners and keep the previous forecast for comparison.
Practical check: Judge the forecast at the horizon it was meant to support.
Read more: AI Variance Commentary: Is the Explanation Supported? · AI for Collections: Check the Account Before the Reminder
The example is fictional. This is general guidance, not financial advice.