AI for finance in Business Central

Accounts receivable

An ageing report tells you what is overdue. It does not tell you which of those balances is a genuine problem, and that distinction is where the analysis time goes today. An assistant reading customer ledger entries can separate the two: a customer 40 days late who is always 40 days late is a payment terms question, not a collections question. A customer who has paid on time for two years and is now 30 days late is the one worth a call.

• Ageing with payment history as context, not just a bucket
• Change in payment behaviour — the signal that usually precedes a bad debt
• Overdue balances cross-referenced with open orders, so you know what to hold
• Credit limit utilisation, and who is approaching it quietly
• Disputed items separated from slow payers, where your setup records that distinction

Our article on receivables control covers the underlying process; the AI layer changes who can ask the question, not what the right answer is.

Accounts payable and cash flow

The payables side is mostly about timing: which invoices are due, which have discount terms worth taking, where payment could be deferred without damaging a supplier relationship. Combined with receivables and open orders, that produces a cash position projection that reflects your actual documents rather than a spreadsheet built in January.

One caution worth stating. A cash flow forecast is only as good as the data behind it, and in most companies some commitments live outside Business Central — payroll timing, tax payments, loan schedules. An assistant that projects from ledger data alone will be confidently short of the full picture unless those inputs are included deliberately. We identify them in discovery rather than discover them in production.

Margin and profitability

The most valuable finance question and the one with the most setup dependency: why did margin move? Answering it means decomposing the change into purchase price movement, discount given, mix shift, and cost adjustments — which requires value entries, consistent dimensions and an agreed definition of margin.

That agreement is not a formality. Business Central can produce at least three defensible margin figures depending on whether you use cost at posting, adjusted cost after the cost adjustment routine has run, or standard cost. If nobody specifies which one, the assistant picks one and disagrees with your management report — and the finance team concludes the tool is broken. Fixing the definition costs an hour at design time and a great deal of credibility afterwards.

Anomaly detection

Postings that do not look like the others: an unusual amount for a vendor, a posting to an account that rarely receives one, a document posted outside normal hours, duplicate invoice numbers from the same supplier. None of these is proof of anything — they are a list worth a human look, which is exactly what a controller wants and rarely has time to produce.

Set expectations honestly here. This is pattern detection, not fraud detection. It surfaces candidates and produces false positives, and its value is in shortening a review, not in replacing one.

What it needs from your Business Central

Three things determine whether finance analysis will work well, and all three are worth checking before committing to a project:

• Dimension consistency. Analysis by department, project or region needs those dimensions filled in reliably. A rule introduced part-way through means analysis crossing that date is wrong in a way that looks correct — see dimensions in Business Central.
• Posting group discipline. If similar transactions land in different accounts depending on who entered them, aggregate figures will not reconcile. Our note on posting groups covers why.
• Cost adjustment running regularly. Margin figures before cost adjustment are provisional. The assistant should say so rather than report them silently.

Where these are not in place, the honest recommendation is to fix the setup first. It is cheaper than building analysis on top of it and produces better reports regardless of whether AI is involved.

From analysis to action

Everything above is read-only, which is where we suggest starting. A finance AI agent extends it: drafting collection reminders for review, preparing payment proposals, submitting approval requests. Payment and credit decisions themselves stay with a person — not because the technology cannot execute them, but because that is where commercial judgement lives.