Cash forecasting agent

What it does

For every open AR invoice, the agent looks at:

• The customer's historical days-to-pay.
• Recent payment trends (slowing? accelerating?).
• Currency, payment terms, and credit memos.

It produces a daily/weekly cash receipt forecast for treasury, with a confidence band, and writes a short narrative explaining why the forecast moved since last week.

Why it matters

Most cash forecasts are still spreadsheet-driven aggregates of payment terms — which assume customers actually pay on terms. They don't. Modeling each customer's actual behavior is the kind of thing humans don't have time for but ML is well-suited to.

What changes for treasury

• A forecast that updates daily instead of monthly.
• An explanation of changes ("Customer Acme is paying ~7 days slower this quarter — accounts for $1.2M of the variance") instead of a flat number.
• More time on liquidity strategy, less on rebuilding the model from scratch each cycle.

Where it can mislead

• Customers with very few invoices get a wide confidence band — treat the forecast as a range, not a point.
• One-off events (a major customer's M&A activity, a regulatory change) aren't in the training data and won't be reflected. Override the forecast in those cases.
• New customers get the segment average until they have 3+ paid invoices.