How to prepare a thirteen-week cash flow forecast with an AI agent

A thirteen-week cash flow forecast projects receipts and payments week by week for a quarter, from actual cash rather than accounting profit. The agent builds it from the ledgers, payment patterns and known commitments, rolls it every week and explains last week's variance. The treasurer or CFO approves the assumptions and signs each version, especially when lenders receive it.

What is a thirteen-week forecast, and who owns it?

The thirteen-week forecast is a direct cash forecast: opening cash, plus expected receipts, minus expected payments, week by week. It is the standard tool when cash is tight, during a refinancing, or when lenders require it as a condition of a facility. It is also simply good practice for any business that wants to see a quarter ahead.

The CFO or treasurer owns it. In stressed situations, an adviser to the lenders may review it every week, and the business may have agreed to report variances against it. That makes accuracy and a clear audit trail more important than polish. The underlying question of whether an agent can prepare a cash flow forecast at all is answered there; this playbook is the method.

What does the agent prepare?

1. Agent, depth 1: sets opening cash. From bank balances at the start of the week, reconciled to the daily cash position.
2. Agent, depth 2: projects receipts. By customer, from the open receivables ledger adjusted for each customer's observed payment behaviour, plus expected new sales from the approved forecast converted to cash on typical terms.
3. Agent, depth 2: projects payments. Supplier payments from open payables and the payment run calendar; payroll and payroll taxes on their dates; rent, loan interest and repayments, tax payments, insurance and capital spend from contracts and schedules.
4. Agent, depth 1: lists irregular items. Quarterly and annual payments, one-off settlements and anything above a set amount, each with its source.
5. Person, CFO or treasurer: sets and approves assumptions. Collection rates for at-risk customers, payment stretching if any, new borrowing or asset sales.
6. Agent, depth 2: builds the forecast. Weekly, with closing cash and headroom against facility limits.
7. Agent, depth 2: prepares the variance report. Last week's forecast against actual, by line, with the reason for each material difference.
8. Person, CFO: signs the version. And sends it to lenders where required.

What does a variance line look like?

Take an illustrative manufacturer that forecast receipts of 600,000 in week five and received 470,000. The agent's variance report traces the 130,000: one large customer, forecast at 100,000, paid nothing because of a disputed delivery, and two smaller customers paid 30,000 later than their usual pattern. It moves the disputed amount out of the forecast until the dispute is resolved, flags it as an assumption the CFO must decide, and keeps the two late payers in week six based on their remittance advice.

What does the CFO approve, and what evidence remains?

The CFO approves the assumptions that make the forecast a judgement rather than a calculation: which customers will pay late, whether suppliers will accept slower payment, what financing will arrive and when. Each is recorded with its owner and reason.

Source for every line. Ledger, contract, schedule or assumption. Evidence: a source column on each line.

Behaviour, not terms. Receipts are timed by how each customer actually pays. Evidence: the payment pattern used per customer.

Weekly variance. Evidence: the variance report, signed alongside the new version.

Frozen versions. Each signed version is kept unchanged. Evidence: version history with approver and recipients.

Treasury system link. Where expected flows are loaded as entries, as in preparing cash positions in your treasury system, the forecast and the system should agree for the first week.

Where does it usually go wrong?

Receipts on due dates. Customers pay late, and receipts on due dates overstate cash every week until the forecast collapses. Use behaviour.

Irregular payments forgotten. A quarterly tax payment or annual insurance premium arrives in week seven and nobody planned for it. The agent's irregular item list prevents this.

No variance discipline. The forecast rolls each week without anyone asking why last week was wrong. Lenders notice. Variance reasons must be specific: "customer A paid 50,000 in week 3, not week 2", not "timing".

Optimism in assumptions. Collection of a disputed debt is assumed in week four. Mark assumption-based receipts separately, so readers see how much rests on them.

Surplus treated as spare. Headroom in week two is not spare cash if week nine is tight. Decisions on placing surplus, as in whether an agent can invest surplus cash, should look across the whole horizon.

What should you measure?

• Forecast accuracy for week one and week four, by line.
• Cumulative variance over the horizon.
• Assumption-based receipts as a share of total, shown as an amount.
• Days from week end to signed version.

Where Pryme fits

In Pryme, projecting receipts, projecting payments and preparing variances are separate capabilities, certified on your own past weeks of actual cash. Ledger data arrives sealed and the CFO's assumptions as declared, so lenders and the board can see which part of the forecast is data and which is judgement. Each signed version is filed. The finance bounds are in AI agents for finance.