By KYN AI Advisory Team — AI implementation specialists, Singapore
You can build a short-term cash flow forecast from three lists: open customer invoices, open supplier bills and recurring commitments such as payroll and rent. Place each item in the week you expect cash to move, add the opening bank balance, and the running total is your forecast. An agent keeps the inputs current and flags the weeks that go negative. A person owns the assumptions.
What is a direct cash flow forecast?
A direct forecast lists the cash you expect to receive and pay, week by week. It starts from what is already known: invoices you have issued, bills you have received, payroll and tax dates. It differs from a profit forecast, which includes accruals and non-cash items. A business can be profitable and short of cash in the same week.
A 13-week horizon is common because invoices and bills already in the system explain most of it. Beyond that, the forecast depends on sales you have not made, and certainty falls.
What inputs do you need?
| Input | Source | Timing question | |---|---|---| | Opening cash | Bank balances, reconciled | As at the start date | | Open invoices | Receivables ledger | When will each customer pay? | | Open bills | Payables ledger | When will you pay each one? | | Payroll and CPF | Payroll calendar | Fixed dates | | Rent, loan repayments, subscriptions | Commitments list | Fixed dates | | Taxes | Filing calendar | Statutory dates, from IRAS and your adviser | | Expected new sales and costs | Pipeline, budget | Estimates, shown separately |
Keep known items and estimates in separate lines. A reader should see how much of next month's cash rests on invoices already issued, and how much rests on hope.
How do you place invoices in the right week?
The due date is not the payment date. Customers pay early, on time and late, and the pattern differs by customer.
Use history. For each customer, take the last six paid invoices and calculate the days between the due date and the payment date. Use the median as that customer's expected delay.
Expected payment date = due date + customer's median delay
A customer who pays 12 days late on average lands 12 days after the due date. A customer with too little history gets a default delay that you choose and state. Disputed invoices go in a separate line with no date until resolved.
This depends on matched payments. If bank lines are not tied to invoices, the history is wrong. See bank reconciliation with an AI agent.
How do you place bills?
Bills are within your control, which makes them easier.
- Use the due date by default, or your payment run schedule: if you pay on the 15th and 30th, a bill due on the 18th goes out on the 30th.
- Mark bills you intend to delay and state the new date.
- Include bills approved but not yet received for known recurring suppliers, as estimates.
The capture side is covered in how an agent reads supplier bills from Gmail.
What does the forecast look like?
A simple weekly layout:
| Line | Week 1 | Week 2 | Week 3 | |---|---|---|---| | Opening cash | A | B | C | | Receipts from open invoices | + | + | + | | Payments of open bills | − | − | − | | Payroll and CPF | − | | | | Rent, loans, subscriptions | − | | − | | Tax payments | | | − | | Closing cash | B | C | D |
Add a minimum cash line: the balance below which you must act. Highlight the first week the closing cash falls under it. That week is the decision. The earlier you see it, the more options you have: chase a large invoice, move a payment run, draw on a facility, or talk to a supplier.
What does the agent do?
Rules keep the forecast current without anyone rebuilding a spreadsheet.
- Refresh daily. Pull the latest balances, invoices, bills and payroll dates.
- Recalculate expected dates from payment history.
- Show the change. What moved since yesterday: a customer paid early, a bill arrived, a large invoice slipped a week.
- Flag the weeks that cross the minimum cash line, and list the invoices that would fix it if paid on time.
- Send a short summary to the finance lead, for example on WhatsApp. The format is the number, the week and the change, not commentary.
- Log the forecast each day so you can compare it with what happened.
The agent does not move money, change terms or contact a customer by itself. It prepares the list of invoices to chase, and a person approves the reminders. See invoice chasing agent templates.
How do you know the forecast is any good?
Compare forecast with actual each week. Record, for each week, forecast receipts against actual receipts, and the same for payments. The differences show where your assumptions are weak: usually a few large customers, or an irregular tax payment. Adjust those assumptions, not the whole model.
Do not treat the output as a prediction. It is a view of what happens if the known items behave as they did before. It improves as the data does.
What does a six-week example look like?
Use a small company with S$40,000 in the bank (illustrative). The direct forecast lays out expected receipts and payments by week.
| Week (illustrative) | Receipts | Payments | Closing cash | |---|---|---|---| | 1 | S$18,000 | S$22,000 | S$36,000 | | 2 | S$9,000 | S$12,000 | S$33,000 | | 3 | S$14,000 | S$31,000 | S$16,000 | | 4 | S$22,000 | S$10,000 | S$28,000 | | 5 | S$6,000 | S$13,000 | S$21,000 | | 6 | S$15,000 | S$11,000 | S$25,000 |
Week 3 is the finding: payroll and a large supplier bill fall together and cash drops to S$16,000. With that in view the owner can move a supplier due date, ask one customer to pay early or schedule a transfer. Without the table, the same dip appears as a surprise on the day.
Show two versions: expected dates and a cautious version where every customer pays ten days late. If the cautious line stays above your minimum balance, you are safe. If not, act now.
How often should the forecast refresh?
Refresh it weekly, and again after any large event: a big invoice paid, a supplier bill added, a payroll run approved. A daily refresh of the opening balance is cheap if the bank feed is reliable. Keep each version so you can compare the forecast with what happened. Where the gap is large, look at the cause: a customer that pays late every month, a bill missing from the list, or a one-off item. Adjust the inputs, not the method. After three months you will know which customers need a longer assumed delay and which weeks carry a surprise.
Where KYN fits
KYN's own books keep receivables, bills and payroll in one system, with bills from Gmail, receivables chased at 3, 14 and 30 days and payroll on 2026 CPF rules. For a manufacturer, a 06:30 morning report goes to each department on WhatsApp at about S$0.05 per report. Systems run in a cloud account in the client's name, every action is logged with the actor, and money waits for a named person.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. See finance systems and reducing days sales outstanding.
Frequently asked questions
How far ahead should a small business forecast?
Thirteen weeks is a common horizon. Update it weekly, or daily if cash is tight.
Can a spreadsheet do this?
Yes, for a small number of invoices. It breaks down when data comes from several systems and needs daily refresh. The method is the same either way.
What if customers pay unpredictably?
Show a range: an expected case using median delay and a cautious case using a longer delay for your largest customers. Plan against the cautious case.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
