AI cash flow forecasting: what AI adds to a 13-week forecast, with a worked £ example
Manoj Tutika · Published 16 November 2026 · Last checked 16 November 2026
Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.
What does AI add to a cash flow forecast?
A 13-week direct cash forecast lists expected receipts and payments week by week and shows the closing balance. The weak point is timing. Most forecasts assume customers pay on the due date and that the payables list is complete. Neither is true, and the gap is usually where the surprise comes from.
AI improves three things.
- Payment behaviour per customer. Instead of the due date, it uses each customer's history: Harlow Retail pays six days late on average, Northgate pays a month late, smaller customers mostly pay within a week of the due date. ICAEW's guidance asks for exactly this, expected, worst-case and best-case dates and amounts for each item, with collection probabilities for debtors.
- Recurring payables. It finds payments that repeat in the bank feed but have no invoice in the ledger yet: software subscriptions, rent, card fees, direct debits. These are the lines a manual forecast forgets.
- Scenarios. It makes best, most likely and worst cases cheap to run: what if the two largest customers pay 30 days late, what if a supplier insists on payment up front.
ICAEW recommends rolling, detailed, weekly cash flow forecasts for 13 weeks out when cash is tight, and scenarios for worst case, most likely case and best case (4 November 2022). AI does not change the method. It makes the method affordable every week.
What does that look like in pounds?
A UK company starts the month with £60,000 in the bank. It has £70,000 of open sales invoices and a normal month of payments.
| Customer | Amount | Due | History | Week by due date | Week by behaviour |
|---|---|---|---|---|---|
| Harlow Retail | £24,000 | Week 2 | Pays about 6 days late | Week 2 | Week 3 |
| Bexley Foods | £18,500 | Week 3 | Pays on time | Week 3 | Week 3 |
| Northgate Ltd | £12,000 | Week 2 | Pays about 30 days late | Week 2 | Week 6 |
| Smaller customers | £15,500 | Weeks 1 and 4 | Most pay within a week of the due date | Weeks 1 and 4 | Weeks 1, 2, 4 and 5 |
Payments are the same in both versions: £16,400 in week 1 (suppliers £14,000, software subscriptions £2,400 found in the bank feed), £38,000 in week 2 (payroll £31,000, rent £7,000), £9,500 in week 3 (PAYE and National Insurance to HMRC) and £16,000 in week 4 (suppliers).
| Week | Receipts by due date | Closing cash by due date | Receipts by behaviour | Closing cash by behaviour |
|---|---|---|---|---|
| Week 1 | £10,000 | £53,600 | £8,000 | £51,600 |
| Week 2 | £36,000 | £51,600 | £2,000 | £15,600 |
| Week 3 | £18,500 | £60,600 | £42,500 | £48,600 |
| Week 4 | £5,500 | £50,100 | £4,500 | £37,100 |
By due date, cash never drops below £50,000. By behaviour, it drops to £15,600 in week 2, the week payroll and rent go out, because Harlow and Northgate have not paid yet. The remaining £13,000 arrives in weeks 5 and 6. Same invoices, same payments, a very different conversation with the bank. That week 2 low point is what the forecast is for: it gives you time to chase Northgate now, move a supplier payment or draw on the overdraft before you need it.
How do you measure forecast accuracy?
AFP's best practices for cash forecasting start with agreeing what the forecast is for and how accuracy will be measured, then running variance analysis on a timely and consistent basis, focused on what is material to cash. In practice:
- Save every forecast. Keep each week's version, not just the latest, so you can compare forecast with actual.
- Measure error by horizon. Week 1 should be close; weeks 5 to 13 will be wider. Track actual minus forecast, and the absolute error as a percentage of actual, separately for week 1, weeks 2 to 4 and weeks 5 to 13.
- Split receipts and payments. Receipts are usually where the error is. Payments are mostly within your control.
- Check bias, not just size. If receipts come in below forecast eight weeks out of ten, the forecast is optimistic, and an AI model should be adjusted for that.
- Explain the big misses. For each material variance, record why: a customer paid late, an invoice was disputed, a payment was brought forward. Those reasons improve the next forecast.
ICAEW makes the same point from the other direction: compare forecasts with actual results to improve them, and have someone other than the preparer review the reports.
Excel or a live ledger?
| Question | Excel with an AI assistant | Forecast on a live ledger |
|---|---|---|
| How current is it? | As current as the last export | Updates as bank lines and invoices arrive |
| Payment behaviour | Only if you export months of history | Read directly from invoice and payment records |
| Recurring payables | Only what you list | Found in the bank feed |
| Scenarios | Easy, the assistant edits the model | Easy, run against current data |
| Accuracy tracking | Manual, if you save versions | Automatic, versions saved each week |
| Errors | ICAEW warns spreadsheets are highly prone to errors under stress | Fewer manual steps, but still reviewed |
Excel is a fine place to start. Claude for Excel and Copilot in Excel can build a 13-week template, explain the formulas and flex assumptions without breaking them. The trouble comes in week three, when nobody has refreshed the export and the forecast quietly goes stale. For Xero users, see cash flow forecasting for Xero; for the basics, see the 13-week cash flow forecast and budget vs cash flow forecast.
What do Thred's agents do with your cash forecast?
Thred is a team of AI agents that handle daily work for finance teams and accounting firms, on your existing accounting software. A cash forecast is only as good as the ledger it reads, so the agents work on both.
The Treasurer manages your cash: it keeps the 13-week view current from the ledger and bank, predicts receipts per customer from their payment history and flags the week cash runs tight. The Collector manages your receivables and chases Northgate before week 2, not after. The Payer manages your payables and times supplier runs around the low point. The Bookkeeper keeps the record up to date, so open invoices and bank lines are current every day. The Analyst monitors performance and variance, including forecast against actual. The Connector finds context in your connected tools, such as the email where a customer promised payment, and The CFO is your finance lead when you have none. The Accountant handles filings and optimises for taxes, so VAT and PAYE dates sit in the forecast.
On Exact Online the agents read and write directly, every write shown as an approval card until you set a rule. Agents for Xero and Moneybird work alongside. See Thred Layer.
Frequently asked questions
What is AI cash flow forecasting?
It is a cash forecast where AI predicts the timing of each receipt and payment from your own history, instead of assuming every invoice is paid on its due date. It also finds recurring payments, flags unusual items and recalculates the forecast as new bank lines and invoices arrive.
Can I do AI cash flow forecasting in Excel?
Yes, to a point. Assistants such as Claude for Excel or Copilot in Excel can build a 13-week template, explain formulas and run scenarios on the data in the workbook. The limit is the data: the workbook is only as current as the last export from your ledger, and payment behaviour needs months of invoice and receipt history.
What is the best AI cash flow forecast tool?
The best tool is the one that reads your live ledger and bank, predicts timing per customer and supplier, and shows you why it moved a receipt. Check that it saves each forecast version so you can measure accuracy, and that it works with your ledger, such as Xero or Exact Online.
How does AI help a 13-week cash flow forecast?
In three ways. It predicts when each customer will pay based on their history, it finds recurring payables such as subscriptions, rent and payroll that a manual forecast misses, and it makes scenarios cheap to run. ICAEW recommends rolling, detailed, weekly forecasts for 13 weeks out when cash is under pressure.
How accurate is AI cash flow prediction?
There is no single number; it depends on your customers, your data and the horizon. Measure it yourself: save each week's forecast, compare it with actual receipts and payments, and track the error separately for week 1, weeks 2 to 4 and weeks 5 to 13. Watch for bias as well as size.
What data does an AI cash forecast need?
Opening bank balances, open sales and purchase invoices with due dates, at least six to twelve months of invoice and payment history per customer, recurring payments from the bank, payroll and tax dates, and any known one-offs such as capital spend or loan repayments.
Sources
- ICAEW, Nine principles for finance professionals (4 November 2022, checked 16 November 2026)
- AFP, 10 Best Practices in Cash Forecasting (checked 16 November 2026)
- Anthropic, Use Claude for Excel (checked 16 November 2026)
Read next
- The 13-week cash flow forecast for small companies, and how to keep it current
- Cash flow forecasting for Xero: what Xero does, the add-ons, and how to keep a forecast current
- AI FP&A use cases: what AI does well in financial planning and analysis, and what it needs underneath
- Thred vs Float: a Float alternative that keeps your cash view current from your Xero books
- Budget vs cash flow forecast: the difference, side by side, with a profitable month that runs out of cash
Where Thred comes in
Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.
