A cash flow forecast is a week-by-week or month-by-month estimate of the money that will actually arrive in and leave your bank account, so you can see a shortfall weeks before it happens. AI helps by drafting the forecast from bank and invoice exports, estimating when each customer really pays and running what-if scenarios, but you still own the assumptions.
The idea people most often get wrong is treating profit as cash. A business can be profitable on paper and still run out of money, because it pays suppliers before customers pay it. A forecast is the tool that shows that gap, and it's only as good as its timing assumptions, which is exactly where AI is both most useful and most likely to mislead.
Profit and cash are different numbers
Take an illustrative nine-person import-export business that buys industrial valves from an overseas manufacturer and sells them to trade customers. One container order looks like this:
| Week | Event | Cash | Running cash position on this order |
|---|---|---|---|
| 1 | Order placed; 30% deposit paid | -$18,000 | -$18,000 |
| 6 | Goods ready; 70% balance paid before shipment | -$42,000 | -$60,000 |
| 12 | Container arrives; freight and duties paid | -$7,000 | -$67,000 |
| 13-14 | Goods sold and invoiced for $95,000 on 60-day terms | $0 | -$67,000 |
| 22-24 | Customers pay (some late) | +$95,000 | +$28,000 |
The accounts will show a $28,000 profit on this order. The bank account shows $67,000 going out and nothing coming back for about five months. Run two or three orders like that at once and a profitable business can't pay its wages. That's what a forecast is for: not predicting the future precisely, but showing when the timing bites so you can arrange cover in advance.
The four lines every forecast has
However it's laid out, every cash flow forecast has the same skeleton for each week or month:
- Opening balance: cash in the bank at the start of the period (the previous period's closing balance).
- Cash in: money you expect to receive, mostly customer payments, plus things like loan drawdowns or asset sales.
- Cash out: money you expect to pay: suppliers, wages, rent, loan repayments, tax payments, freight and anything else that leaves the account.
- Closing balance: opening plus in minus out. This is the number you watch.
Two choices shape the rest. The horizon: a 13-week forecast, updated weekly, is the working tool for businesses where cash is tight or lumpy; a 12-month forecast, by month, is for planning hires, investment and borrowing. Many businesses keep both. And the method: the "direct" method lists actual receipts and payments, which is what owners need week to week; the "indirect" method starts from profit and adjusts for timing, which accountants use for longer-range and statutory work. Everything below uses the direct method. For a full build of the weekly version, see building a 13-week cash flow forecast with AI help.
A six-week forecast, line by line
Here's the same import-export business's forecast for the next six weeks, with a $25,000 overdraft available:
| Wk 1 | Wk 2 | Wk 3 | Wk 4 | Wk 5 | Wk 6 | |
|---|---|---|---|---|---|---|
| Opening balance | 42,000 | 58,700 | 68,200 | 46,400 | 21,700 | 44,300 |
| Customer receipts | 31,000 | 18,500 | 22,000 | 12,000 | 27,500 | 19,000 |
| Supplier payments | -8,000 | 0 | -42,000 | -6,500 | 0 | -18,000 |
| Freight and duties | 0 | -7,200 | 0 | 0 | -3,100 | 0 |
| Wages | 0 | 0 | 0 | -26,000 | 0 | 0 |
| Rent and overheads | -6,300 | -1,800 | -1,800 | -1,800 | -1,800 | -1,800 |
| Loan repayment | 0 | 0 | 0 | -2,400 | 0 | 0 |
| Closing balance | 58,700 | 68,200 | 46,400 | 21,700 | 44,300 | 43,500 |
On these assumptions nothing goes wrong, though week 4 (payroll plus a supplier payment plus the loan) is clearly the pinch. The forecast's value shows when one assumption moves. If the $22,000 expected from the largest customer in week 3 arrives three weeks late, week 3 closes at $24,400 and week 4 at minus $300: payroll week, in the overdraft. Nothing about the business's profitability changed; one customer's timing did. Seeing that on a Monday in week 1 gives the owner three weeks to chase, ask the supplier for a week's grace, or confirm the overdraft.
Building your first forecast in an afternoon
You don't need special software to start. A spreadsheet and three exports (six months of bank transactions, your list of unpaid customer invoices and your list of unpaid supplier bills) are enough. Ask an AI assistant to build the empty structure first, then fill it with your own figures:
Create a 13-week cash flow forecast layout for a spreadsheet. Rows:
opening balance; customer receipts; other receipts; supplier payments;
freight and duties; wages; rent and overheads; loan repayments; tax
payments; other payments; closing balance; overdraft limit; headroom.
Columns: week commencing dates starting Monday [date]. Give me the
formulas for closing balance and headroom, and nothing else.Then work down the rows: unpaid invoices go into receipts in the week you expect them (using example 2 below for timing), unpaid bills into payments by their due dates, wages and rent on their fixed dates, and everything else from the bank history. The first version takes two to three hours; most of that is deciding when each receipt will really arrive. Label every figure you estimated rather than knew, so you can see which assumptions to watch.
Five ways AI speeds up cash flow forecasting
1. Sorting bank history into forecast categories
A forecast built on history needs your past bank lines grouped into the categories above. Export six months of bank transactions and ask a business-plan AI assistant to categorise them:
Categorise each bank line into: customer receipts, supplier payments,
freight and duties, wages, rent and overheads, loan, internal transfer,
annual or quarterly bill, other. Add a column "recurring?" (weekly,
monthly, quarterly, annual, one-off). Flag anything you're unsure of.An illustrative extract of the output: "CARRIER INV 5521: freight and duties, monthly-ish; FX TRANSFER TO USD ACC: internal transfer; INSURANCE PREMIUM: annual bill (next due in March?); REFUND C-0209: customer receipts, one-off, unsure." Two fixes are typical. Transfers between your own accounts, including currency accounts, must be excluded from cash in and out, or they double-count. And the "annual bill" lines are gold: the AI has found the payments you'd forget, and each needs a real next-due date from the paperwork. Categorising transactions with AI and checking its work covers this step in detail.
2. Working out when customers really pay
Most forecasts assume invoices are paid on their due date. Most customers don't pay that way. Upload a year of invoices with issue dates, due dates and paid dates (customer names replaced by codes) and ask for each customer's actual payment behaviour:
| Customer | Terms | Average days to pay | Range | Invoices |
|---|---|---|---|---|
| C-0118 | 30 days | 47 | 38-61 | 14 |
| C-0342 | 60 days | 58 | 55-63 | 9 |
| C-0077 | 30 days | 33 | 29-41 | 22 |
| C-0209 | 30 days | 72 | 44-110 | 6 (2 disputed excluded) |
That table changes the forecast: C-0118's invoices belong 47 days after issue, not 30, and C-0209 needs a cautious assumption or a conversation. Check two or three customers by hand against your ledger, and make sure disputed and part-paid invoices are handled consistently. The same data feeds predicting late payers and acting early.
3. Turning supplier terms into payment dates
For an importer, the biggest payments sit in purchase orders and pro-forma invoices, in sentences like "30% deposit with order, balance against copy of shipping documents, estimated ready date 14 November". Paste the terms and dates for open orders into the assistant and ask for a table of expected payment dates and amounts, with the assumption behind each. An illustrative line: "PO-3187: $42,000 balance, expected week of 17 November, assumes goods ready on estimated date; if production slips, payment slips with it." That last clause is the useful part: it tells you which dates depend on someone else's schedule.
4. Running what-if scenarios
Once the base forecast is in a spreadsheet, AI is quick at stress-testing it. Share the forecast and ask:
Using this 13-week forecast, show the weekly closing balance under three
scenarios: (a) our two largest customers each pay 3 weeks late; (b) freight
and duties rise 25%; (c) the November container is delayed 4 weeks, so its
balance payment moves but its sales also move. List the lowest balance in
each scenario and the week it happens. Show the changed rows.An illustrative answer: "(a) lowest balance minus $18,600 in week 4; (b) lowest $17,900 in week 4, a small effect; (c) lowest $31,200 in week 9, but cash from those sales moves out of the 13-week window." The insight in scenario (a) is that customer timing matters far more than cost inflation for this business, which tells the owner where to spend effort: credit control, not freight negotiations. Always check one scenario's changed rows by hand, because models sometimes shift a receipt without shifting the matching cost.
5. Explaining forecast against actual each week
Every week, replace the forecast figures with what actually happened and ask the AI to list the differences over a threshold. An illustrative note: "Receipts $9,500 below forecast: C-0118 paid $12,000 of $21,500 expected; C-0077 paid on time. Payments $2,100 above forecast: unplanned courier charges." Keep the model to what changed and by how much; the why is yours to find out, usually with a phone call. Explaining budget-versus-actual variances with AI applies the same discipline to monthly accounts.
Forecasts already built into accounting software
Before building anything, look at what your accounting package already does. Xero's short-term cash flow view projects your position 7 or 30 days ahead from bank balances, outstanding invoices and bills, including repeating ones, and lets you adjust dates and add one-off amounts; its Analytics Plus add-on, available on some plans, extends projections to 90 days. QuickBooks Online's Cash Flow Planner forecasts 90 days from your history and lets you add future events without touching your books. Both assume a lot less about customer behaviour than a custom forecast, but for a business with steady income they may be all you need. If you use Xero, Xero's JAX assistant for invoices and cash-flow questions shows how to query this data in plain English, and AI cash flow forecasting tools compared covers the dedicated apps.
What AI gets wrong in a cash forecast
A realistic example of how these errors show up: an illustrative packaging supplier asked an assistant to "build a cash flow forecast from these unpaid invoices and bills" and got a tidy table with every receipt on its due date and a comfortable balance throughout. Three weeks later the account was overdrawn on payroll day. Two large customers were paying at around 50 days on 30-day terms, as they had all year, and the forecast had no way to know. The fix took ten minutes: the days-to-pay table from example 2, applied to each customer's invoices. The error wasn't the AI's arithmetic; it was an assumption nobody had checked.
- Everyone pays on time. Unless you give it payment history, a model assumes due dates. Spot it: receipts in the forecast bunch on the 30th day after each invoice.
- Irregular bills vanish. Insurance, annual software licences, quarterly tax payments and equipment servicing don't appear in a three-month history. Spot it: a forecast month with no unusual payments at all. Keep a list of known irregular bills with next-due dates and add them by hand.
- One-offs become recurring. A large one-time receipt, such as an insurance payout or asset sale, gets projected every month. Spot it: income that looks better than any real month last year.
- Currencies get mixed. An importer paying suppliers in one currency and selling in another needs every line in one currency at a stated, cautious rate. Spot it: supplier payments that look suspiciously small.
- Growth gets invented. Ask for "a realistic forecast" and some models add a few per cent a month. Spot it: every line creeping up. Tell it to project only what you give it.
A 20-minute weekly routine that keeps the forecast honest
- Monday, 5 minutes: export last week's bank lines and replace last week's forecast figures with actuals.
- 5 minutes: ask the AI for the variance note (example 5 above) and read it.
- 5 minutes: update the assumptions that changed: a customer who promised payment, a container that slipped, a new order placed.
- 3 minutes: roll the forecast forward a week, adding the new week 13.
- 2 minutes: look at the lowest closing balance in the next eight weeks. If it's below your comfort level, decide one action today.
The routine matters more than the tool. A simple spreadsheet updated every Monday beats a sophisticated model updated whenever someone remembers.
What a 12-month version looks like for a seasonal business
The weekly forecast shows next month's pinch; the monthly one shows next season's. For an illustrative packaging supplier whose sales peak from October to December, the monthly view makes the problem obvious:
| Aug | Sep | Oct | Nov | Dec | Jan | |
|---|---|---|---|---|---|---|
| Customer receipts | 88,000 | 92,000 | 104,000 | 131,000 | 158,000 | 149,000 |
| Board and materials | -71,000 | -96,000 | -102,000 | -84,000 | -52,000 | -47,000 |
| Wages and overtime | -31,000 | -33,000 | -38,000 | -41,000 | -41,000 | -32,000 |
| Other costs | -12,000 | -12,000 | -13,000 | -13,000 | -13,000 | -12,000 |
| Net cash flow | -26,000 | -49,000 | -49,000 | -7,000 | +52,000 | +58,000 |
The business spends $131,000 more than it receives from August to November, building stock and paying overtime for sales it won't be paid for until December and January. The year is profitable; the autumn is a cash hole. Seeing that in June gives the owner time to arrange a seasonal facility, negotiate longer terms with the board supplier, or ask the biggest customers for part-payment on large orders. AI can spot the seasonal shape from two or three years of monthly sales, but the purchase plan behind the materials line comes from the owner and the production manager.
Which kind of forecast suits which business
| Business (illustrative) | Cash pattern | Forecast to keep | Where AI helps most |
|---|---|---|---|
| Import-export business | Large supplier deposits months before sales cash | 13 weeks, weekly, plus 12 months monthly | Supplier terms into dates; scenarios |
| Packaging supplier | Seasonal peaks; raw material bought ahead of the peak | 12 months monthly, switching to weekly before the peak | Seasonal patterns from sales history |
| Testing laboratory | Steady monthly billing, some slow-paying clients | Monthly, 12 months | Customer days-to-pay |
| Courier firm | Weekly wages and fuel, customers paying monthly | 13 weeks, weekly | Categorising fuel and repair spending |
Whichever you choose, start with the four lines, add your real payment timings, and update it weekly. AI makes each of those steps faster; it can't make the forecast right on its own. The judgement about which customer will really pay and which supplier will really ship on time is still the part that keeps a profitable business solvent.
Cash flow forecast questions owners ask
How often should I update a cash flow forecast?
Weekly if cash is tight or you have large, lumpy payments such as supplier deposits or quarterly bills; monthly if cash is comfortable and your income is steady. Each update replaces forecast figures for the week just gone with actuals, rolls the forecast forward a week, and revisits the assumptions that changed. A forecast updated once and left alone is out of date within a fortnight.
Can I give an AI-built forecast to my bank or lender?
You can give them a forecast you built with AI help, as long as you understand and stand behind every assumption. Lenders will ask why receipts arrive when they do and what happens if they don't. Keep the forecast in your own spreadsheet, list the assumptions on a separate tab, and have your accountant review it before it goes to anyone who will lend against it.
What should I do if the forecast shows a shortfall?
Act early, which is the whole point of forecasting. The usual levers, roughly in order of speed: chase overdue invoices, ask good customers to pay on time or early, talk to suppliers about moving a payment, delay non-essential spending, and arrange or extend an overdraft or facility before you need it. Banks respond far better to a request made weeks ahead with a forecast than to one made the day before payroll.
Further reads
- How to Chase Late Payments With AI Reminders That Sound Human — The fastest lever when a forecast shows a gap.
- How to Use AI to Understand Your Profit and Loss Statement — The profit side of the profit-versus-cash story.
- Break-Even Analysis With AI: A Worked Example for a New Product — Check a new product pays before it hits your cash flow.
- How to Prepare a Business Loan Application With AI Help — Use your forecast in a loan or facility application.
- Quote to Cash: Connect Quotes, Invoices and Payments With AI — Shorten the time from quote to cash in the bank.
- Can AI Do My Bookkeeping? What Still Needs an Accountant — Clean books make every forecast more reliable.
- How to Use AI for Scenario Planning: Best, Worst and Likely Cases — Build best, worst and likely cases from your own numbers, use AI to challenge the assumptions, and turn each case into triggers and pre-agreed actions.
- Chasing Unpaid Invoices With AI: Polite Reminders for Trades — A reminder ladder for trade invoices, where to switch it on in Xero, QuickBooks, Jobber and Housecall Pro, and AI prompts for the awkward replies.
- Can AI Help a Charity Forecast Demand for Its Services? — Use Excel's Forecast Sheet or a chat assistant to turn two years of service counts into a winter range, then add what the model can't see.
- How to Forecast Next Quarter's Sales With AI Using Your History — Three baselines, a backtest that exposes over-confident models, an adjustments log, and a wine merchant's festive quarter forecast worked end to end.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Xero pages on short-term cash flow and Analytics Plus; QuickBooks pages on Cash Flow Planner; facts sheet for plan names. The business, figures, customer codes and AI outputs are illustrative.