A practical, no-jargon walkthrough for building a forecast you can actually trust, without hiring an analyst.
You do not need finance software or a data science degree to get a usable cash flow forecast. Export your transactions and AR/AP aging, feed them to ChatGPT or Claude with a structured prompt, and ask for a best case, base case, and worst case. Sanity-check the output against what you actually know about your customers, then repeat it every Monday. That is the whole system.
I’ve sat in on enough finance reviews over the years to know exactly how this goes. Someone builds a cash flow tab a year or two back, wires in a few formulas, and for a while it works fine. Then a customer starts paying net-60 instead of net-30, a new vendor contract gets added, and nobody circles back to update the assumptions. Nobody decides to let the forecast rot, it just quietly stops matching the business while everyone is busy doing other things.
Pulling transactions, reconciling AR and AP aging, and rebuilding scenarios by hand eats hours most people do not have on a Monday morning, so the forecast only gets touched once there is already a scare. That is backwards, and I have watched it play out the same way more times than I can count. You want to see the shortfall three weeks out, not three days out, because by three days your options shrink to calling the bank or calling the client and asking nicely.
Cash flow problems, not a shortage of orders or revenue, are consistently cited as the top reason small and mid-size businesses run into trouble.[1] Honestly, I have yet to see a business go under because the owner lacked discipline. I have seen plenty go under because nobody had a spare afternoon to look three weeks ahead, which is a much easier problem to fix.
Let’s be honest about what AI is and is not good at here. It will not know that your biggest client mentioned they are switching vendors next quarter. It does not have a gut feeling about whether a slow month is seasonal or a real warning sign. That judgment is still yours.
What it is genuinely good at is the part that eats your Monday: pulling numbers together, spotting patterns across months of transactions, and generating multiple scenarios instead of one brittle guess. Paste in a spreadsheet export and ask a general model like ChatGPT or Claude to identify your recurring payables, your average days-to-collect by customer, and your seasonal spikes, and it will do in two minutes what used to take an afternoon.
Finance teams that still do this reconciliation by hand report spending a large chunk of their week on repetitive data entry and matching, time that could go toward actually deciding what to do about the numbers instead of just producing them.[2] That is the trade AI is actually offering: less time assembling the picture, more time acting on it.
This is the version that works whether you are using a general chat tool or dedicated forecasting software. Do it in this order.
Say you run a 12-person agency with $180,000 in monthly recurring revenue, but two of your biggest clients pay net-45 instead of net-30. You export the last four months of transactions plus your AR aging report and run the standing prompt.
The AI flags something you half-knew but had not quantified: those two clients alone account for 40% of your receivables, and their payment lag means a typical month shows healthy revenue on paper while your actual bank balance dips low around the 20th, right before payroll. The base-case scenario shows you covered. The worst-case scenario, where both clients pay a week late in the same month, shows a gap of roughly $9,000 against your payroll obligation.
That is the whole point of running scenarios instead of asking for a single number. A single forecast would have shown “healthy,” full stop. The worst-case scenario gives you three weeks of runway to move money from a reserve account, ask one client to pay a partial invoice early, or simply know it is coming instead of getting blindsided on payroll day.
You have two real options, and the right one depends on how often you need this.
General AI tools (ChatGPT, Claude). If you are forecasting for one business and you are comfortable exporting and uploading a spreadsheet, this is the cheapest and most flexible path. You already have a subscription, there is nothing new to learn, and you can ask follow-up questions in plain English (“what happens if that client pays two weeks late every month?”). The tradeoff is that you are doing the data export and re-upload manually each time.
Dedicated forecasting tools (Float, Fathom, Pulse, or your accounting software’s built-in AI features). These connect directly to your bank feed and accounting system, so the forecast updates itself without you exporting anything. Worth it once you are checking this more than weekly, managing multiple entities, or the manual export step is the thing stopping you from actually doing this consistently. Expect a real monthly cost for this tier, not a free add-on.[1]
AI forecasts are only as honest as the data and the prompt behind them. A few specific ways this goes sideways:
None of this means skip AI forecasting. It means treat the output the way you would treat a sharp but new analyst’s first draft: useful, fast, and still worth a second pair of eyes before it goes in front of your bank or your board.
The forecasting method matters less than whether you actually repeat it. This version is small enough to survive a busy week:
Minutes 1 to 5: export last week’s transactions and your current AR/AP aging report from your accounting software.
Minutes 5 to 12: upload to your AI tool with your standing prompt (see the workflow above). Skim the output for anything that looks off.
Minutes 12 to 18: add anything the AI could not know: a client conversation, a delayed invoice, a new expense you already committed to.
Minutes 18 to 20: note the worst-case number somewhere your team actually sees it. Not because you expect it, but because you want zero surprise if it starts trending that way.
This matters because the forecast number was never really the point. The lead time is the point. A cash gap you spot five weeks out is one phone call to a vendor about payment terms. The same gap spotted three days out is a genuine crisis, and I have sat across the table from founders living through both versions of that conversation.
No, and you should not try to make it. Your bookkeeper or accountant understands the specific context of your business: which clients are reliable, which expenses are one-time versus recurring, and what is coming that has not hit the books yet. AI is genuinely useful for the mechanical part, pulling patterns out of transaction data fast, but the judgment layer still needs a human who knows your business.
At minimum, 3 months of transaction history and a current accounts receivable and accounts payable aging report. More history helps the AI spot seasonality, but even a rough 3-month picture beats no forecast at all. If you only have a bank statement export, start there and add the aging reports once you can pull them.
For most small businesses forecasting a single entity, a general AI tool like ChatGPT or Claude with a good prompt is genuinely accurate enough, provided you are giving it clean data and checking the output. Dedicated forecasting software earns its cost once you are managing multiple entities, need the forecast to update automatically from a live bank feed, or are checking it more than weekly.
Weekly, at minimum, for any business with tight margins or lumpy revenue. Monthly is the bare minimum for anyone. The value of AI here is that a weekly routine that used to take half a day now takes about 20 minutes, so there is not much excuse to fall back to quarterly.
Yes, but you need more historical data for it to be useful, ideally at least one full seasonal cycle, and you need to say explicitly in your prompt that the business is seasonal so the AI weighs recent months correctly instead of assuming a flat trend. If this is your first season using AI forecasting, treat the output as a rough guide rather than gospel until you have a cycle of data to compare it against.
This guide draws on how Future Factors trains finance and operations teams to build practical AI workflows, plus current industry reporting on AI-assisted cash flow forecasting for small and mid-size businesses.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.