Build your cash flow forecast from your project pipeline — not your bank history. Three numbers: receivables due, invoices outstanding, projected inflows by month.
You've got a model deployment wrapping up next month, two retainers running, and a new build in scoping. The bank account looks fine today. Ask what it looks like in six weeks — after you've paid the team and before that deployment invoice clears — and the honest answer is: you're not sure.
That's the cash flow problem. Not a shortage. An absence of visibility.
The real reason most forecasts get abandoned
Most founders who've tried cashflow forecasting built it from accounting history. Past transactions. What came in last month. That's a rearview mirror.
It doesn't show you that the enterprise client on net-60 terms won't pay until the 14th. Or that the milestone trigger for the RAG pipeline is tied to a client sign-off that keeps slipping. Or that inference costs doubled this month because a client went live at scale.
A cash flow projection built from transactions tells you what happened. One built from your project pipeline tells you what's coming.
The 3 numbers AI agencies need to track
1. Receivables due
Invoices already sent. Not yet paid. What's owed, by whom, by when. This number is sitting in your inbox or your billing tool. You just haven't pulled it into a single view.
2. Invoices outstanding
Work completed but not yet invoiced. Deployments at 80%. Milestones hit but not yet triggered. Prompt library delivery done, sign-off pending. Real money you've earned and haven't claimed.
3. Projected inflows by month
Work in progress, mapped to the payment trigger. Not "this project is worth $40K." What you need is: "$12K at kickoff (received), $16K at dev completion (expected next month), $12K at deployment (month three)." That third number is where most AI agencies have the most uncaptured visibility.
How to actually build it
Step 1 — List every active project and retainer. For retainers: monthly amount + expected payment date. For projects: every remaining milestone, the trigger condition (client approval, deployment, eval completion), and the expected date.
Step 2 — Pull all open invoices. Every sent invoice, with its status. Due date. How many days overdue.
Step 3 — Map the next 90 days. Three columns: Month 1, Month 2, Month 3. Confirmed inflows: retainer payments, invoices at their due date. Conditional inflows: milestones pending client sign-off, eval set reviews, go-live approvals.
Step 4 — Lay your fixed costs against it. Payroll. Contractor fees. API and model inference costs. Anything that clears whether or not a client pays.
"Your P&L might show $80,000 in revenue for the month. The enterprise client is on net-60 terms. The RAG pipeline milestone hasn't triggered yet. That cash isn't here. Payroll is."
One place to track it
The Finance + Invoices module in Agency OS tracks every invoice by status — Draft, Sent, Due, Overdue, Paid — linked to the client record and project. Your receivables due and outstanding invoices are always visible. MRR is calculated automatically from active client records.
When invoice status is accurate, the forecast almost builds itself.
Why this structure works for AI agencies specifically
AI project billing is uneven by design. A model deployment might pay 40% at kickoff, 40% at deployment, 20% at a 30-day performance review. An eval set build pays per phase. An inference-cost retainer might carry variable overage billing.
If those triggers are tracked in a project record and linked to invoice status, you can see the pipeline — not just the bank balance.