Every finance team already knows the opening balance. The hard part is when the money actually arrives, and when it has to leave.
A customer who always pays ten days late is not the same as one who pays on the due date. Payroll, GST, TDS and a loan instalment do not move because a spreadsheet hoped they would. Most cash forecasts still drop every open invoice onto its due date, then get surprised in week six.
A Cash Forecast agent closes that gap. Not a chatbot that writes a plausible cash story. An agent that times each open invoice from that customer's own payment history, places the outflows you already know, and shows a range for every week, including the chance the balance falls below the minimum you set. The default horizon is 13 weeks. This is the difference between finding a shortfall on the morning it happens and seeing it while you can still act.
Stop thinking of it as a cash formula. Think of it as a treasury analyst that rebuilds the next 13 weeks from how your customers and suppliers actually behave.
A Cash Forecast agent, built the right way, gives you:
Including the hard cases: invoices already overdue, customers with a short history, and payments that arrive in two parts.
P10, P50 and P90 for receipts, payments and closing cash, so you see the likely week and the weeks that can miss.
Payables by your own days-to-pay or the due date, plus payroll, recurring items, loan repayments and statutory dates you supply. The agent never computes interest.
You set the floor. The agent names the earliest risky week and the probability the balance goes under it in any week.
Ask which customers and which bills move that week. The answer is the drivers, not a paragraph of guesswork.
Delay one customer, assume they pay on time, slow a supplier, or add an inflow. The comparison uses the same paths, so the difference is the scenario.
MAPE, WAPE, bias and a tracking signal, with a timing correction carried forward so the next run starts from what the last one got wrong.
This is what separates a cash forecast agent from a due-date spreadsheet: it does not put every invoice on the day it is due. It uses how that customer has paid before, keeps invoices that are still open in the picture, and shows a range. It never tells you to borrow, invest, or change a rate. Your treasury team decides.
This is built for the people who find out about a cash gap when the balance is already short.
If you're looking for advice on borrowing, investing, or interest, this isn't that, and it won't pretend to be. It forecasts timing and balances from the rows you send. It doesn't connect to your bank. It doesn't post entries, and it doesn't tell you what to do with a gap. This is for people who want a checkable 13-week range, not a chatbot that invents a cash story.
| Generic AI Agents / Chatbots | Us | |
|---|---|---|
| Consistency | Different answer each time you ask — probabilistic, hard to trust twice | Deterministic — same input, same output, every time |
| Cost to run | Re-processes your entire dataset through an LLM on every query | Purpose-built models, minimal token usage — cheap enough to run on the open book; pipelines larger than 500 leads can score locally |
| Reliability | A 0–100 score invented in the chat — confidently wrong when the closed sample is thin | Fits conversion from your outcomes, shows a range when it should, and refuses to fit when there isn't enough data — honesty instead of a bluff |
| Data privacy | Your data often leaves your systems, sits with a third party | Full SOC 2 compliance — no data ever leaves your premises |
In short: Other agents give you an answer. This gives you conversion probabilities, money at risk, and a daily queue you can defend, reproduce, and put in front of your security team without a fight.
We didn't build new models and hope your customers pay on time. Everything underneath is established statistics, applied to the way money actually arrives and leaves.
| What It Does | In Use Since |
|---|---|
| Times each open invoice from how late that customer has paid before, and keeps invoices that are still open in the curve — Kaplan-Meier survival on payment delay | 1958 |
| Builds many paths of receipt timing and reports a range for receipts, payments and closing cash — Monte Carlo (Metropolis-Ulam) with a Gaussian copula | 1949 |
| Compares a previous forecast with what actually arrived, and carries a timing correction into the next run — Trigg tracking signal, with MAPE, WAPE and bias | 1964 |
Why this matters to you: every layer here has been published and used on real data for decades. We didn't ask a language model to guess next quarter's cash. Customers with only a few paid invoices borrow their segment's curve, weighted by how much history they have. Overdue invoices use the chance they pay in each remaining week, given how late they already are. The agent shows a range, and it tracks its own accuracy. It never computes interest, and it never advises you to borrow or invest. Your results depend on the history you send and the outflows you include.
That's what a range you can re-run actually means. Not a marketing word. A math word.
ChatGPT and other LLMs are built to talk. Ask them when a customer will pay and they generate the most plausible-sounding week, which is exactly why they're brilliant at explaining cash flow and unreliable at timing thousands of invoices from history.
Here's the practical difference:
In Short: ChatGPT can explain what a 13-week forecast is. It can't reliably tell you which week the balance is most likely to break your minimum, and why. That's the gap this agent exists to close.
In plain terms, it takes your open invoices, how those customers have paid before, and the outflows you already know, and does three things:
That's the whole point of the agent: it turns a hoped-for due-date total into a week you can plan around.
No. If you can export open invoices and a payment history from your accounting software, you already have the core of what the agent needs, plus an opening balance and the outflows you want included.
Just follow the step-by-step tutorials we provide, and you're up and running in minutes: no models to train yourself, no code to write. There's a free tier you can try without a sales call. The default horizon is 13 weeks. You can ask for a different number of weeks when you need it.
The short version: if you can export a spreadsheet, you can have this running today. Follow the step-by-step tutorial.
Yes. This is built to be fully SOC 2 compliant, and your data never leaves your premises.
The agent is stateless: your data is processed in the request and never stored, and each user signs in with their own OAuth login. It never connects to your bank. You send exports, and nothing is held. Customer names of sole proprietors count as personal data, so references are pseudonymised where possible. This was built with data protection as a requirement from day one, not an afterthought.
The short version: you get the intelligence, not the exposure.
It's based on well-established methods that have been researched, tested, and used for decades, not a new algorithm hoped to work.
The agent looks at three things: how late each customer has paid relative to the due date, the outflows and dates you supply, and many paths of that timing so each week is a range. Open invoices stay in the picture. Customers with little history borrow their segment's curve.
The short version: it isn't guessing from patterns in text, like a general AI chatbot would. It's calculating, using methods that have had decades to prove themselves.
Non-deterministic means: ask the same question twice, get two different answers. This is how most general AI chatbots work. They generate the most likely-sounding response each time, even when nothing about your data changed. Fine for drafting a note to the board. Risky when the answer decides which week you might run short.
Deterministic means: same data in, same answer out, every single time. If a week changes, it's because your invoices, your history, or your outflows changed, not because the model rolled different dice.
What we use: the same inputs give the same weeks, every time. The ranges this agent gives you come from fixed methods with a fixed seed, not a language model guessing at what sounds tight. That's what lets you show a week to a controller and know it holds up if someone re-runs it tomorrow.
The short version: we don't give you a week you have to hope is right. We give you the same, re-runnable range, every time you ask.
No. It never advises you to borrow, invest, or change a rate, and it never computes interest. It shows the weeks, the range, and the chance of falling below the minimum you set. Your treasury team decides what to do about a gap.
In Short: it times the cash, and your team stays in charge of the decision.
Ready to See the Tight Weeks Before They Arrive?
You already have the data. Your open invoices, payment history and known outflows have everything this agent needs; you just need to know how to connect it. We'll walk you through exactly how, step by step.
How to Build an AI Agent →No credit card. No treasury-systems project. Just your exports and a few minutes.