NineAnts · Cash forecasting
Cash flow forecasting for treasury teams
AI cash flow forecasting uses your historical transactions and known future cash flows to predict where your cash will be, by category, entity and currency. NineAnts builds a continuous, bottom-up forecast from the data already in your banks, ERP and TMS, picks a forecasting model per cash flow category, and measures its own accuracy against actuals.
- Forecasting model chosen for each cash flow category
- Per category
- Rolling view, plus daily and monthly
- 13 weeks
- Of clean history we recommend to start
- 24 months
- Actuals compared with forecast
- Every day
The problem
Why cash flow forecasting is still hard
Ask a CFO's favourite question: where will our cash be in thirteen weeks? In most groups the answer starts with a round of emails. Treasury asks every subsidiary for its numbers. Half respond on time. The files arrive in different formats and currencies, someone forgets the tax payment, and the team spends two weeks consolidating a forecast that is out of date by the time it reaches the board.
The cost of that is concrete. Surplus cash that nobody saw coming sits in a current account. A shortfall that nobody saw coming gets funded at emergency rates. Currency mismatches go unhedged because the timing of the cross-currency flows was never clear.
Most treasury teams know this. The problem is rarely the model. It is the data underneath it: bank statements from several banks in several formats, transactions without a category, ERP invoices whose due date is not the date the customer actually pays, and a spreadsheet that only one person fully understands.
How NineAnts does it
How NineAnts forecasts your cash
NineAnts treats forecasting as a system, not a single model. It starts with the data foundation, adds what you already know about the future, and lets AI fill the gap between the two.
Start from your actuals
Every forecast starts with history. NineAnts connects to your banks through API or SFTP and collects balances and transactions every day. Each transaction gets a cash flow category through rule-based categorisation, so the model learns from customer receipts, supplier payments, salaries and taxes, not from raw bank lines.
We recommend at least 24 months of clean, categorised history. That is the minimum to see seasonality year on year: the tax months, the Q4 receipts peak, the August dip in supplier payments.
A forecasting model per category
Customer receipts with hundreds of transactions a month behave very differently from capex with five payments a year. NineAnts does not force one model onto every category.
The Treasury AI Agent first classifies each category as stable, seasonal, trending or volatile. It then picks the model that fits: time-series models such as Prophet for categories with strong seasonality and enough volume, recurring-pattern detection for stable flows such as rent, schedule-based models for sparse payments such as quarterly taxes or dividends, and moving averages where data is limited. You can see which model was chosen for each category and why.
The analysis runs fresh every time a forecast is requested, so when a stable category turns volatile the agent notices on its next run. There is no model to retrain and redeploy.
Add what you already know
Your bank data tells you what happened. Your business systems tell you what will happen. NineAnts brings in known cash flows from your ERP, TMS, files and manual inputs: open invoices, payroll runs, tax obligations, lease and loan payments. These form the baseline of the forecast. For the near term they are often quite accurate.
Different teams across the organisation can add their own cash flow inputs, so local knowledge ends up in the group forecast without a round of emails.
Structural change or one-off?
Numbers alone cannot tell you whether a jump in rent is a new lease or a one-time deposit. The agent reasons about what kind of cash flow it is looking at and whether a change is structural or temporary. When the signal is ambiguous, it raises the change with the treasurer instead of guessing.
Actuals vs forecast, continuously
Every day new actuals arrive and NineAnts compares them with what was forecast, at category and horizon level. You see where the forecast drifts, by how much, and why. Patterns in the errors themselves show up too, such as a category that is consistently over-forecast.
Each forecast is stored as a versioned snapshot that can be locked once finalised. Comparing versions over time shows whether your forecasts converge as the period approaches, and which categories need attention.
Scenarios and the 13-week view
Look at the forecast day by day for liquidity, week by week in the 13-week view, or month by month for board reporting. NineAnts stores data at the most granular level and aggregates up, so switching views loses nothing.
For what-if questions, ask the Treasury AI Agent to apply a scenario to your forecast, such as customer receipts 20% lower next quarter or a major supplier moving from 30 to 60 days. The agent recalculates on top of your base case, so you can compare the two.
What you get
A forecast your team can act on
- One continuous, bottom-up cash flow forecast across entities, banks and currencies
- A forecasting model chosen per cash flow category, with the choice explained
- Known cash flows from ERP, TMS, files and manual input as the baseline
- Contributions from local teams in one shared forecast
- Actuals vs forecast at category and horizon level, measured continuously
- Forecast versions you can lock and compare
- What-if scenarios applied by the Treasury AI Agent on request
- Daily, 13-week and monthly views from the same data
- Early sight of excess cash, funding gaps and FX exposure, through the Treasury AI Agent
Built on clean data
A forecast is only as good as its categories
- historical transactions learned from
- 181,031
- categorised automatically, up from 58%
- 89%
- less manual categorisation work
- 67%
A forecast built on badly categorised actuals inherits every one of those errors. That is why every NineAnts implementation starts with the client's history.
In one recent implementation, our agents learned a treasury team's categorisation logic from 181,031 transactions they had already categorised by hand and in Excel. The resulting rule set categorised 89% of transactions automatically, up from 58% with the team's Excel formulas, and agreed with the team's own historical choices 99.3% of the time.
One platform
How it connects
Forecasting sits on top of the rest of NineAnts. Each module feeds the next.
- Feeds it
Data automation
Puts your bank data and forecast in the Excel workbooks and Power BI reports your team already uses.
Learn more - Sets its start
Cash visibility
Today's cash position across every bank, entity and currency: the starting point of every forecast.
Learn more - Shapes it
Transaction categorisation
Turns raw bank lines into the cash flow categories the forecast is built from.
Learn more - You are here
Cash flow forecasting
A bottom-up forecast per category, measured against actuals.
- Acts on it
Treasury AI Agent
Reads the forecast and flags what needs action: excess cash, shortages, FX exposure.
Learn more
NineAnts sits on top of your existing TMS and ERP. It does not replace them. Want the full method? Read our guide on how to do AI cash flow forecasting.
Built security first
Frequently asked questions
Straight answers to the questions treasurers ask us
What is AI cash flow forecasting?
AI cash flow forecasting predicts future cash positions from historical transactions and known future cash flows. Instead of one spreadsheet model for everything, an AI system analyses each cash flow category separately, detects seasonality, trends and recurring patterns, and picks a suitable forecasting model per category. NineAnts adds continuous actuals vs forecast measurement, so the forecast improves with every cycle.
What is the best AI tool for cash flow forecasting?
It depends on your situation. If you already run a traditional TMS and it covers your needs, adding a forecasting module there may be enough. If your data is spread across several banks, entities and systems, NineAnts builds a bottom-up forecast on top of what you have and picks a model per category. We are happy to give an honest view on which route fits you best.
How accurate is AI cash flow forecasting?
Accuracy depends mostly on the quality and length of your history and on the category. Near-term forecasts built on known cash flows are usually more accurate than forecasts a year out, and stable categories forecast better than volatile ones. NineAnts measures actuals vs forecast continuously at category and horizon level, so you always see how accurate each part of your forecast is.
How much historical data do I need for AI cash flow forecasting?
We recommend at least 24 months of categorised transaction history. That is the minimum to see seasonality year on year, such as tax months or a peak in customer receipts in Q4. Thirty-six months is better, because more history makes it easier to separate real patterns from one-off events. NineAnts can help categorise that history during implementation.
Can NineAnts produce a 13-week cash flow forecast?
Yes. NineAnts stores cash flows at daily level and aggregates them, so the same forecast can be viewed day by day for liquidity, as a 13-week forecast for short-term planning, or monthly for board reporting. Because the forecast is continuous, the 13-week view rolls forward as new actuals and known cash flows come in.
Does AI cash flow forecasting replace the treasury team?
No. NineAnts does the data work and the analysis, and treasury stays in control. The AI provides a baseline per category, and your team adds the context the data cannot show, such as a planned acquisition or a customer you are about to lose. Every AI-generated entry is marked, so you always know which numbers came from the AI.
Is cash flow forecasting included in every NineAnts plan?
Cash flow forecasting is part of the Enterprise plan, which starts from €24,000 per year and also includes the Treasury AI Agent. The Starter and Professional plans cover data connectivity, automation and delivery to Excel, Power BI and other systems.
Related reading
Blog
Aug 31, 2026 · 3 min read
67% less manual categorisation
During a recent NineAnts implementation, our categorisation engine learned a treasury team's own logic from 181,031 historical transactions and cut their manual categorisation work to roughly a third.
Read the articlePodcast
Episode 297 · Nikos Polymenakos · 51 min
How Bad Financial Decisions and Poor Cash Management Are Destroying Businesses
Nikos Polymenakos lived through Greece's decade of bankruptcies and capital controls. He explains why companies that could not read their own balance sheet went under, and why AI belongs at the end of a data journey rather than the start.
Listen to the episodePodcast
Episode 293 · Marianna Polykrati · 38 min
AI in Treasury for Finance Leaders to Boost Forecast Accuracy by Cleaning Data
Marianna Polykrati automates everything she can, but not the cash flow forecast. She explains why automating a bad process only produces bad results faster, and what she learned connecting bank APIs in Greece.
Listen to the episode
See your own cash forecast in NineAnts
Book a 30-minute session. Expect questions about your banks, entities and current forecasting process. If NineAnts fits, we show you how. If it does not, we tell you.