You invested in a TMS to get treasury out of spreadsheets. Yet for many teams, forecasting is still the last major workflow living in Excel.
Cash positioning, payments, and bank connectivity run through the system. Somewhere between the ERP extract, bank file, and the regional team's emailed assumptions, the process quietly drifts back into Excel and the forecast quietly becomes a parallel universe. Two sources of truth, one of them held together by tabs and tribal knowledge.
This isn't a criticism of TMS platforms. Most were designed around visibility, controls, payments, and workflows. Forecasting was often bolted on later.
What actually breaks in spreadsheet forecasting
It starts with the data. A 13-week forecast pulls from your TMS, your banks, your ERP, sometimes your billing system. Then you layer on inputs received from other teams or subsidiaries. Then, bring in debt servicing, FX settlements, intercompany activity and other treasury owned inputs. Each of those exports lands in a different format. Someone has to flatten them, align headers, scrub duplicates, and paste them into the master workbook before any forecasting happens. By the time the inputs are ready, the data is already stale.
Then come the formulas. Every forecast model is a stack of nested calculations one delete-key away from disaster. Anyone who has owned a forecast has experienced this: the model worked perfectly until one formula got overwritten six tabs deep. In treasury, the worst ones are silent. You don't see a broken VLOOKUP until the CFO does.
Version control is its own disaster. Forecast_v7_FINAL_FINAL.xlsx is not a joke, it's a liability. With four people editing the same workbook in different windows, somebody overwrites somebody else's update every week. Nobody is sure which file is the truth. Reconciling versions replaces reconciling transactions.
Then there's human risk. A typo in row 412. A formula dragged one cell too far. A hardcoded number from last quarter that nobody remembered to clear. The forecast is only as accurate as the most tired analyst on a Monday afternoon. And when that analyst leaves, the model leaves with them.
Read: Manual vs Automated treasury forecasting: what's the real cost?
How does AI forecasting work?
The goal isn't to automate Excel. The goal is to stop building forecasting processes around Excel altogether. It's a different shape of system. One that learns your business, explains its answers, and gets better every week.
Data flows in continuously from your TMS, banks, and ERP, no exports, no pasting. And those spreadsheet submitted forecasts? One thing large language models are unexpectedly good at: dealing with messy inputs. The spreadsheet forecast submitted by Germany never looks like the one from APAC. AI can normalize those differences without forcing every team into the same rigid template. Instead of asking treasury to manually encode assumptions, the system starts with observed behavior: customer collections, payment cycles, seasonality, supplier timing, recurring treasury activity and the dozens of small signals a spreadsheet flattens into a single average. The model updates daily, not quarterly.
You stay in control of the inputs. Configure recurring flows the model can't infer, like payroll, and scheduled tax payments, and they're folded into the baseline automatically. Layer manual entries on top for one-off events: a planned acquisition, a board-approved bonus, a tax settlement. Mark outliers, and those transactions are ignored when the model retrains. Every adjustment is tracked, attributed, and visible. Nothing happens in a black box, and every forecast is more accurate than the one before, as it continuously learns your business.
Variance Analysis closes the loop. Each week you see actuals against forecast with the drivers of the gap surfaced automatically. No more end-of-quarter post-mortems trying to remember what you assumed three months ago.
The bottom line
Your TMS solved cash visibility. AI forecasting closes the gap it left open, on top of the TMS you have, with no manual layer to break and a forecast the business can actually trust.
Spreadsheets are remarkably resilient. Most treasury teams get a lot out of them. But complexity compounds: more entities, more systems, more assumptions, more stakeholders. Eventually the process becomes harder to maintain than the forecast itself. The question becomes less "Can Excel do this?" and more "Should treasury still be operating this way?"
