Forecast accuracy baseline
Current vs. actual variance over the last 12 weeks, by entity, by category. Where the model (or the spreadsheet) is consistently wrong — and where it's accidentally right.
For corporate treasurers · Cashflow & liquidity
Cashflow forecasting & liquidity planning
From a spreadsheet nobody trusts to a forecast the board acts on. Diagnostic, integration and build-out — delivered alongside your team, not behind a curtain.
Most cashflow projects fail at the data layer long before the model. Subsidiaries produce 13-week numbers under pressure, each with a different risk appetite. Spreadsheets multiply. The board sees a number it doesn't challenge because it can't trace it. Even a 10% accuracy uplift is worth real money at scale — but you only get there by fixing the data and the process, not just adding a model on top.
Step 1 · Diagnostic
Where is the forecast actually breaking?
Two to three weeks, fixed scope. A hands-on diagnostic of accuracy, data, coverage and governance. Delivered as a written report with a prioritised action list — not a slide deck and a proposal for another engagement.
Goal: establish the baseline and tell you honestly whether ML is what you need next, or whether a process fix would move the number further, faster.
Data availability & tagging
Bank feeds, AR/AP, payroll, treasury movements. What's machine-readable, what's still manual, what's tagged to a chart-of-accounts a model can use.
Coverage & risk-appetite gaps
Which entities are forecast, which are inferred, which are guessed. How aggressive vs. conservative the underlying assumptions are — and whether that's policy or personality.
Governance & sign-off chain
Who owns the number, who reviews, who challenges. The path from a subsidiary submission to a board-pack figure, and where breaks happen.
Step 2 · Integration
Fix the data, then the model
The boring 80% of the work is where the accuracy gain actually comes from. We stand up the data spine, tag the history, engineer the features and validate the baseline so the model has a chance.
- 01
Systems connected
Bank APIs, ERP, TMS and payroll routed into the treasury data spine. No more exports, no more monthly refreshes — the forecast sits on a live substrate.
- 02
Data cleaning & tagging
The 80% of the work nobody owns. Dedupe, standardise, tag to a stable chart of accounts, resolve entity / currency / account breaks, make the history defensible.
- 03
Feature engineering
The variables a model can actually reason about — day-of-week, month-end seasonality, counterparty payment patterns, large-and-unusual flags. Separates what ML should forecast from what judgement should.
- 04
Scaling & validation
Feature scaling so one outlier doesn't dominate the model. Training / validation / testing split, explicit measures of precision, recall and MAPE. A baseline everyone trusts.
Step 3 · Build-out
Short-range, medium-range, explained and owned
ML handles the regular, explainable flows. Judgement handles the one-offs — M&A, dividends, bond coupons. The forecasting stack knows which is which, and so does the team.
Short-range: 13-week
Rolling 13-week direct cashflow forecast per entity, with ML-driven category-level estimates and judgement overlays for the one-offs (M&A, dividends, bond coupons). Live liquidity buffer tracking against policy.
Medium-range: 3-to-12 month
Indirect forecast driven by P&L assumptions, working-capital swings and financing decisions. Used to stress-test covenant headroom and shape the financing roadmap.
Variance & commentary
AI-drafted variance explanations that tie the forecast back to the numbers, in your tone of voice. The board pack writes the first draft of itself.
Handover & enablement
Runbooks, notebooks, model cards and training so the team can interrogate the forecast themselves — no black box, no permanent consultant dependency.
Proven
The approach that has already taken the awards home
Adam Smith AwardsAI cashflow forecasting — Treasury Today
TMI AwardsTechnology innovation in treasury
Economist ImpactAI in corporate finance programme
Start a conversation
Ready to put your treasury to work?
Book a 30-minute diagnostic call. We'll tell you within the hour whether we can help, and where the biggest wins likely sit.
Or email us directly at [email protected]