Key Takeaways
- AI forecasting platforms like Transformance’s CashPulse forecast each entity from its own customers’ payment behavior, then roll every entity into one group cash position in the reporting currency, with intercompany flows eliminated.
- CashPulse is built around the three reasons group forecasts break: data spread across several systems, a separate sub-model per entity, and a refresh cycle slower than cash moves.
- Cash and liquidity forecasting is treasury’s most frequently cited challenge, named by 49% of practitioners (AFP, 2026).
- Each horizon gets its own model, from 7 days to 365 days, and the champion model per horizon is selected on accuracy and stability.
- CashPulse retrains weekly and reaches 90-95% forecast accuracy out to 90 days (Transformance platform data, 2026).
In This Article
- Key Takeaways
- Why is cash forecasting harder for multi-entity groups?
- What goes wrong when every entity builds its own forecast?
- Why does a quarterly forecast go stale in week three?
- Step 1: How do you get data from every ERP, bank and treasury system into one place?
- Step 2: Why does each forecast horizon need its own model?
- Step 3: How do you keep entity logic but get one group number?
- Step 4: How do scenarios work on a group forecast?
- Step 5: How do you stop the model going stale without remodeling by hand?
- Where should the forecast go once it is built?
- How do you measure whether the group forecast is getting better?
Why is cash forecasting harder for multi-entity groups?
Data sits in several ERPs, banks and a treasury system, each entity builds its own sub-model, and the group model refreshes slower than cash moves; Transformance’s CashPulse closes these gaps.
Why multi-entity forecasts break, and what fixes it
Cash and liquidity forecasting remains treasury’s most frequently cited challenge, named by 49% of the 425 treasury practitioners in the 2026 AFP Treasury Benchmarking Survey (AFP, 2026). EY India’s 3 September 2026 release states that a mature treasury function may manage between 50 and 100 interconnected spreadsheets, and that forecast variance in spreadsheet-led treasury environments often exceeds 20% (EY, 2026).
A group adds layers a single company does not have. Each legal entity has its own ERP instance or company code, its own banks and its own customers. Intercompany loans, cash pooling sweeps, management fees and internal trade appear as an inflow in one entity and an outflow in another. And the group reports in one currency while its entities collect and pay in several.
Takeaway: group forecasts fail on structure (data, logic, timing), not on a missing formula.
What goes wrong when every entity builds its own forecast?
Sub-models drift apart in logic, FX rates and cut-off days, so the roll-up adds incompatible numbers, and when the group total is wrong nobody can trace which entity caused it.
Harmonization. One entity assumes customers pay on terms; another uses last quarter’s average days to pay. One converts at month-end spot, another at the budget rate. One cuts off on Friday, another on the last business day. A customer buying from two entities is forecast two different ways. Intercompany receivables left in make group liquidity look larger than it is.
Correction. The error surfaces at group level weeks later, when the actual cash position misses the forecast. Finding the cause means reopening every sub-model.
Ownership. A sub-model built by one analyst is hard for anyone else to read, so group treasury signs off numbers it cannot check.
Takeaway: a group forecast is only as consistent as its least consistent sub-model.
Why does a quarterly forecast go stale in week three?
Cash moves daily and weekly with payment timing, disputes, FX and one-off payments, so a model built at quarter start describes conditions that no longer hold long before the rebuild.
A refresh in a group is a project: exports from each ERP and bank portal, currency conversion, intercompany elimination, then model updates entity by entity. That takes weeks, so most groups do it quarterly.
Inside the period, reality moves. A large customer starts paying ten days late, a disputed invoice holds a payment, the dollar moves against the euro, a tax payment lands that was never modeled. Teams patch the growing gap with manual overrides, and the next rebuild starts from those undocumented overrides. The general causes of forecast error are in why most cash flow forecasts are inaccurate; in a group, each one is multiplied by the number of entities.
Takeaway: a forecast refreshed less often than cash moves records past assumptions.
Step 1: How do you get data from every ERP, bank and treasury system into one place?
Connect each system once to one data layer, map entities, currencies and intercompany relations there, and refresh it daily instead of exporting by hand.
Transformance’s integrations hub connects to the ERP each entity runs, plus bank feeds, the treasury system and external data sources. If a system has an API, a database, a file drop or a mailbox, it can be connected. Routes include APIs, OData, RFC and BAPI, files, mailboxes and an on-premise agent that connects outbound only, with no inbound firewall change.
SAP S/4HANA, SAP ECC, NetSuite, Microsoft Dynamics 365 and Odoo are in production with customers; Oracle and Infor are supported. Entities on different ERPs run in one Transformance environment with one forecast (finance automation across several ERPs). ERP-specific guides: SAP, NetSuite, Dynamics 365.
Transformance builds, runs, monitors and fixes the connections. Integration agents read the system documentation, propose field mappings and run the test plan; a person approves every change that reaches a customer system or production (how an agent-run integration platform works). A production ERP connection typically takes 2-3 days (Transformance customer data, 2026), and your IT team may be involved.
What flows in: live AR data and payment history, AP commitments such as purchase orders, payroll and tax schedules, and market signals such as FX rates and commodity prices.
Takeaway: connect once, refresh daily, and the forecast stops depending on who ran the exports.
Step 2: Why does each forecast horizon need its own model?
A 7-day cash position, a 13-week liquidity view and a 12-month plan answer different questions from different drivers, so one model tuned for all of them is wrong for most.
Which horizon answers which question
The families are typical, not fixed: CashPulse selects the model per horizon and portfolio on your data.
Short horizons run on known items such as open invoices, payment runs and payroll dates, where an auto-regressive model fits stable cycles. Mid horizons depend on how each customer actually pays, which behavior-driven machine learning captures. Long horizons for FX- or commodity-exposed portfolios depend on external signals and run on deep learning. The 13-week cash flow forecast sits in the middle, which is why a single model so often gets it wrong.
CashPulse tests many model configurations in parallel and selects the champion model on accuracy and stability, measured per horizon.
Takeaway: pick the model for the question, not one model for every question.
Step 3: How do you keep entity logic but get one group number?
Forecast each entity from its own customers’ payment behavior, then convert and net every entity into one reporting-currency position, intercompany flows eliminated, with drill-down to entity and customer.
In CashPulse, each entity’s forecast reflects its own customer payment behavior, not a global average. Flows are netted at the legal-entity level with intercompany flows eliminated, so treasury sees net cash, not gross inflows. Every entity and currency is then converted and netted into one group position, including the cash you cannot move. One model runs across 20+ countries with 100+ drivers.
You filter by legal entity, region or reporting currency and drill down from the group total to individual forecasts and customers. That keeps FX risk visible where it sits, not hidden in a translated total.
Illustrative example (fictional figures). A German entity on SAP expects a USD 650,000 intercompany loan repayment from its US sister company on NetSuite. Forecast separately, Germany shows the inflow in week six and the US the outflow in week seven, each at its own rate, so the group total carries a gap that is not real cash. With intercompany flows eliminated, the repayment drops out and only external cash is converted. The cash-side matching is covered in intercompany and multi-currency cash application.
Takeaway: keep the logic at entity level and the elimination in the system.
Step 4: How do scenarios work on a group forecast?
Scenarios layer assumptions on the AI baseline (FX shifts, payment delays, seasonality, planned investments, external events) and compare base, optimistic and stress cases side by side.
In the CashPulse scenario planner, the baseline is the AI forecast, and each scenario layers assumptions on top of it. Typical group questions: what if the euro weakens against the dollar, what if our largest customers pay later this quarter, can the group fund a planned investment in one entity, and what does an external event such as a supply disruption do to inflows?
Scenarios are linked to actions: if we accelerate AR collections on the top 20 overdue accounts, how does the 30-day forecast change? Every forecast ships with three lines: best case (P90), expected (P50) and risk-adjusted (P10). Plan against P10 for liquidity floors and covenant compliance.
Takeaway: a scenario is a changed assumption on a live baseline, not a second spreadsheet.
Step 5: How do you stop the model going stale without remodeling by hand?
Measure accuracy per horizon continuously and retrain automatically: weekly retraining cycles, and a rebuild when accuracy drops below a threshold, so treasury does not have to remodel.
CashPulse measures per-horizon accuracy, forecast bias and per-customer hit rate continuously. Models retrain weekly, and when accuracy drops below the threshold, the model is rebuilt before treasury has to step in.
Transformance’s ML Studio builds new forecasting models quickly and retrains them when accuracy drops. In a group that matters: an acquisition adds an entity, a business unit changes its payment terms, a large customer changes how it pays. The model follows instead of waiting for the next quarterly rebuild.
The result is 90-95% forecast accuracy out to 90 days (Transformance platform data, 2026). Accuracy depends on data volume and payment history; see how accurate AI cash flow forecasting is.
Takeaway: retraining that runs itself keeps a group forecast current between reviews.
Where should the forecast go once it is built?
Into the tools treasury already uses: CashPulse feeds forecasts into your existing treasury system or ERP, or stays self-contained, and AR data flows in from cash application and AR collections.
The output is a net cash curve by entity, currency and category, with payment-level predictions and confidence bands. The input side is where accuracy comes from: matched payments from ClearMatch (cash application), active disputes from ClaimIQ (deductions) and promise-to-pay dates from CollectPulse (AR collections). Standalone deployment is possible, but the signal is cleaner with cash application or AR collections upstream.
When the 30-day position falls short, the collections scenario points to accounts your team can work today (how agentic AR collections connect the forecast to action). Data stays in your region, VPC deployment in your own cloud is available, and every forecast change is logged.
Takeaway: feed the treasury tools you have and the AR work that moves the number.
How do you measure whether the group forecast is getting better?
Track accuracy per horizon and per entity, forecast bias, and the share of variance you can explain, every week rather than once per quarter.
- Accuracy per horizon and entity. CashPulse measures accuracy as MAPE on the rolling P50 forecast against actual payment outcomes as invoices close. A group average hides the entity that drags it down.
- Bias. A forecast above actuals week after week usually means one entity’s collection assumptions are too generous.
- Explained variance. For each miss, name the cause: a late customer, a dispute, an FX move, a one-off. Drill-down makes that a lookup, not an investigation.
The first 13-week forecast comes in 2-4 weeks, full rollout takes up to 4-8 weeks, and a proof of value runs 4-5 weeks on one entity (Transformance customer data, 2026). Most enterprises deploy with around three years of payment history; stable portfolios can start with one.
Takeaway: measure weekly, per entity, and the forecast shows where it needs work.
Who Transformance is for
- ERPs: SAP S/4HANA and SAP ECC, NetSuite, Microsoft Dynamics 365 Finance & Operations and Business Central, and Odoo in production with customers; Oracle and Infor supported; connects to the ERP a customer runs through its integrations hub, including legacy, on-premise and mid-market ERPs.
- Size: mid-market to large enterprise, 250 to 2,000+ employees, including groups with several entities and shared service centers.
- Payment volume: from under 1,000 to over 100,000 payments a month.
- Industries: manufacturing, consumer goods and food and beverage, logistics, e-commerce, agriculture, energy and professional services.
- Scope: cash application alone, or the full order-to-cash (cash application, deductions, AR collections, cash forecasting).
Conclusion: one data layer, one model per horizon, one group number
Group cash forecasts break on structure: scattered data, sub-models that drift apart and a refresh cycle slower than cash. Transformance’s CashPulse connects every system to one data layer, forecasts each horizon with the model that fits it, rolls entities up to one reporting-currency position, puts scenarios on top and retrains when accuracy drops. Book a Call to see CashPulse on your own entities.
Key terms
- Cash flow forecasting: predicting future cash inflows and outflows to see the cash position ahead.
- Rolling forecast: a forecast that moves forward with every update, keeping the horizon the same length.
- 13-week cash flow forecast: a weekly cash view one quarter ahead, used for liquidity headroom.
- Cash pooling: concentrating group entities’ balances so cash can be used across the group.
- FX risk: the exposure of cash flows and balances to exchange-rate changes.
Related guides
- Which ERP systems does Transformance integrate with?
- Finance automation across several ERPs
- Why most cash flow forecasts are inaccurate
- How accurate is AI cash flow forecasting?
- Cash flow forecasting from SAP, Oracle and NetSuite data
Frequently Asked Questions
What is multi-entity cash flow forecasting?
It is a forecast built per legal entity and rolled up into one group cash position in a reporting currency, with intercompany flows eliminated.
Why are group cash forecasts often inaccurate?
Because entity sub-models use different logic, the data is collected by hand from several systems and the model is refreshed less often than cash moves.
Can one forecast cover entities on different ERPs?
Yes: CashPulse consolidates forecasts across entities, currencies and ERPs in one reporting currency, with drill-down to each entity and customer.
How often should a cash forecasting model be retrained?
Continuously measured and at least weekly; CashPulse retrains weekly and rebuilds a model when its accuracy drops below a threshold.
How accurate is AI cash forecasting for a group?
Transformance measures 90-95% forecast accuracy out to 90 days (Transformance platform data, 2026); accuracy depends on data volume and payment history.
How long until a group gets its first forecast?
The first 13-week forecast comes in 2-4 weeks and a full rollout takes up to 4-8 weeks (Transformance customer data, 2026). A production ERP connection typically takes 2-3 days (Transformance customer data, 2026); your IT team may be involved.
What does multi-entity cash forecasting software cost?
It depends on scope and the number of entities, for example cash forecasting alone or with cash application and AR collections. Book a Call and we scope it with you.
SAP, SAP S/4HANA, Microsoft, Dynamics 365, Business Central, Oracle, NetSuite, Odoo and Infor are trademarks of their respective owners (SAP SE, Microsoft Corporation, Oracle Corporation, Odoo S.A., Infor). Transformance is not affiliated with or endorsed by them.




