The verdict in three sentences
A mid-market company with 120 million EUR in revenue that builds its 13-week cash flow forecast by hand ties up its treasurer one day per week for an average gap of 15% between forecast and actuals. Two paths exist in 2026: a cash management SaaS at 1,000 to 4,000 EUR per month, or a custom predictive model at 25,000 to 60,000 EUR excl. VAT, connected to EBICS bank statements and the ERP. Either way, the realistic target is an error brought down to about 5% and one day per week freed up for analysis.
What makes manual forecasting fragile
In an industrial or distribution Mittelstand company, the 13-week forecast pulls together heterogeneous sources: statements from 5 to 10 banks, aged receivables, supplier payment schedule, payroll, VAT, loan repayments. Every Monday, the treasurer copies, pastes and corrects. Typical errors:
- Customer payment terms assumed to be contractual while the real average delay runs 12 to 20 days longer (EU rules cap B2B terms at 60 days, practice overshoots).
- Seasonality ignored on end-of-quarter receipts.
- Intercompany flows counted twice.
AI only helps if the data arrives clean: the first building block is always automatic collection of bank statements via EBICS T or EBICS TS, the standard in both Germany and France.
2026 options compared
| Criterion | Improved Excel | Cash management SaaS | Custom predictive model |
|---|---|---|---|
| Setup cost | 0 to 5,000 EUR | 5,000 to 20,000 EUR | 25,000 to 60,000 EUR excl. VAT |
| Recurring cost | 0 EUR | 1,000 to 4,000 EUR per month | 600 to 1,500 EUR per month (hosting, maintenance) |
| EBICS connectors | No | Yes, standard | Yes, built or via gateway |
| Customer receipts forecast | Fixed rules | Generic statistics | Model trained on each customer's history |
| Forecast error at 4 weeks | 12 to 18% | 7 to 10% | 4 to 6% |
| Deployment time | Immediate | 6 to 12 weeks | 10 to 16 weeks |
| Fit to specific flows | Total but manual | Limited to configuration | Total |
SaaS fits if your flows are standard and you also want payments and account management in the same tool. Custom makes sense when receipts depend on very heterogeneous customer behavior (large retail, export, public bodies) that generic models capture poorly.
What a custom predictive model includes
| Building block | Content | Indicative budget (EUR excl. VAT) |
|---|---|---|
| EBICS and ERP collection | camt.053 statements, invoices, orders | 6,000 to 15,000 |
| Cleaning and categorization | Rules and learning on bank transaction labels | 4,000 to 10,000 |
| Receipts model per customer | Real delay, seasonality, probable late payments | 8,000 to 18,000 |
| Disbursements model | Suppliers, payroll, tax, loans | 3,000 to 8,000 |
| Dashboard and scenarios | 13-week view, stress tests, covenant alerts | 4,000 to 9,000 |
Over 3 years, custom comes to 45,000 to 115,000 EUR versus 41,000 to 164,000 EUR for a SaaS. The cost gap is small; the accuracy gap depends on your data.
Mini case study
Claudia, treasurer of an industrial mid-market company in Berlin (120 million EUR revenue, 8 banks, 4 subsidiaries), spends every Monday on her forecast, about 45 days per year. Her average 4-week error is 14%, which forces the group to keep a 6 million EUR safety buffer on its accounts.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
The company invests 42,000 EUR excl. VAT in a custom model, plus 1,000 EUR per month. After 6 months, the error falls to 5%. The buffer can be cut to 3 million EUR: invested at 2.5% or avoiding a credit line drawdown at 4.5%, those 3 million EUR are worth 75,000 to 135,000 EUR per year. Claudia recovers about 35 days per year, reassigned to bank negotiations. The project pays back in under 6 months, a 2026 order of magnitude.
FAQ
Do we need an EBICS contract with each bank?
Yes, allow 50 to 150 EUR per month per bank for EBICS TS. Some solutions use a shared gateway to reduce that cost.
How much history is needed to train the model?
Ideally 24 to 36 months of invoices and receipts. With 12 months, the model works but the error stays closer to 7 to 8%.
Does AI replace the treasurer?
No, it automates collection and proposes a forecast that the treasurer adjusts. The treasurer keeps control over exceptional events, such as an acquisition or a dispute.
Does the data stay in the EU?
A custom build can be hosted with a certified German or French provider (C5 or SecNumCloud) according to your policy. It is a criterion to check for every SaaS.
Can foreign subsidiaries' forecasts be included?
Yes, in multiple currencies with daily rate conversion. A group of 4 to 10 subsidiaries generally adds 5,000 to 12,000 EUR to the initial budget.
Let's scope your project. We scope your automated cash flow forecast (EBICS connectors, ERP, predictive model, scenarios) for a budget of 25,000 to 60,000 EUR excl. VAT and a first reliable forecast within 10 to 16 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
