Recap & Recording: AI Cash Flow Forecasting — The Ultimate Playbook

Key Takeaways

Jan-Willem Attevelt

AI in forecasting only works when built on structured and enriched data

“It is just the fact that you need to gather 2 years of clean historical data, and you need to categorize it. So think about the cash flow categories which are important to your business.”

Jan-Willem emphasises that the real barrier to AI adoption is not the model itself, but the lack of structured historical data. Without consistent categorisation and clean datasets, forecasting models cannot identify meaningful patterns or seasonality.

Agentic AI should support decision-making, not replace treasury judgement

“It is really going to guide you through the whole process… this is what I see in the patterns, and I’ve selected this model for this specific cash flow forecast category.”

He highlights that modern agentic AI can support forecasting by suggesting models and identifying patterns, but it still operates as a decision-support layer. The key emphasis is on transparency, explainability, and human oversight in every forecast output.

---

Annette Gilles

Fragmented data and unclear ownership are the biggest forecasting blockers

“If you pull from multiple ERP subsidiaries, manually, by email, whatever you have, this can turn into a sort of nightmare, so you have to sort that out.”

Annette stresses that most forecasting issues stem from operational fragmentation rather than tooling. Multiple systems, manual inputs, and unclear responsibilities create inconsistency that no automation layer can fully resolve.

Technology cannot compensate for weak processes or definitions

“Neither a TMS or AI is magically solving any forecasting problems you might have.”

Her core message is that technology only amplifies existing processes. Without clear forecasting definitions, ownership structures, and governance cycles, AI or treasury systems will not deliver reliable improvements.

---

Marianna Polykrati

Treasury teams are still heavily anchored in Excel-driven expertise

“What we have done, all treasurers we have built in the Excel models, we have built our own experiences and the knowledge that we have, and we are very good at firefighting.”

Marianna explains that Excel remains dominant because it embeds years of tacit treasury knowledge. This makes transformation difficult, as organisations are not just replacing a tool, but an entire way of working.

AI implementation must start with mapping and understanding the full cash flow ecosystem

“Don’t start with AI. So, as I said, map your cash data. Where it comes from.”

Her key message is that AI adoption must be preceded by deep process understanding. Teams need to map data flows, align stakeholders, and understand cash movements before introducing any automation layer.