“Sibos 2026 Miami in one word: exceptional.” That is how our CEO Niels van Daatselaar summed up the week. In four days he met with over 30 banks and more than 100 people, presented our new ERP integration proposition with Apideck, and joined the EY panel in the Meet the experts series.
This article covers what we took away on AI in business banking: what the most innovative banks are building, why the ERP matters, and where TreasurUp stands on the questions the EY panel put to the room.
Sibos Miami 2026: EY panel “Meet the experts: Will AI become the new partner for corporate treasury?”
AI becomes a real partner for business clients when it prepares the work and a person approves it, inside the bank’s own channels, including the ERP. The agents prepare. People decide. The bank that hosts the agent keeps the relationship.
What the most innovative banks are building in AI
Several presentations at Sibos showed what the most innovative banks are working on in AI, including client MCP layers, payment agents and agentic cash pool optimisation. Each maps onto a part of the platform TreasurUp runs for banks.
| Theme at Sibos | Where it sits in TreasurUp’s platform |
| Client MCP layers | Connection Centre: banking in the business client’s channel of choice, through 25+ REST APIs, MCP, files and micro apps |
| Payment agents | Cross-border payments: the agent combines FX rate sourcing with routing rules and compliance checks, and prepares the payment for approval before anything settles |
| Agentic cash pool optimisation | Liquidity management: multi-scenario cash flow forecasts from ERP and bank data, feeding sweep and pooling recommendations the owner approves |
The common thread is the approval gate. Each of these prepares an action. A person at the business client or the bank approves it.
Why the ERP matters for AI in business banking
The business client’s moment of need happens when an invoice is booked, an order is placed or a cash surplus appears. That moment often sits in the ERP, not in the business banking portal.
That is why Niels spent much of the week with Gertjan “GJ” De Wilde of Apideck, meeting banks about our new bank-ERP integration proposition. Apideck connects 50+ accounting and ERP systems, including QuickBooks, NetSuite and Microsoft Dynamics 365 Business Central, through one API. TreasurUp provides the bank’s capabilities in or next to the ERP. One integration instead of a project per ERP.
For AI in business banking, this is the groundwork. An agent is only as good as the data and connectivity beneath it.
The EY panel: from portal to partner
On the EY panel, Niels presented TreasurUp’s vision on AI from a business banking perspective. He shared the stage with James Sankey and Matt Cox (EY), Surath Sengupta (Lloyds Banking Group) and Aser Blanco (NVIDIA).
These are the five questions the session worked through, and where TreasurUp stands on each.
1. Can a business banking portal become a partner?
Business banking portals were built to let clients see information, access services and initiate transactions. A partner prepares the next step before the business client asks. Business clients no longer want to run the workflow themselves. They want the outcome. The hedge placed on time. The cash positioned before the weekend.
2. Where should the first dollar of AI investment go?
Where the daily relationship is most contested: cross-border payments, liquidity management and FX trading. Business owners already route these to fintechs and ERP or accounting tools, one workflow at a time. These are also the categories where TreasurUp’s composable platform already runs live at other banks.
3. One copilot or many specialised agents?
Scoped agents, each tied to one transaction category, on one shared intelligence engine. Whoever hosts the agent owns the relationship. From our perspective, the bank is the trust layer, which is also why TreasurUp’s platform has no cross-institution agent-to-agent flows.
4. Are business clients ready, or is connectivity still the constraint?
Connectivity comes first. In the Nordea case, that meant connectivity to 900+ ERP systems (on-premises) and 20+ (cloud hosted).3 Banks that solve connectivity first give their business clients something an agent can actually work with.
5. Where should the approval gate sit?
Between proposal and execution, on every material step. This is structural, not a setting. Where automation increases over time, it should come from better agent quality and tighter scope, never by removing people from material decisions. TreasurUp’s non-goals stay fixed: no autonomous trade execution, no autonomous limit changes, no autonomous credit decisions and no cross-institution agent-to-agent flows.
FAQ
1) Can an AI agent execute a payment, hedge or sweep on its own?
No. In TreasurUp’s platform, every action with financial, regulatory or accounting impact passes through an explicit human approval. The approval is logged with the agent’s reasoning chain, the input data, the alternatives considered and the decision. This holds across cross-border payments, liquidity management and FX trading.
2) Who owns the AI agent in business banking, the bank or the software vendor?
The bank. TreasurUp agents are bank-tenanted by default: branding, data, model choice and deployment topology are configurable per bank. The bank can run the platform in its own cloud, bring its own LLM, or choose a hybrid set-up with bank-hosted data and inference.
3) What is a client MCP layer in business banking?
MCP lets AI assistants and agents connect to a bank’s services in a structured, governed way. In TreasurUp’s Connection Centre, MCP sits alongside 25+ REST APIs, files and micro apps, so business clients can reach payments, FX and liquidity services in the channel they choose, under the bank’s controls.
4) How does approval-gated AI hold up against DORA and the EU AI Act?
Governance is platform specification. Components are catalogued by criticality and resilience-tested under DORA alignment. Model risk management covers inventory, validation, drift monitoring, override and rollback. Every query, proposal, approval and executed action is captured end to end.
Missed us at Sibos in Miami? Bring one workflow.
If your bank is weighing where AI in business banking starts, bring your cross-border payments, liquidity management or FX trading workflow to a working session. You leave with a scoped proof-of-concept plan.
Curious where the agent should sit at your bank?
Book a meeting