Rabobank Deepens Its Bet on Expert.ai's Explainable, Hybrid Multi-Agent Architecture
Source: PR Newswire, via Morningstar — 2026-07-15
Summary
Rabobank is strengthening its long-standing partnership with Expert.ai to expand use of the "EidenAI Suite," a hybrid architecture that combines natural language understanding, knowledge graphs, classical machine learning, and LLMs within a governable multi-agent framework. The pitch is explainability by design — a real constraint in regulated banking, where an "I don't know why the model said that" answer isn't acceptable for many compliance and customer-facing decisions.
Key Takeaways
- The EidenAI Suite is explicitly hybrid: it combines NLU, knowledge graphs, traditional ML, and LLMs rather than routing everything through a single LLM — using each component where it's strongest.
- It's framed as a multi-agent system built for governability, meaning explainability and auditability are architectural goals, not an afterthought layered on top of a black-box model.
- Rabobank's continued, deepening use of this stack (rather than switching to a pure-LLM approach) signals that regulated banks are prioritizing explainability and auditability over raw LLM capability for at least some of their AI workloads.
Reel Script
Hook (~16s): Most companies are racing to put a single LLM in charge of everything. A regulated bank just deepened its bet on the opposite approach — a hybrid system where the LLM is only one piece of the puzzle.
Core Concept (~80s): In a regulated industry like banking, "the model said so" isn't an acceptable answer when a customer or a regulator asks why a decision was made — you need to be able to trace exactly how a conclusion was reached. That's a real constraint pure LLM systems struggle with, because a large language model's reasoning is much harder to audit step-by-step than, say, a knowledge graph lookup or a classical machine learning classifier with clear feature weights. Expert.ai's EidenAI Suite is built around that constraint directly: instead of one LLM handling everything end to end, it's a hybrid stack that combines natural language understanding, knowledge graphs, traditional ML models, and LLMs — using each piece for what it's actually good at, and keeping the overall system's reasoning traceable.
Hands-On (~90s): Think of the architecture as a multi-agent pipeline where each agent has a distinct, explainable role instead of one generalist model doing everything. NLU components handle parsing and understanding incoming text or requests. A knowledge graph component supplies structured, verifiable facts and relationships — the kind of thing you can point to directly as "this is where that answer came from." Traditional ML models handle classification or scoring tasks where a clear, auditable decision boundary matters more than open-ended reasoning. And LLMs get used specifically where flexible language generation or understanding is genuinely needed, rather than as the default answer to everything. The result is a system where, if a regulator or auditor asks "why did this get flagged," there's an actual traceable path through named components — not a black box you have to reverse-engineer.
Takeaway (~22s): If you're building AI for a regulated industry and reaching straight for "wrap everything in an LLM," this partnership is a reminder that the winning architecture there is often hybrid, not purist — and that explainability designed in from the start beats explainability bolted on after a compliance team asks hard questions.