Denodo & AWS: Powering Agentic AI for Banking

WEBVTT

00:00:00.000 -> 00:00:10.000

Narrator: In this demo, we're showing how Denodo and AWS together power agentic AI for banking, specifically stopping account takeover fraud in seconds, not hours. Let's get into it.

00:00:10.000 -> 00:00:27.000

Narrator: Account takeover fraud is hard to stop because fraud data is siloed. Ecosystems are fragmented across on-prem and cloud, and enriched signals arrive 8 to 14 hours late from overnight batch jobs. By then, the money has moved. This is what we're solving.

00:00:27.000 -> 00:01:10.000

Narrator: Together, Denodo and AWS provide the unified data foundation required to power agentic AI in banking. This foundation enables AI agents to securely access and reason over enterprise data across multi-cloud and hybrid environments without requiring data movement or duplication. A context-rich semantic layer ensures that agents operate with consistent business meaning and shared definitions across all data sources. Zero-copy data federation delivers real-time access to both operational and analytical data, allowing agents to retrieve and act on the most current information. And strong, fine-grained governance and security across AWS and non-AWS systems ensures that every agent action is controlled, auditable, and compliant at enterprise scale.

00:01:10.000 -> 00:01:45.000

Narrator: Now let's dive into the agentic AI demo scenario. Five AI agents handle this end-to-end. Agent 1 monitors behavior in real time and scores anomalies. Agent 2 weighs all signals and makes a block or escalate decision. Agent 3 assembles the complete case file automatically. Agent 4 executes containment across every system in under 30 seconds. Agent 5 handles compliance, populating the suspicious activity report and routing for human sign-off. Humans stay in the loop for exceptions and regulatory approvals; everything else runs autonomously.

00:01:45.000 -> 00:02:03.000

Narrator: Under the hood, Denodo federates five separate on-premises source systems into a single governed layer, live, zero copy, in milliseconds. Amazon Bedrock powers the agents through Amazon Quick, and the Denodo Data Marketplace makes those data products discoverable and reusable. Let's see it running.

00:02:03.000 -> 00:02:27.000

Narrator: [UI showing Amazon Quick Automations workflow] Here's the automation in Amazon Quick, five agents running sequentially, each one's output feeding the next. What makes this work is that every agent is reasoning over semantically enriched context, not raw data pulled from individual source systems, but governed, business-ready meaning that Denodo assembles in real time. That semantic layer is what allows each agent to make a confident, explainable decision.

00:02:27.000 -> 00:02:49.000

Narrator: [UI displaying Denodo Data Marketplace view] Now let me show you what's powering these agents. In the Denodo Data Marketplace, here's the transaction risk logical data product, a real-time, semantically enriched view that federates core banking, card management, digital banking, and KYC into a single governed data product, trusted, tagged, categorized for financial crime and fraud.

00:02:49.000 -> 00:03:09.000

Narrator: [UI highlighting Data Lineage feature] Every field in this view traces back through the transformation layers to the raw source systems: core banking, card management, digital banking, all federated in a single query, enriched in real time. This is the semantic layer turning fragmented source data into AI-ready context.

00:03:09.000 -> 00:03:50.000

Narrator: The fraud risk decision, and the most important moment in the workflow. What you're seeing is agentic AI reasoning powered by Denodo's real-time semantic layer. The agent isn't scoring a raw transaction; it's reasoning over enriched context assembled live from multiple source systems. The fraud score hits 0.97, well above the auto-block threshold: a confirmed mule account sharing the same device and IP, a transfer 19.59 times Maria's historical maximum, new device on VPN from Berlin, 932 kilometers from home, every anomaly firing simultaneously. That explainability is only possible because Denodo is providing real-time semantic meaning to every signal.

00:03:50.000 -> 00:04:05.000

Narrator: From there, Agent 3 assembles the complete case narrative from live Denodo data: timeline, a breakdown of exactly which signals drove the fraud score, and a full evidence summary before the analyst opens it.

00:04:05.000 -> 00:04:14.000

Narrator: Agent 4 executes the containment: blocks card, suspends online banking, and notifies customer in one step.

00:04:14.000 -> 00:04:22.000

Narrator: And Agent 5 runs the compliance close: SAR fields populated, routed to a human approver for sign-off.

00:04:22.000 -> 00:04:42.000

Narrator: [UI showing generated Suspicious Activity Report PDF] And here's the output: a fully populated suspicious activity report. All 14 fields auto-generated: subject details, financial summary, AML typology signals, the full suspicious activity description. Every field carries machine-readable provenance back to its source. Regulators verify facts, not narrative.

00:04:42.000 -> 00:04:53.000

Narrator: In summary: fraud stopped before funds leave the account, case investigation time from hours to seconds, and every decision fully governed and traceable.

00:04:53.000 -> 00:05:04.000

Narrator: Unified Data. Powering Agentic AI at Scale. Learn more at denodo.com/aws. Thank you.