Kyc aml · Production

Adyen builds AI Applied Research Engineering team for Integrity Risk automation, foundational transaction models, and Uplift optimization

The problem

Traditional integrity risk processes (AML, KYC, CDD, CRR) require large human teams to manually research, analyze, and label cases; even Adyen's existing Graph Neural Network framework cannot automate those human review steps. A production readiness gap also exists for AI data analysis agents, with state-of-the-art models achieving only around 16% accuracy on multi-step reasoning tasks.

First attempt

Adyen's existing ML integrity risk framework using Graph Neural Networks surfaces cases with optimized precision but cannot automate the human analysis steps; and current state-of-the-art AI agents achieve only around 16% accuracy on multi-step reasoning benchmarks requiring structured and unstructured data.

Workflow diagram · grounded in source
1
GNN surfaces integrity risk cases
ai_action
“ML models, like Graph Neural Networks, surface cases to human analysts. The framework is designed to optimize precision such that the cases being worked on are worthy of human time”
2
Human analysts review cases
human_review
“employ a sizable team of humans that manually research, analyze and label cases”
3
LLM agents automate human-heavy steps
ai_action
“the advent of LLMs and especially AI agents with access to tools like database queries and web search presents a great opportunity to automate those human-heavy workflows even further”
4
Operational feedback and RL training
feedback_loop
“This team provides feedback to measure, fine-tune, and reinforce our algorithms. This data not only provides a way to measure and harness the performance of AI Agents, but it contains a wide set of signals that can be used to accelerate …”
5
Uplift AI decides transaction outcome
ai_action
“an AI trained on a massive dataset decides the course of every transaction by balancing cost, risk and conversion in an optimal outcome and a minimal overhead for the operations team”
Reported outcome

Adyen established a research engineering team bridging AI research and production, co-built the DABStep benchmark with Hugging Face using over 450 real-world tasks, and runs Adyen Uplift in production platform-wide with an AI deciding the course of every transaction.

Reported metrics
SOTA agent accuracy on DABSteparound 16%
Adyen payment volume processed (2024)more than USD 1.4 trillion
YOY payment volume growth~25%
DABStep benchmark task countover 450
Reported stack
Graph Neural NetworksLLMsDABStepAdyen UpliftPPOHugging Face
Source
https://medium.com/adyen/unlocking-value-through-ai-applied-research-engineering-3dc3997a0bbd
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Adyen established a research engineering team bridging AI research and production, co-built the DABStep benchmark with Hugging Face using over 450 real-world tasks, and runs Adyen Uplift in production platform-wide wi…

What tools did this team use?

Graph Neural Networks, LLMs, DABStep, Adyen Uplift, PPO, Hugging Face.

What results were reported?

SOTA agent accuracy on DABStep: around 16%; Adyen payment volume processed (2024): more than USD 1.4 trillion; YOY payment volume growth: ~25%; DABStep benchmark task count: over 450 (source-reported, not independently verified).

What failed first in this deployment?

Adyen's existing ML integrity risk framework using Graph Neural Networks surfaces cases with optimized precision but cannot automate the human analysis steps; and current state-of-the-art AI agents achieve only around…

How is this kyc aml AI workflow structured?

GNN surfaces integrity risk cases → Human analysts review cases → LLM agents automate human-heavy steps → Operational feedback and RL training → Uplift AI decides transaction outcome.

Related kyc aml cases