AI Applied Research Engineering for Payment Platform Value Creation

adyen · Finance · 2025 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.2 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Payments AI research engineering

Adyen needed to apply AI research meaningfully in critical large-scale payment flows while recognizing that exploratory work lacks guaranteed production outcomes.

Constraints:

regulated fintech
real-time payments
large-scale experiments
research-to-production gap

Architecture

Adyen describes an applied research-engineering operating system rather than a single agent product. Research engineers start from business-valued hypotheses, implement frontier AI research as robust production code, run experiments on Adyen’s large payments/data platform, use operational-team feedback to measure and improve algorithms, and partner with product teams to productionize. Current agentic work focuses on integrity-risk workflows where existing graph/ML systems surface cases to humans, and LLM agents with database/web-search tools may automate more of the research/analyze/label loop under regulated feedback.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  business-valued-hypothesis["business-valued-hypothesis"]
  research-engineering-team["research-engineering-team"]
  experiment-platform["experiment-platform"]
  payment-transaction-data["payment-transaction-data"]
  operational-feedback["operational-feedback"]
  integrity-risk-framework["integrity-risk-framework"]
  agentic-workflows["agentic-workflows"]
  genai-platform["genai-platform"]
  product-engineering-teams["product-engineering-teams"]
  business-valued-hypothesis --> research-engineering-team
  research-engineering-team --> experiment-platform
  experiment-platform --> payment-transaction-data
  payment-transaction-data --> integrity-risk-framework
  integrity-risk-framework --> agentic-workflows
  agentic-workflows --> genai-platform
  operational-feedback --> experiment-platform
  operational-feedback --> integrity-risk-framework
  agentic-workflows --> operational-feedback
  research-engineering-team --> product-engineering-teams

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`business-valued-hypothesis`	Impact-oriented hypothesis aimed at creating products, improving products, or changing finance/company operations.	`research-engineering-team`
`research-engineering-team`	Bridges AI research and production by implementing papers/ideas as scalable production-ready code.	`business-valued-hypothesis`, `experiment-platform`, `product-engineering-teams`, `genai-platform`
`experiment-platform`	Adyen’s large-scale platform for deploying and evaluating experiments on payments/data traffic.	`research-engineering-team`, `payment-transaction-data`, `operational-feedback`
`payment-transaction-data`	Structured and unstructured payments data flowing through Adyen APIs.	`experiment-platform`, `integrity-risk-framework`
`operational-feedback`	Signals from skilled operations teams reviewing cases and outcomes.	`experiment-platform`, `integrity-risk-framework`, `agentic-workflows`
`integrity-risk-framework`	Existing AI-first AML/KYC/CDD/CRR framework with graph models surfacing cases to humans.	`payment-transaction-data`, `operational-feedback`, `agentic-workflows`
`agentic-workflows`	Exploratory LLM/RL workflows with database-query and web-search tools for integrity-risk automation.	`integrity-risk-framework`, `genai-platform`, `operational-feedback`
`genai-platform`	Platform hosting open-source/open-weight models that can be contextualized, fine-tuned, and distilled.	`agentic-workflows`, `research-engineering-team`
`product-engineering-teams`	Production teams that partner with research engineering to convert exploratory initiatives into product value.	`research-engineering-team`

Data flow

A research hypothesis tied to product or finance impact is selected. Research engineers prototype and experiment on Adyen’s payments/data platform, using transactional data plus operational feedback. For integrity risk, existing ML such as graph neural networks surfaces cases to analysts; proposed LLM agents use database queries and web search to automate more investigation; operational teams review cases and feed performance signals back into measurement, fine-tuning, reinforcement learning, or product deployment.

4 supporting quotes

“we start with a hypothesis that would benefit any of the avenues above and iterate, build and experiment to either accept or reject that hypothesis.”article, stated

“Our data includes not only the transactional information flowing through our API, but also insights from a highly-skilled operational team. This team provides feedback to measure, fine-tune, and reinforce our algorithms.”article, stated

“ML models, like Graph Neural Networks, surface cases to human analysts.”article, stated

“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.”article, stated

Why this architecture

Adyen’s architecture is an organizational/technical bridge: pure research is too detached from business outcomes, while product teams may not have bandwidth for high-risk cross-team AI exploration. Research engineering stays exploratory but anchored to production value, supported by a massive data/experiment platform and operational feedback loops that can measure agent performance in regulated integrity-risk work.

4 supporting quotes

“The goal of the research engineering team is to bridge the gap between cutting-edge AI research and the use of this research in production applications, by taking research papers and implementing them into robust, scalable, production-ready code.”article, stated

“That is fundamentally different from unguided long-term, highly exploratory research that is not anchored on a business goal”article, stated

“the research engineering team and product engineering teams partner up to bring these exploratory initiatives into production value.”article, stated

“Integrity Risk is an ideal field to introduce research engineering because its stakeholders are internal customers, providing a direct feedback loop from operational teams”article, stated

Models used: AI/ML models · Graph Neural Networks · LLMs · open source/open weight models · Reinforcement Learning

Scale

Adyen processed more than USD 1.4 trillion in payment volume in 2024, with about 25% year-over-year growth.
“processed more than USD 1.4 trillion in payment volume in 2024 alone, with a steady growth of ~25% YOY.”article, stated
Adyen frames itself as a 4,500-person company operating from 28 offices and serving customers over 115 nationalities.
“We are a team of 4500 people who operate in 28 offices worldwide, serve customers in over 115 nationalities”article, stated
The experiment platform sits over more than $1.4T yearly traffic, giving research work massive experimentation potential.
“All this over a platform that processed more than 1.4 trillion USD in traffic yearly (2024).”article, stated
Research on multi-objective optimization is deployed platform-wide and affects every transaction.
“The research work here is deployed platform-wide, affecting every single transaction, allowing for massive impact and experimentation potential”article, stated

Alternatives considered (roads not taken)

Unguided long-term research detached from business outcomes

Rejected because: Adyen starts with hypotheses tied to products, product improvement, or finance/company transformation.

“That is fundamentally different from unguided long-term, highly exploratory research that is not anchored on a business goal that would (1) create new products, (2) improve our products or (3) fundamentally change the way finance and companies operate.”article, stated

A sizable human team manually researching, analyzing, and labeling integrity-risk cases

Rejected because: Adyen already uses AI-first case surfacing and sees LLM/tool agents as a way to automate human-heavy workflows further.

“The traditional angle from finance companies would be to employ a sizable team of humans that manually research, analyze and label cases.”article, stated

“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.”article, stated

Patterns

`eval-driven-development` (research-engineering)

Adyen emphasizes hypothesis rejection and experimentation on a high-scale payments platform.

“This means having the mindset of trying, the engineering to deploy and perform experiments and the scientific mind of rejecting hypotheses.”article, stated

`data-quality-loop` (model-operations)

Operational-team feedback is used to measure, fine-tune, and reinforce algorithms.

“This team provides feedback to measure, fine-tune, and reinforce our algorithms.”article, stated

Tradeoffs

How to pursue AI in payments.

Chose research engineering plus production AI products over passive observation or pure academic publication.

Rationale: Adyen positions itself to both ship AI-first products and explore emerging agentic/RL workflows.

“We cannot simply remain passive and observe.”article, implied

How to organize AI research for payments value creation.

Chose Applied research engineering anchored to business-valued hypotheses and product-team partnership. over Unguided long-term research or centralizing all innovation away from product teams..

Rationale: Adyen argues research engineering should remain exploratory but fixated on business outcomes and production value.

“we start with a hypothesis that would benefit any of the avenues above and iterate, build and experiment to either accept or reject that hypothesis.”article, stated

“The research engineering team does not behave like a product engineering team, bound to standard software development practices. As such, they work in a more exploratory, iterative way while staying fixated on business outcomes.”article, stated

Gotchas (what bit them)

Exploratory AI research does not promise production outcomes.

Root cause: Research engineering must bridge fast-moving community work with payment-platform constraints and network effects.

Resolution: Separate exploratory research from product execution while investing in experiment platforms and data access.

“Exploring, contrary to executing, does not have guaranteed production outcomes and requires ideating, experimenting, iterating and engaging with the AI community in order to learn and reach better outcomes.”article, stated

Evaluation

The article discusses platform experimentation and research posture; it does not provide a specific LLM benchmark.

Metrics: $1.4T yearly traffic platform · operational feedback · experiment capability

“All this over a platform that processed more than 1.4 trillion USD in traffic yearly (2024).”article, stated

Outcomes

Adyen is researching agentic workflows and reinforcement learning for integrity risk, but the case is strategic rather than a deployed LLM agent story.

This is not clearly an implemented LLM agent; marked non-agentic.

“AI Agentic workflows and Reinforcement Learning for Integrity Risk.”article, stated