Enfuce is a Finnish company that provides cloud-based payment processing and card-issuing infrastructure, operating in the fast-moving and highly regulated fintech and payments industry. In this environment, innovation with AI must go hand in hand with security, scalability, and compliance.
To move to production-grade AI use cases faster, Enfuce wanted to build a robust and scalable foundation for agentic AI. The ambition was not only to deliver one use case, but to establish the architecture, infrastructure, governance, and development patterns needed to support future business-facing AI solutions.
Recordly joined the collaboration through a senior Data & AI consultant who worked as an embedded member of Enfuce’s team and was the sole external expert in the delivery. Working closely with Enfuce’s data scientists, analytics engineers, and business subject matter experts, Recordly helped build a secure and scalable platform foundation that enables Enfuce’s teams to develop new agentic AI use cases faster.
The first use case was fraud dispute automation: a customer support process where agents need to perform several checks, review available information, and apply human judgement before moving a case forward. After less than two months, Enfuce had an AI agent in production that can process fraud dispute tickets in the background, prepare a preliminary assessment, and support human agents with detailed reasoning.
.jpg)
The solution
Instead of a large delivery team, Recordly and Enfuce’s Data & AI engineers were able to move quickly from prototype to production while building both the Agentic AI platform foundation and the first fraud dispute automation use case.
The project was kicked off with an on-site workshop that resulted in a working prototype: an AI agent that could fetch data from Snowflake through a model context protocol (MCP) server and categorize fraud disputes. It leverages strong data models from the analytics engineering team, reducing rework and improving speed.
After the prototype, the project moved into the main implementation phase. Two workstreams progressed in parallel. First, the production infrastructure and platform foundation were put into place. Second, a focus on translating fraud dispute rules, scheme rulebooks, and subject matter expertise into logic the agent could use to support case assessment.
A key design principle was to use large language models (LLMs) only where they could provide real value, as not every step in an agentic workflow needs a language model. For deterministic checks, such as verifying whether a submitted transaction exists in the database, plain code is cheaper, faster, and more reliable. LLMs are most valuable when ambiguity is involved, such as interpreting natural language input from a human or reasoning over less structured information.
As a result, the final workflow combined deterministic code with LLM-based inference. This made the solution more reliable, cost-efficient, and suitable for a production fintech environment.
The business impact
Fraud dispute handling requires careful review, multiple data checks, and human judgement. With the AI agent in place, the initial assessment can be prepared in the background, allowing the human agent to focus on validating the recommendation, applying judgement, and ensuring quality.
Early estimates indicate that the time required for the initial assessment can be reduced by 80%. The value is expected to grow as dispute volumes increase, because the agentic solution gives Enfuce a way to scale recurring checks and information gathering without increasing manual work at the same pace.
Beyond the first use case, Enfuce now has a reusable foundation for future agentic AI solutions. The natural next step is to expand the platform with additional business-specific AI agents and use cases. Potential future areas include fraud monitoring and review, compliance-supporting workflows, operational case handling, internal knowledge assistants, and customer support use cases where AI can answer recurring questions, prepare case context, and flag edge cases for human review.
This allows support teams to spend more time on complex cases, customer-specific situations, and decisions that require human judgement or escalation.
What Enfuce said
“Recordly brought deep senior expertise that complemented our in-house capabilities. Their consultant worked as part of our team, collaborating closely with our data scientists, analytics engineers, and domain experts to build a secure and scalable foundation for agentic AI in a demanding fintech environment.
The first use case, fraud dispute automation, moved from prototype to production in around two months, creating a strong foundation for future agentic AI solutions.”
— Joren Hoog Antink, Head of Data & AI, Enfuce
Ask our experts more
.jpg)
See more of our work
