Case Studies | FinTech

Scaling a virtual CFO for production

Anthropic to Bedrock Migration

About

A FinTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

Challenge

The migration addressed four focus areas that had to be met before the assistant could scale in production.

Agentic orchestration, not manual routing

The assistant needed to accept a natural instruction, for example generating a spend-by-category query for the last thirty days, and autonomously select the correct action, whether generating a query, creating a category, setting a rule or showing a spend breakdown. Reasoning by the model should only kick in when deterministic checks flag genuine uncertainty, keeping behaviour predictable and cost-controlled.

Structured output and safe execution

Every action had to return well-formed JSON that downstream Lambda functions could parse and act on, executing sanitised SQL against the data store using the end user's own bearer token. The system had to reject SQL injection, unauthorised schema discovery or anything outside permitted scopes, so that automation never became a security liability.

Guardrails and data protection

Running in a regulated, financial context meant Bedrock Guardrails had to prevent any prompt or response containing personally identifiable information or leading to data leakage, and block toxic or prompt-injection attempts. Raw financial data had to stay within the client's systems throughout.

Observability and audit

To be production-grade and compliant, every request needed an unique request ID, with the prompt, response tokens, invoked tools, Lambda inputs and outputs and any guardrail events logged and retained for at least ninety days. Every automated decision had to be fully reconstructible.

Solution

Cloud Combinator delivered the migration in clear phases so that the client could see the architecture prove itself before it carried production traffic, with a controlled pilot ahead of full cutover.

1,000

Business customers targeted at scale (projected)

90-day

Audit trail retention on every request

~10 weeks

Phased delivery to final acceptance

By the numbers:

  • 1,000 - Business customers targeted at scale (projected)
  • 90-day - Audit trail retention on every request
  • ~10 weeks - Phased delivery to final acceptance
Changes

The engagement delivered an EU-hosted, agentic virtual CFO on Amazon Bedrock against the functional and non-functional success criteria in the Statement of Work, validated through a controlled pilot and signed off after hyper-care. The figures below describe the target scale and design commitments rather than an approved production benchmark.

  • Autonomous, predictable orchestration — The Bedrock Agent selects the right action group for each instruction without manual routing, reserving model reasoning for genuinely ambiguous cases to keep behaviour and cost under control.
  • Safe, structured execution — Actions return well-formed JSON that Lambda parses to run sanitised SQL under the user's own bearer token, with injection, unauthorised schema access and out-of-scope commands rejected.
  • Compliance built in — Bedrock Guardrails block PII leakage and prompt-injection, EU-region hosting supports data residency, and raw financial data stays within the client's systems.
  • Fully auditable — Every request is tagged with an unique ID and its prompts, tokens, tool calls and guardrail events are retained for ninety days, so any automated decision can be reconstructed.
  • Handover — The client received the solution as infrastructure as code with an operational runbook, incident-response procedures and a knowledge-transfer workshop, plus two weeks of hyper-care support.

With an agentic, guardrailed foundation in place, the client can add new tools simply by defining a Bedrock action group and wiring a Lambda, scaling the virtual CFO to its growing customer base without rearchitecting the core.

AWS Stack

Amazon Bedrock Agents

For autonomous orchestration of the virtual CFO's action groups.

Amazon Bedrock Guardrails

For enforcing PII and PCI safety and blocking prompt-injection.

AWS Lambda

For serverless microservices for instruction classification, query execution and rule application.

Amazon CloudWatch

For request-tagged logging and observability with ninety-day retention.

AWS CloudFormation

For infrastructure as code for repeatable, version-controlled deployment.

Amazon EventBridge

And Amazon S3 for event routing and durable storage for audit artifacts.

YOU MIGHT LIKE

Related success stories

View all case studies

Case Studies | Insights

Utilising Language Recognition, Speed, and Enhanced Security to Make Social Media a Force for Good

  • Here, we take a detailed look at how the Cloud Combinator team collaborated with another cutting-edge AI service provider that provides intelligent systems to “make social media more social” for brands and users alike.
  • Arwen AI is a UK-based startup specialising in AI solutions to manage and enhance brands’ social media interactions. Founded in 2020 by Matt McGrory, Dr. David Cole, and Joel Bailey, Arwen. AI focuses on using AI to automatically detect and remove spam, toxic comments, and other unwanted content from social media platforms.
  • The team at Arwen have three core products. ‘Moderate’ is focused on identifying and removing toxic content from social media channels. ‘Engage’ helps brands identify and engage with meaningful conversations on social media, and ‘Customize’ allows brands to apply bespoke algorithms to their channels - creating an even more effective moderation and engagement.
Read more
CONTACT US

Ready to turn AI into impact?

We'll help you spot the highest-value opportunities, reduce risk around your first AI initiative, and define a clear path to results from day one.

Why talk to us:

Outcome-driven recommendations

AWS-recognised delivery expertise

Risk-aware AI adoption

Clear next step, not a sales pitch

Start with a focused 20-minute conversation about your goals — no pressure, no commitment.

We use cookies

This website uses cookies to enhance user experience and to analyze performance and traffic on our website.

See our Privacy Policy for details.