Skip to content

Case study

A generative AI compliance assistant for MyComplianceOffice

A working compliance assistant inside the client’s own Oracle cloud, five months from signature.

The client

MyComplianceOffice is a Dublin headquartered compliance technology company and the largest Irish owned RegTech firm. Its platform runs more than 30 compliance products on a single system for regulated financial services firms, serving over 1,300 firms in 105 countries from Dublin and offices in New York, Chicago, Fort Worth, London, Singapore and Hyderabad.

The challenge

Every employee of a regulated firm faces the same questions. Can I accept this dinner invitation? Is this security on our restricted list? What does the rule require before I trade outside the firm?

The answers sit in two places, the regulation and the firm’s own policies, and they differ by jurisdiction, by firm and by the role of the person asking. Compliance teams answer them by hand, one email at a time.

MCO wanted to know whether a generative AI assistant could answer those questions directly from the source documents, and tailor each answer to the person asking.

Two constraints shaped the work. MCO runs its platform on Oracle Cloud Infrastructure, so the assistant had to run inside MCO’s own tenancy with client documents never leaving it. And because MCO serves its client firms from shared infrastructure, the assistant had to keep each firm’s policies and answers strictly separate.

What we did

We started with an options paper, comparing a packaged Oracle generative AI agent against a build on Oracle’s data science service. We recommended the build: lower cost, working output sooner, and a clean path onto Oracle’s later generative AI services.

A first prototype was running three days after MCO handed over the content. It ran on our own environment, so MCO could judge the approach before any of their infrastructure was touched.

Getting into MCO’s tenancy was the harder part. The generative AI models were available only in a region their compartment did not use. We set out the options, recommended the one that kept documents inside MCO’s Oracle estate, and worked alongside MCO’s IT team and Oracle’s specialists to make the change.

We then rebuilt the assistant on Oracle Cloud using retrieval augmented generation. It ingests the regulations, policies and restricted lists, finds the passages that answer a question, and has the language model write the answer from those passages alone. Every answer is scoped to the company, region and role of the person asking, so a UK employee sees UK rules and their own firm’s policy.

The outcome

Five months after signing, MCO had a working assistant inside its own cloud, with every document staying in its own estate. Their product team tested it hands on and kept using it.

  • A repeatable route into Oracle’s generative AI service from an existing tenancy, which MCO’s own IT team can run again
  • A reusable test set of questions and expected answers by company, region and role
  • A costed plan for the next phase

Oracle’s Ireland and UK team invited us to present the approach, and Oracle’s AI specialists joined the planning for what came next.

Where MCO took it

The assistant was a starting point rather than a destination. MCO built the retrieval augmented design out into an AI capability that now runs across the platform: a conversational entry point to more than 30 products, backed by assistants that answer from a firm’s own knowledge base, its policies and its live compliance records, with AI embedded in the compliance workflows themselves.

The principles still hold in production today. Retrieval from the client’s own documents, answers grounded in those documents alone, every answer scoped to company, region and role, and client data staying inside MCO’s own cloud.

“Idiro did an excellent job for us. They were fast, and they were straight with us about the trade offs at every decision point. Keeping client documents inside our own estate was non negotiable, and Idiro built to that constraint rather than around it. That original work became the foundation for everything we have done in AI since.”

John Kearney, Head of Product, MyComplianceOffice

Visit MyComplianceOffice

Next step

Could your data do this?

Tell us what you are trying to achieve. We will tell you plainly whether your data supports it, and what it would take to get there.

Book a discovery call