4most: PRA Rulebook answers, with the sources attached.
Fifty One Degrees partnered with 4most, which describes itself as the UK's largest independent credit risk and actuarial consultancy, to build a proof-of-concept AI assistant for PRA regulatory queries: grounded in the PRA Rulebook, with links and source documents attached to every answer.
Regulatory guidance consultants can trust, at the speed they work
The PRA Rulebook is large, dense and constantly referenced. For a consultancy advising banks and lenders on credit risk, every engagement leans on it, and a wrong or out-of-context reading carries real consequences. 4most wanted to know whether generative AI could give its consultants faster access to grounded regulatory answers without sacrificing the accuracy the work demands. The right way to find out was a proof of concept built on real queries, not a slide deck about potential.
An assistant grounded in the Rulebook
Answers with the sources attached
The assistant responds to PRA regulatory queries with relevant information plus the links and source documents behind it, so a consultant can verify every answer against the Rulebook itself rather than trusting a summary.
Retrieval over the PRA source book
Retrieval augmented generation grounds each response in the PRA Rulebook's actual text, built to handle the Rulebook's structure rather than treating it as generic text.
A feedback loop by design
The system collects queries, responses and user feedback from 4most's consultants, so accuracy and usefulness improve with use.
Two specialist firms, one build
Fifty One Degrees built the engineering
Design, development, testing and deployment sat with Fifty One Degrees: a Python build hosted on Microsoft Azure, using natural language processing and generative AI techniques.
4most brought the domain expertise
4most's consultants defined the real use cases, tested the assistant's answers against their own regulatory knowledge and fed the results back into the build.
Proof of concept before commitment
The engagement was structured to prove value first: a working assistant within a one-month implementation timeline, evaluated on genuine consultant queries before any decision about a wider rollout.
A strategic partnership, publicly announced
Both firms announced the partnership publicly in mid-2024, with 4most describing the work as part of its exploration of generative AI for regulatory horizon scanning. Full disclosure: Fifty One Degrees co-founder Mark Somers co-founded 4most and led it as Co-CEO until 2023, which is how each firm knew exactly what the other could do.
Agree the use case with 4most's consultants: PRA Rulebook queries, answered with sources.
Fifty One Degrees designs, develops and deploys the assistant on Azure within a one-month implementation timeline.
4most's regulatory specialists evaluate answers against real engagement questions and feed back.
Both firms assess the results, with 4most publicly reporting promising initial results from the proof of concept.
Initial results have been promising, demonstrating that generative AI could indeed accelerate and improve our content generation capabilities.Craig Watters, COO, 4most
What did Fifty One Degrees build with 4most?
A proof-of-concept AI assistant for PRA regulatory queries, jointly developed: Fifty One Degrees designed, built and deployed the Python-based assistant on Microsoft Azure, and 4most provided the credit risk and regulatory domain expertise and tested it on real consultant queries. Answers come grounded in the PRA Rulebook with links and source documents attached.
What results did the proof of concept show?
4most publicly reported that initial results were promising, demonstrating that generative AI could accelerate and improve its regulatory content generation capabilities. The proof of concept was evaluated by 4most's own regulatory specialists against real use cases.
Why is grounding in the PRA Rulebook important?
Because regulatory advice fails if the source is wrong. Retrieval augmented generation grounds every answer in the Rulebook's actual text and attaches the citations, so a consultant verifies rather than trusts, which is the standard regulated work demands.
Can Fifty One Degrees build a regulatory AI assistant for my firm?
Yes. Fifty One Degrees builds retrieval-grounded AI assistants over regulatory and technical source material, from the PRA Rulebook to internal policy libraries. Book a Discovery Call and we will scope a proof of concept on your firm's real queries.
Want grounded AI answers from your rulebook?
Book a 30-minute Discovery Call and we'll scope a proof of concept on the regulatory or technical source material your firm relies on.