Financial Services Knowledge Base for Policy Questions

Industry: Financial services · Built: a knowledge base that answers from the firm's own documents · Team size: about 50 advisers

What was the problem?

Advisers had questions every day about product terms, policy wording and eligibility rules. The answers existed, but they were spread across thousands of pages of policy documents, product guides and internal circulars. Finding one meant searching shared folders or asking a senior colleague, which pulled the senior colleague away from their own clients.

What did we build?

An assistant that answers questions using only the firm's own documents. Advisers ask in plain English and get an answer with the document and page it came from, so they can check it before relying on it. It also:

• pulls specific details out of policy pages, such as limits, exclusions and waiting periods
• compares two or more policies side by side, so differences are listed rather than hunted for
• keeps a log of what people ask, so managers can see which topics come up most

This kind of system is often called retrieval-augmented generation. The assistant first finds the relevant passages in the firm's documents, then writes its answer from those passages only.

What changed?

The team's adviser puts the time saved at up to 150 hours a week across the team, about 3 hours per adviser. The time comes from looking up answers, pulling out policy details and comparing documents. That figure is the team's own estimate, and we did not measure it ourselves.

What did the questions reveal?

The question log turned out to be as useful as the answers. When the same question is asked dozens of times, it usually points to a gap in training or a document that is unclear. Managers now use the log to decide which training to run, based on what advisers ask. They also use it to see whether a topic stops coming up after the training.

What are the limits?

The assistant only knows what is in the documents it has been given. An outdated document produces an outdated answer, so a named person keeps the library current. Every answer shows its source, and advisers check it before advising a client. It does not give advice to clients directly.

Could this work for your team?

It suits teams that answer the same kinds of questions from a large set of documents: policies, product sheets, procedures or contracts. If your documents are few and short, a well-organised shared folder may be enough, and we will say so. See how we connect systems like this on our AI integration page, or read AI for Singapore SMEs: where to start.