The problem
The NBA is the professional body for accountants in the Netherlands. Its technical helpdesk receives thousands of questions a year about laws and regulations, answered by a team of specialists who combine this work with other duties. The knowledge was there, spread across the audit standards (HRA), the Dutch accounting guidelines, guidance notes, handbooks and previously answered questions in the helpdesk software, but finding it took time. An unexpectedly complex case could take hours of searching, similar questions were researched all over again and the agreed response time was under pressure. Meanwhile members saw the standards as a last resort rather than a reference work.
The solution
We start with knowledge management: all sources in one knowledge base, kept current through links with the NBA website and the document management system, including the anonymised helpdesk history. On top of that we are building an AI assistant: specialists ask a question in plain language and get a substantiated answer with direct links to the exact article or paragraph. The assistant answers only from the knowledge base, asks follow-up questions when context is missing and says so when it is not sure. We measure quality up front on a golden dataset of difficult helpdesk questions, with fixed thresholds before the next user group comes on board. The professional judgement stays with the specialist.
The result
The first prototype is in place and is now being sharpened in iterations together with the helpdesk specialists. The aim is to have the relevant regulations and comparable earlier cases at hand within seconds instead of hours of searching, with answers that are consistent in style, depth and citation. Once the quality threshold on the golden dataset is met, the assistant goes live internally. After that comes the roll-out to all members, so that a large share of questions no longer needs to reach the helpdesk and the specialists can focus on the complex cases.
