Technical support workload cut by 20%.
That’s 8 FTE saved outright.

The answers already existed. Finding them didn’t.

Customers submitted technical requests by phone, platform, and mobile device, and the volume, product diversity, and case complexity kept growing. New technical content was added daily, which meant response quality could quietly degrade, outdated solutions could resurface, and there was limited ability to audit how a case was actually resolved.

The information SMEs needed to do their jobs was dispersed across multiple systems and formats, written in several languages, inconsistent in structure and depth, and often missing entirely for rare or outlier cases. In a safety-critical industry, SMEs were also reasonably unwilling to act on model output they couldn’t verify — trust depended on accurate information, clear sourcing, and an easy way to check the recommendation.

Three constraints were non-negotiable:

• Technical content scattered across systems, formats, and languages
• SMEs unwilling to trust opaque AI output in a safety-critical setting
• Rare and outlier cases with missing or incomplete documentation

A ranked list of real cases, never a single unexplained answer.

The team built a centralized technical-support knowledge base combining a data collection module, a retrieval component, and a scoring algorithm that ranks historical claims by similarity and relevance. Instead of returning one opaque answer, the system surfaces a priority-ranked list of comparable past cases, scored on vector similarity, with the original source material attached to every result.

A Python module automated claims ingestion and translation so multilingual records could be worked with consistently. Topic modelling pulled out structured metadata — product type, time in service, environment, issue category, root cause, resolution — to make search sharper and cases easier to reuse. Where metadata was incomplete, NLP-based completion suggested likely values, but these were always clearly labelled as suggestions, never mistaken for verified fact.

Quality was built into the workflow rather than bolted on: SMEs rated the usefulness of each result, that feedback fed back into the ranking, aggregated feedback was monitored over time, several distance metrics were tested for accuracy, and automated audit logs captured how each case was resolved. SMEs kept full control of the final decision throughout — the system accelerated research, it didn’t replace judgment.

One-fifth of the workload, gone.

Beyond the immediate time savings, the centralized knowledge base and feedback-driven ranking gave the organisation stronger auditability and a scalable foundation for long-term knowledge management. By keeping every result ranked, source-linked, and SME-verified, the programme delivered faster, more consistent technical support without giving up the traceability a safety-critical industry demands.