Thousands of tickets as a knowledge source — how the service desk stopped asking the same person
The service desk logged several unresolved tickets each day, even though the answers existed in an archive of thousands of past cases. We processed the archive for semantic search and embedded the feature in the ticketing system, so the answer suggests itself before escalation.
Knowledge exists but nobody reaches for it
The service desk logged several tickets a day that couldn't be resolved at the first line. Each one was escalated to a more experienced person, lengthening response time and loading the team's bottleneck. The knowledge needed to solve them mostly existed — but scattered across three places: company notebooks, the system manufacturer's documentation, and an archive of thousands of tickets that nobody actually searched, because searching took longer than asking a colleague. Giving the team a conversational tool without connecting it to those sources wouldn't solve the problem — the system has no access to ticket history and generates generic rather than implementation-specific answers.
What we didSemantic search embedded in the working tool
- Processed the ticket archive into a format enabling semantic search — the one-time cost of processing the entire set turned out to be marginal relative to its value.
- Embedded the feature in the tool the team already uses — the ticketing system gained an option to suggest solutions based on similar past cases.
- Took a hybrid approach: we embedded answers to recurring questions as fixed scripts in the system, and gave the tool access to the ticket archive on a broader range of questions via a prompt with defined boundaries — a structured query pattern limiting the search to specific areas.
- Adopted the approach of releasing at partial quality — a feature providing any value is better than one waiting for full coverage. The organization verified this approach works before committing to broader integration.
What changed
The service desk received a tool that suggests solutions without leaving their main interface — first-line employees solve more tickets independently, without escalation. The archive of thousands of cases, previously dead data, became a working knowledge base updated with each new resolved ticket. The team began recording ticket resolutions as reusable items — a practice that further extends the knowledge base, independent of any AI tool.
“Access to documents isn't enough — the tool still doesn't know which version applies, which answer is current, or which case is the closest match. The difference between access and usefulness is structure: the archive must be processed for semantic search, not just connected.”
Let's write down how your company actually works.
- 45 minutes, and no preparation needed
- An honest read on where you sit today
- The first process worth automating — and why that one
