At a glance
95%
of technical inquiries resolved without human intervention
Minutes
to resolve a technical inquiry, down from a few days
0
dedicated support hires, a year and a half after launch
- Customer
- A global medtech manufacturer (name withheld)
- Industry
- Medical technology
- Challenge
- Sales staff were fielding technical support questions, and each took days to resolve
- Solution
- An AI support agent built on a knowledge graph of the company’s technical documentation, running in Microsoft Azure and integrated with HubSpot and the website contact form
The challenge: salespeople doubling as technical support
~50
technical inquiries a month
A few days
to resolve a typical inquiry
The company makes medical devices used by hospitals and labs around the world. When a user has a question about how a product works or how to use it correctly, it contacts the manufacturer. In healthcare, a slow or unclear answer to that kind of question matters more than it does in most industries.
For years, those questions landed with the sales team. Roughly 50 technical inquiries arrived each month. Salespeople answered what they could from the product’s instructions for use and other technical documents, much of it in legacy formats that were hard to search.
When the answer wasn’t in the documents, or wasn’t easy to find, salespeople had to go to technicians in manufacturing. That pulled people off production to answer support questions. A typical technical inquiry took a few days to resolve.
The arrangement was reaching its limit. Leadership was considering splitting sales and support into separate roles, which meant building a support function from scratch.
Why Wizard Labs
Wizard Labs was already advising the company on its broader AI transformation. As we mapped where AI could make the biggest difference, customer support stood out. The problem was well defined, the knowledge needed to solve it already existed in the company’s documentation, and it bore directly on the staffing decision leadership was weighing.
The same team that scoped the opportunity went on to prepare the data and build the system.
The solution: an AI support agent proven in shadow mode before it went live
Shadow mode for 9 months
The agent answered real tickets silently and was scored against the human team. It went live once it resolved 80% of them.
Live for 6 months
The 20% it couldn’t resolve guided each improvement. It now resolves 95% of technical inquiries.
Turning legacy documents into usable knowledge
The answers to most technical questions already existed in the company’s documentation. The problem was getting to them. Wizard Labs used multimodal AI with computer vision to ingest the legacy-format technical documents and convert them into structured, searchable data inside the company’s own secure cloud environment.
Testing against real tickets, with no disruption
We didn’t put an unproven agent in front of customers. Instead, we connected a shadow agent to the company’s existing ticketing system. As technical inquiries arrived, the agent drafted its own answers, and we compared them with the responses the human team actually sent. Customers saw no change, and staff kept working exactly as before.
We started by scoring the agent directly against those human responses. We then built an internal evaluation suite from those real cases and used it to benchmark every change to the system.
The obstacle: early results weren’t good enough
At first, the agent resolved too few inquiries to be useful. Standard retrieval wasn’t matching the way customers actually phrase technical questions. The team rebuilt the retrieval-augmented generation (RAG) architecture around a graph of entities that tags each piece of documentation by product, use-case category, and the clinical conditions the product relates to.
That last link moved the needle most. Users often describe a question in terms of a condition rather than a product name, and the graph lets the agent connect the two. Over about nine months, the agent’s resolution rate in shadow testing rose to 80%.
Going live, then using the gaps as a roadmap
At 80%, the agent went live. We integrated it with HubSpot and the website contact form so technical inquiries reach the agent before a person. A classifier screens every incoming inquiry first; only technical ones go to the agent, and everything else goes straight to the team. The system runs in Microsoft Azure.
The 20% the agent couldn’t yet resolve became the guide for the next six months of work.
How an inquiry is routed
Inquiry arrives through the website contact form or HubSpot
A classifier decides whether it’s a technical question
Technical
The AI agent answers in minutes; 95% need no human
Everything else
Goes straight to the sales team, where a conversation can open a sale
The results: 95% of technical inquiries resolved in minutes
Value from the first day live
At launch, the agent was already resolving four out of five technical inquiries with no human involvement. Most technical questions were off the sales team’s plate from day one, months before the system reached its final performance.
95% resolved without human intervention
After six more months of improvement, the agent now resolves 95% of technical inquiries without a person stepping in. That figure, from Wizard Labs’ own tracking, counts only technical inquiries. Other inquiries are routed to people by design and are excluded from the figure.
From days to minutes
A technical inquiry that once took a few days to resolve now gets an answer in a couple of minutes.
No support team needed
A year and a half after launch, the company still has no dedicated support staff. Salespeople still handle some service requests, but these are now mostly questions that lead naturally into a sales conversation.
Human contact where it counts
Non-technical inquiries go straight to people on purpose. Some come from people in customers’ purchasing departments, and a human conversation there builds the relationship and opens sales opportunities. The AI handles the routine technical questions, so the team can spend its time on the conversations that grow the business.
