Your Expertise Has a Headcount. Your Demand Doesn’t.
The most valuable asset in a technical distribution or manufacturing business usually isn’t the catalog. It’s the team. A handful of people who know the catalog inside and out. Which interface plate ships with which master, what changes above a 300-kilogram payload, when a configuration needs a second look before it goes out the door. That knowledge takes decades to build, and it is the hardest thing in the business to replace.
It also has a hard limit. Your expertise has a headcount, and your demand doesn’t.
That isn’t a criticism of anyone. It’s arithmetic. Even on a large team, the hardest configurations route to the same few people, because that’s where the deepest pattern recognition sits. Customers are in other time zones, they ask on Friday evenings, and all of them want that person. Hiring is the obvious response, but that judgment takes years to build, and the people who have it are retiring faster than they can be replaced.
Putting a general chatbot on the website is the other obvious response, but most chatbots fail for two reasons. They’re built for general support, not the specifics industrial companies need. And they run on general-purpose language models trained to give the most probable answer rather than the most accurate one. The more technical and niche the question, the more likely a confident wrong answer. This is a hallucination. It isn’t a defect waiting on a patch, it’s how they work. A model with no access to your product data fills the gap with something plausible and has no way to tell you which answers are which. For a part number, plausible ships the wrong part.
Novanta took a third path. Its ATI tool-changer catalog runs to millions of valid combinations across five module categories (electrical, pneumatic, control and signal, high-current, and servo), each tied to a specific robot from FANUC, Universal Robots, Kawasaki, Denso, Yaskawa, EPSON, ABB or KUKA. The people who could resolve a customer’s application into a validated part number were a small team of application engineers carrying decades of pattern recognition.
Rather than route around them, Novanta started with them.
Their institutional knowledge was drawn straight from the engineers who held it and modeled into a knowledge grounding layer, a knowledge graph the agent draws from rather than improvises against, reached through a purpose-built tool harness. Those same engineers then tested it every week. A new version shipped every Monday. When one of them noticed the compatibility matrix had an adapter relation wrong, the correction was queued that day and went out in the next release. Their judgment stayed the authority the whole way through. The system was how it got recorded.
Quick Consult went live on ati.novanta.com in April 2026. A customer describes the robot, the payload and the module mix, and gets back a validated Novanta part number, in any of seven languages, at any hour. Every part number is cross-checked against Novanta’s canonical part-number reference before a customer ever sees it, and across more than 400 customer sessions that check has held at 100%. Evals hold the agent’s behavior to real configurations before any change ships, and continuous sync keeps the graph current as the catalog moves. Anything it can’t validate goes to an engineer with the whole conversation attached, a path designed in from the first week.
Ajaiey Sharma, the Novanta VP who sponsored the work, put the result plainly: “We didn’t just expand capacity—we made our expertise available to every customer, instantly. That’s how you turn complexity into an advantage and create real, scalable growth.”
The engineers are content owners now. Their expertise reaches every customer who asks rather than the fraction who made it to the queue, and their own hours go to the configurations that genuinely need a person. Routine tickets stopped being the price of admission for the interesting work.
That outcome is rarer than the market makes it sound. MIT’s NANDA “GenAI Divide” study found roughly 5% of enterprise generative-AI pilots reach production with measurable impact, and that vendor-led, partnership-driven projects succeeded in about 67% of cases against 33% for internal builds. Novanta staffed it accordingly: a decision-maker, a subject-matter expert and a project owner on their side, a forward-deployed engineer from Reshape, one shared backlog.
If you’re looking at the same constraint, start by finding out what you would lose. Take the last month of configuration requests your team resolved, and mark each one according to whether the answer existed in a document or only in somebody’s head. That second number is the expertise that walks out the door at retirement. It’s also what any system would have to be built on, and most businesses have never counted it.
Web link
https://www.reshapex.com/en/case-studies/quick-consult-novanta
About the author
James Sugrue is Head of Sales Operations and Customer Success at Reshape Automation (ReshapeX), which builds the knowledge grounding layer for AI deployed in industry. With 30 years in manufacturing, design, and technology, James works with distributors and manufacturers on getting their own product knowledge into a form their customers can reach. Reach him at james@reshapeautomation.com or visit reshapex.com.
Contact
James Sugrue
Head of Sales Operations and Customer Success
Reshape Automation
james@reshapeautomation.com
Wednesday, August 26, 2026
by: James Sugrue
Section: Press Releases
