How North Atlantic Industries Can Trust Its AI to Answer Customer Questions across 4.2 quadrillion product configurations

How do you keep an AI assistant accurate when your product line can be configured 4.2 quadrillion ways? North Atlantic Industries (NAI), a defense technology company that builds ruggedized compute for larger platforms, had to answer that question for its sales managers, engineers, and admins.

The answer starts with AI-ready product data: product records that are accurate, complete, structured, and well-governed enough for an AI assistant to answer from without a person correcting the output. NAI holds that data in Pimly, and its AI assistant reads it directly.

Before Pimly, a single spec change touched several places and three people, and it took three days to reach the website and docs. NAI's AI assistant was only as current as that data.

Here is what NAI achieved, and what it means for any team that sells a configurable product.

Key takeaways

• Regional sales managers answer spec questions in front of customers, on their phones, from current specs instead of saved files or printed manuals that go stale.
• A spec change that took three days and three people now takes five to ten minutes and one person.
• NAI's Agentforce assistant reads all 54,662 NAI products from one governed record in Pimly, so every answer traces back to a single source of product data.
• Reps, engineers, and admins ask the same assistant and get answers from the same record, and NAI built it to serve its public website and customer portal too.

Want the full story behind these results? Download the complete NAI white paper.

Why did NAI's product data slow down its AI?

NAI builds its products from modules, like building blocks. "We never ship the same thing" to two different customers, says Tim Campbell, who leads IT and software engineering at NAI. That makes product data hard to keep straight.

Before Pimly, that data lived in a product builder application, the public website, manuals, spec sheets, and data sheets. On the website side alone, three people touched every spec change. One wrote it, one updated the website by hand, and a documentation writer updated the docs site. "It could take one to three days," says Damian DiCocco, NAI's Software Engineer Lead. Tim describes life before Pimly this way: "Keeping those all in sync was not easy because there's a ton of places to update it."

The AI had the same problem. NAI's support assistant started as a prebuilt chatbot, then became a retrieval database over NAI's manuals. Both needed constant upkeep. "We got tired of supporting the thing because we'd have to change every few months," Tim says. The assistant also lived outside Salesforce, which left the people using it with, in Damian's words, "a lot of runaround."

An assistant is only as current as the data behind it. This is the same readiness gap described in why Agentforce fails without product data readiness, and NAI needed the assistant and the data in one place.

What changed when NAI's assistant read directly from Pimly?

NAI went live on Salesforce in April 2025, and Tim's engineers build and integrate business systems there every day. Pimly Product Data Layer (PIM) runs natively in Salesforce, which fit how they already work. "It was very convenient and ideal that it was built on Salesforce," Tim says.

Damian loaded the full ERP catalog, 54,662 products, in one to three days. "I don't remember hitting any crazy roadblocks," he says.

NAI then rebuilt its assistant on Agentforce, inside Salesforce, as one central assistant. Pimly stores product data natively in Salesforce, so Agentforce reads it like any other record. There is no pipeline to build and no integration to maintain. Tim says of the assistant, which reads Pimly product data, "We like the fact that it was able to work with our very configurable product, which is not easy to do." Damian describes the difference in upkeep this way: "We don't have to worry about, oh, this brand new model's out, we need to update our whole system now. It's just been cleaner."

Now every rep, engineer, and admin asks the same assistant and gets answers from the same product record. People outside engineering also pull product data into their own Claude projects without asking engineering for a connection.

Curious how NAI structured a product line this configurable? Get the full walkthrough in the NAI white paper.

What results did NAI get from one governed product record?

A spec change reaches the AI in minutes

The person who writes a spec now changes it once, in Pimly, and the website, the docs site, and the assistant all draw on that record. "Now the person who's writing the spec originally can just update it in Pimly and it's done," Damian says. "It depends how complex the change is, but I would assume like 5, 10 minutes, which is awesome."

A change that took up to three days and three people now takes five to ten minutes and one person. That speed matters most for the AI, because the faster the data updates, the more current every answer is.

Sales managers answer customers in the room

NAI's regional sales managers spend much of their time on the road. They used to pull up a saved file or a printed manual in front of a customer, one that might have changed since they last downloaded it. Now they ask the assistant on their phones, on the spot, and quote from current specs. Agentforce is the interface, and the answer comes from Pimly product data. Damian says reps check the assistant before they commit to an answer with a customer: "Let's ask the bot really quick. That way they're not promising customers things that are outdated or not supported anymore." That is what it looks like when a rep becomes a product expert on technical SKUs.

What is next for NAI?

NAI plans a portal for its reps and distributors around the world, with product information, manuals, and new launches. "Every four years we'd bring them in and talk to them and tell them about new products. That's not often enough," Tim says. Damian aims to connect more systems through MCP servers, with Pimly as the product record underneath.

What does this mean for teams that sell configurable products?

NAI's results come from one change: product data lives in one governed record that its AI reads directly. The person who writes a spec updates it once, and reps in front of customers get current answers. Any manufacturer with a catalog that no two customers see the same way faces the same choice.

The result: every rep becomes a product expert, answering confidently, quoting accurately, and winning more deals.

See what your AI knows about your catalog. Book a scoped demo built around your SKU complexity.

Frequently asked questions

What is AI-ready product data?

AI-ready product data is product information that is accurate, complete, structured, and governed well enough for an AI assistant to answer from without a person correcting the output. At NAI, all 54,662 products live in one governed record in Pimly, and the Agentforce assistant reads that record directly. The outcome is that regional sales managers quote from current specs in front of customers.

How did North Atlantic Industries speed up spec changes?

NAI moved its product data into Pimly so the person who writes a spec updates it once. Before Pimly, a spec change took up to three days and three people, and now it takes five to ten minutes and one person. The website, the docs site, and the AI assistant all draw on that same record.

How does an AI assistant read product data in Pimly?

Pimly stores product data natively in Salesforce, so Agentforce reads it like any other record. NAI has no pipeline to build and no sync job to maintain between Pimly and its assistant. The outcome is an assistant that stays as current as the product data behind it.

Why isn't a chatbot trained on manuals enough for configurable products?

NAI tried that path first. It started with a prebuilt chatbot, then built a retrieval database over its manuals, and both needed constant upkeep. NAI then moved to an assistant that reads current product records, so the answers stay as current as the data.

How does a PIM help sales reps answer spec questions in front of customers?

A PIM gives the AI assistant one governed record to answer from, so a rep gets a current spec instead of a saved file or printed manual that might have changed. NAI's regional sales managers ask the assistant on their phones and quote from current specs on the spot. Reps become product experts even when no two customers get the same configuration.