The AI Conversation in Distribution Isn't Really About AI

July 27, 2026

I've spent a lot of time over the past month on product demos and discovery calls with distributors. Plumbing, industrial supply, roofing, insulation, safety, you name it. Some were larger organizations. Others were family businesses with a few branches. Some were single branches. It was a really good spread of business types and sizes.

Obviously, every company is different. Different ERP. Different catalog. Different customers. Different process for getting a quote out the door. Or even a direct order in.

But after enough calls, you start to hear the same things.

The interesting part is that prospects aren't really asking whether AI can quote anymore. They're asking whether it can work inside their business without making everyone's job more complicated.

That's a much better question.

Nobody's asked me which model is running behind the screen. They ask:

• Will it understand the shorthand our customers use?
• Can my team stay in email and the ERP?
• What happens when the request starts with a phone call?
• Can the salesperson check the match before anything is created?
• How does it handle the line it doesn't understand?
• Will it do anything without my approval?

The technology matters, of course. But that's not where most of these decisions are getting made.

The Skepticism Is Real, but It's Specific

I don't think most distributors are afraid of AI. They're skeptical of software companies telling them a clean demo will work exactly the same way inside their actual workflows.

That's fair.

Most prospects have used ChatGPT. Many have seen AI pull information from a PDF, spreadsheet, email, or drawing. One AI transformation leader told me companies that aren't using data extraction today are missing out.

The question now is whether it can understand their version of the problem.

Can it work through a catalog pushing 100,000 items? Can it understand that two descriptions use different words for the same product? Can it handle the temporary item a salesperson created three years ago for one customer? Can it catch a unit-of-measure mismatch without pretending it knows the answer?

One distributor was skeptical before a controlled test. After we reviewed examples together, including the misses, his verdict was "guardingly optimistic."

I love that damn answer.

It wasn't blind excitement. It meant we had earned the right to keep working on the problem. The conversation immediately moved to what actually mattered: show the customer's description next to the catalog description, explain what was selected, and make it easy for the salesperson to correct it. They don't want to create more hassle for their team. Neither do we.

The misses were just as useful as the matches because they showed us what the product needed to do next.

Nobody Is Asking Us to Remove the Salesperson

Here's another thing I haven't heard once: "Please take my sales team completely out of this process."

Prospects are comfortable letting AI read the request, pull out the line items, suggest products, check inventory, and prepare a quote or order. They still want a person to review what matters and approve the result.

That's not resistance to AI. It's just the right way to split the work.

A 100-line RFQ isn't 100 difficult decisions. Most of the work is reading, lookup, matching, and entry. A smaller group of lines needs product knowledge, customer context, or a couple phone calls.

AI should clear the routine work so the salesperson can focus on those exceptions.

One prospect wanted a field salesperson to prepare a quote from the road, then send it to someone in the office for approval. Another distributor wanted users to see exactly what the system matched before anything moved forward.

That's the pattern I keep hearing. Prepare the work. Flag what is uncertain. Keep the person in control.

A Good Model in the Wrong Workflow Is Still a Bad Product

A lot of distributors live in email. The request comes into a shared inbox, someone forwards it, and the team wants the final record in the ERP. For those companies, an email-first workflow makes sense because it doesn't ask the salesperson to learn another place to work.

One enterprise AI leader described it perfectly: "Don't make me swivel chair. I want to sit in my ERP."

But not every distributor works that way.

One distributor gave me the most useful no I heard all month. A large share of its quoting happens outside email. Reps are on the phone, at job sites, or standing with the customer. A lot of the time, they skip a formal quote and create the order directly in the ERP.

Even if our product worked exactly as shown, an email-first workflow would have added another step.

So the model could work and the product could still be wrong for that sales team.

That call forced me to rethink the problem. Instead of asking how to automate a quote, the better question was how to move faster from a customer conversation to a placed order.

Maybe that means voice. Maybe it means an SMS confirmation after the call. I don't know exactly what the answer is yet, but we've started prototyping different options. The last thing I want is to force people to download software on their phones or add a recorder to every Zoom meeting.

But I know we can't force every distributor through the same front door.

AI Is Starting to Look Like a Knowledge Layer

The public AI conversation usually gets pulled toward job replacement. That's not the conversation I'm having with distributors.

They are worried about the knowledge walking out the door when experienced people retire.

One operations manager pointed to an aging workforce and said the knowledge those experienced employees carry can't simply be replaced.

Every distributor has some version of this. One rep knows the nickname a customer uses for a product. Another knows that a certain account always wants one manufacturer unless it's out of stock. Someone else can look at five words in an email and understand the missing specifications because they have seen the request a hundred times.

That knowledge is valuable, but most of it isn't written down.

If the system can remember corrections at the account level, the next person doesn't have to start from zero. New employees get help. Experienced reps answer fewer repetitive questions. The business becomes a little less dependent on one person knowing everything.

That doesn't replace the experienced salesperson. It makes that person harder to replace.

Another family-owned distributor said, "People buy from people."

Exactly. The point of AI should be to remove the administrative work that keeps salespeople from talking to customers.

Nonstock Is Where People Really Lean In

Quoting and order entry are easy for prospects to understand. The conversation gets more interesting when we start talking about nonstock.

One distributor told us nonstock represents roughly a third of its business. Some of those products are repetitive nonstocks that already exist in the ERP. Others are real one-offs that may require overseas sourcing and months of lead time.

Another industry leader described businesses where more than 30 percent of sales are nonstock. Today, the salesperson may own the entire search. Find possible suppliers. Check availability. Compare alternatives. Confirm whether the vendor is approved. Ask for pricing. Wait for the email to come back. One ISR told me he once booked an Uber Connect to get a part to a job site as quickly as possible. I love that. Customer service is what matters in this business, and it's a core value I want to uphold in my own company.

One reaction to autonomous sourcing was pretty direct: "That's where big value is. I'll pay extra."

The pattern is hard to ignore.

Reading a document and matching a stocked item is becoming expected. Helping a distributor move from "we don't stock this" to a reliable customer answer is harder, and it's probably more valuable.

That's where quoting starts to cross into purchasing, supplier relationships, and real coordination work.

What I Took Away From the Month

Going into these calls, I thought the main AI story was speed. Get the quote out faster. Save the salesperson a few hours. Respond before the competitor does.

All of that's still true.

But it's not the whole buying decision.

Prospects want to know whether the product fits the way their team already works. They want to see what happens when the model isn't sure. They want control over the final decision. They want the company selling the AI to be honest about what's built and what isn't. That's how you build trust.

At the end of the day, they aren't buying AI because they want more AI.

They are buying time back. They are buying less rekeying. They are buying faster answers for customers. They are buying a way to preserve knowledge that took decades to build. They are buying the chance to be the first distributor to respond to the customer. That's where the value is.

The best AI product in distribution should eventually start to feel pretty boring. It shows up where the team already works, handles the tedious part, and asks for help when it's not sure.

That's what prospects are actually asking for.

If any of this matches what you're seeing in your own business, I'd like to hear about it.

-N

---

Based on conversations with distributors and manufacturers in June and July 2026.