Benj Cohen grew up going on ride along sin a sales rep's car.
His father runs Benco, a dental distribution business that's been in the family for three generations, and believed that "the magic happens on the customer's ground." Benj remembers visting dozens of dental offices, watching a rep work, listening to customers, and asking questions.
A family trip to Korea was also a vendor visit. Benj was when he sat across from the CEO of Vatech, an early entrant to the U.S. dental X-ray market. "I remember distinctly the guy saying, 'We will be the number one player in this market,'" Benj said. Years later, Vatech had become one of the largest suppliers to his family's business.
Distribution, curiosity, and applied math eventually led Benj to start Proton.ai, the AI platform built exclusively for distributors.
For this episode of *In the Mind of a Distributor*, Dasha Shakov turned the microphone around. She interviewed Benj about growing up in the family business, the model release that changed how Proton builds software, and where AI creates real value for a distributor.
Below are four lessons from the conversation.
Lesson 1: Stay close enough to the customer to find the next business
Benj's path into applied math started with a semi-fake internship.
He was rowing for the junior national team in high school. The schedule left room for little beyond rowing, eating, and sleeping, so Benj asked his father for work he could do remotely. His father sent him to the head of data analytics within marketing at the family business.
That's where Benj got his first hard problem: how to predict customer churn from a large data set. Benj downloaded RStudio and got started.
"What I found so satisfying about that summer was I was using math," Benj said. "I was able to follow my curiosity to explore new technology, explore the data, and then actually deploy something that someone would actually use on the other side."
The work stayed close to a real user and a real commercial problem. Benj carried that approach through college, where he worked with manufacturers on models that predicted which products a customer might buy. His father saw the larger opportunity first: "You should do this for distributors. This is way more interesting than doing it for a manufacturer."
The project became Proton.ai.
The company changed shape several times. It started by using transactional and customer data to tell sales reps who to call and what to sell. Customers began using it like a CRM, so Proton built a CRM around the distribution workflows traditional CRMs missed. Agentic AI then made it possible to connect more work across the business.
The common thread was proximity. Benj sat with the person doing the work to learn the process, watching where it broke and building against that problem.
If you cannot name the person doing the work or the customer outcome you want to improve, keep looking for the right first project.
Customer proximity is also why he only half believes the standard founder advice to work on the business instead of in it.
"My experience has been I personally learn the most by being super, super, super close to the problem and the customer," he said. "If I'm just working on the business, somehow that doesn't capture where I learn the best, which is going and visiting the customer, going and talking to the person who's closest to the thing and trying to understand the problem."
Before you schedule a vendor demo, sit with the rep, customer service agent, buyer, or pricing manager. Find the work that takes time without improving the answer a customer gets.
Lesson 2: Connect the four profit levers
Benj has a simple way of describing the economics of distribution. There are four ways to make more money:
- Increase price.
- Improve product mix.
- Sell more.
- Reduce operating expense.
"That's basically it," he said.
Most distributor systems split those levers apart. Your sales activity sits in the CRM. Pricing rules live somewhere else. Purchasing and inventory run through the ERP, while rebates may depend on another set of files and people. The decisions affect each other even when the systems do not.
If a sales system knows what a customer is likely to buy, the same demand signal should improve inventory purchasing. Better purchasing means more of the right inventory and less dead stock. Inventory position and willingness to pay affect pricing. Rebates can change which product produces the best margin, which should change what the rep sells.
"All these levers are fundamentally related," Benj said.
The software problem is coordination. Your people should not have to act as the integration layer between sales, pricing, inventory, and rebates. An AI platform can read the systems you already run and act across the same data, with a person reviewing the work where judgment belongs.
When you evaluate an AI project, you should be able to name the lever it moves. If the project does not increase price, improve product mix, sell more, or reduce operating expense, it is hard to justify and harder to scale.
Lesson 3: AI-drafted emails are just the beginning
Most people first meet generative AI through a blank text box. They ask it to draft an email, summarize a document, or answer a question. Benj thinks that experience sets the ambition too low.
"I think that's missing the point," he said. "Instead of just drafting an email, you can take a workflow fully end to end. The opportunity is to actually do the whole process end to end."
Proton went through the same learning curve internally. The team first added coding agents to its existing software development process. The agents made individual steps faster, but the process around them stayed the same.
Then a model release around Thanksgiving 2025 changed what Benj thought was possible. He built Proton IQ, an internal application that pulls together Gong calls, CRM records, email, and Proton's knowledge base. A team member can ask about customer feedback. Agents analyze the customer record, pass the finding into the development process, and carry the work toward deployment.
Proton changed its software development process around that capability. A human not only hand-rolls custom code, but now manages and reviews work completed by several agents.
"We had to go through the work of rethinking how the process worked in a world where agents can do tasks end to end," Benj said.
Adding AI to one step gives you a faster version of the same workflow. The larger gain comes from rebuilding the workflow around what the agent can complete. Hook it into the right data and point an expert to verify the results.
Take a common distribution task such as preparing for a customer visit. A chat tool might summarize notes that a rep finds and pastes into a prompt. An agentic workflow works different. It pulls account history, finds reorder gaps, checks open quotes, and prepares the account brief before the rep leaves. The rep reviews the process and decides what to do.
The unit of change is the workflow.
Lesson 4: Your seasoned reps may be your best AI users
You may assume the most tenured employees will be the hardest people to bring into an AI rollout. Benj is seeing the opposite.
Traditional software has grown harder to use as vendors add features. "In Salesforce you can literally do anything, and also it's really hard to do anything," he said. Every new capability adds another field, menu, or sequence of clicks for the user to learn.
Agents change the interface. A rep asks for the outcome and leaves the navigation to the agent.
"They just ask the agent, 'Can you go take care of these five things for me?' And the agent does the hard work of doing that," Benj said.
Seasoned reps already know what a good answer looks like. They know which customer details matter before a visit, which product substitutions are realistic, and when a recommendation makes no sense for the account.
"Some of the biggest users of some of our agentic products are actually seasoned reps because they have the advantage of knowing what they want," Benj said. "They're like, 'I'm about to go visit this customer. Here's the shit I need to know,' and then the system just gets it for them."
The practical move is to recruit experienced operators into the first pilot. Ask them to define the outcome, test the agent's work, and identify the points where a person needs to review or approve. Their experience makes the workflow better, and their credibility gives the rest of the team a reason to try it.
Where should a distributor start with AI?
Start small enough to prove something.
You can find an AI use case in almost every function. Product data needs cleaning. Sales teams need better account coverage. Inventory decisions still depend on gut feel. The size of the opportunity makes it tempting to launch several projects at once.
Benj recommends a sequence instead. "One of our values is think big, start small," he said. "Think about the bigger picture of all the different ways you could leverage AI in the business, but start with some small use case, prove value with that use case, use the value from that use case to pay for the next thing, and so on down the chain."
Pick a workflow with a clear owner, data you can reach, and an outcome the business already measures. Prove that the new workflow works. Then carry the learning into the next use case.
For sales, the first use case may be building a call list from reorder gaps before the rep starts the day. For an operations team, it may be finding a high-cost administrative workflow that an agent can complete with a person approving the result.
The first project has one job: earn the second project.
What stays the same in distribution
Benj expects distribution to play the same role 10 years from now. Suppliers will still need a path to customers. Customers will still need local knowledge and product expertise in someone who makes the problem right.
The work around those relationships will change.
"I think we will look back 10 years from now and think, 'Oh my God, I can't believe that we had human beings doing some of the work that today is manual,'" he said.
Benj frames the change as an OpEx question: what percentage of the dollars you spend actually creates value for the customer? Entering data into a system does not. Calling several vendors to track down an answer takes time. The answer matters; the hunt does not.
"We'll be spending the same dollars, but just doing activities that make the experience for our customers way, way, way, way, way better," Benj said. "We'll be like, 'I can't believe we used to spend dollars on these tasks and activities that don't actually matter for our end customer.'"
One thing Benj is still trying to figure out
The podcast ends with a question about curiosity. Benj went deeper than software or distribution.
"The thing that I'm really curious to understand better is how to think about feeling satisfied in your life," he said. "So much of culture, and especially in the United States, is all about how you've got to figure out how to make more money. That's the thing that will make you more satisfied. And my experience so far is that's not really true."
He does not have a neat answer. "I'm still very much trying to figure out what is it exactly that gives me that deep sense of satisfaction," he said. "I suspect I'll be curious about that for my entire life."
Watch the full episode of *In the Mind of a Distributor* to hear Benj explain Proton's origin, how agentic AI changed the company's product strategy, and why he still learns best beside the person doing the work.
.png)
