Generative AI has been the hottest thing in technology since the introduction of ChatGPT in 2022. This year, gen AI will drive $200 billion in global investment, Goldman Sachs says—even though studies suggest most companies haven’t profited much from their gen AI projects, at least so far.

Given the significant cost and risks associated with this new form of AI, should distributors invest in it? Some distributors that have jumped in say the answer is yes, because gen AI can do things that no other technology—or human beings—can do. But they make clear that projects must be thought out carefully and tailored to match each company’s needs and capabilities.

Among the distributors already using generative AI is Parts Town, which supplies equipment and replacement parts to restaurants. It’s developed a tool called PartPredictor that helps technicians quickly identify the part they need to fix a problem. It’s also enhanced its live chat with a gen AI version that’s produced a 900% increase in conversion—customers using the tool are 47 times more likely to purchase than others—and increased average order value by 20%.

And this is only the beginning, because the large language models that underpin generative AI are improving at a rapid pace, says Emanuela Delgado, The Revolution, group vice president of growth and innovation at Parts Town.

“LLMs are much, much better at reasoning and creating content than they were a couple of years back, and in six months, even two months, they’ll be better than they are now. That’s how rapidly things are moving,” Delgado says.

As part of that rapid evolution, software vendors are increasingly incorporating AI features into their products, rather than offering them as add-ons, says Janet Zelenka, a consultant who formerly was both chief information officer and chief financial officer at distributor Essendant and Stericyle, a provider of medical waste disposal services.

Companies that have deployed modern software will be able to take advantage of these innovations in coming years, giving them an edge over those using older, internally developed code. “Distributors with a technology debt will fall farther behind those who have upgraded their systems,” Zelenka says.

Those more technologically advanced companies are the best position to take advantage of the enormous potential of generative AI, which can create original text, images, video, and software code by analyzing and manipulating vast quantities of data. But gen AI projects are costly, introduce new security risks and can radically impact a company’s workforce—replacing some workers while requiring others to learn new skills.

In short, we are at the beginning of what many believe will be a fundamental transformation of business operations. And that poses a big challenge for leaders of all companies, including distributors.

Fast adoption, rapidly improving results

A global survey of 1,500 business executives by consulting firm McKinsey and Co. in July 2024 illustrates how fast companies are moving to test and deploy generative AI.

78% of respondents in the July 2024 survey said their companies were using gen AI in at least one business function, an increase from 72% earlier in the year and 55% a year earlier, according to McKinsey’s “State of AI” report released in March 2025. To be sure, it’s early days. More than 80% said gen AI has yet to significantly impact profits and only 1% called their projects mature.

But the results are improving, and fast. For example, in supply chain and inventory management 67% of those using gen AI in those functions reported improved revenue generation in the second half of 2024 versus 53% in the first half, while 61% saw lower costs in the second half of the year versus 46% in the first half.

Big companies, those with annual revenue above $500 million, are deploying this new form of AI more aggressively: 52% of larger companies have created resolute AI teams to drive adoption versus only 23% of smaller ones. But, in a sign of how early adoption is, only 18% of big companies have developed KPIs to measure results and 16% of smaller ones.

Regardless of size, the most common applications were in marketing and sales, product and service development, service operations and software engineering. But McKinsey notes, there is considerable variation by industry: “Organizations are applying the technology where it can generate the most value,” the report says.

Standardized data makes gen AI work for Parts Town

Two distributors, Parts Town and Southern Glazer’s Wine and Spirits, are examples of companies introducing generative AI in ways designed to meet their unique needs. At the same time, they emphasize they are implementing technology carefully, mindful of the need to protect company data and keep costs in check.

Parts Town uses generative AI in several ways, including its enhanced live chat system that can answer questions in 27 languages and is available on the company’s website, PartsTown.com, at all hours of the day and night.

But its most ambitious project is PartPredictor, which began building in 2022. The goal is to enable customers to access PartsTown.com and quickly identify the parts they need to fix a particular equipment problem.

Aiming to use gen AI to address that problem, the company built its large language model, the dataset AI uses to provide answers to customers’ queries—with information gleaned from tens of millions of work orders for previous repairs.

Now, with PartPredictor, Delgado says, “you can say select the equipment issue, for example, my fryer isn’t heating or there’s an oil leak, and the technology will drive you to the accurate part based on your exact piece of equipment and the issue that’s happening. We wouldn’t have been able to do that successfully without generative AI.”

To get started, Parts Town collaborated with an outside firm = to name to modify a prebuilt large language model to the distributor’s needs. As proof of concept, Parts Town started with parts from a handful of manufacturers.

One big problem it faced was that the information in its millions of work orders was not in a standardized format. Since AI—both the “traditional” versions and the newer generative AI tools—work based on pattern recognition, if data is not in a common format, AI will struggle to see the patterns.

According to Delgado, Parts Town used a second large language model designed to turn unstructured into structured data to cleanse the data and create a dataset AI can work with. The fact that there are now LLMs for specific purposes like standardizing unstructured data illustrates the rapid evolution of AI.

It took about six months to input the data from the first small group of suppliers and another six months to incorporate parts from a larger group of manufacturers, Delgado says. “But those cycles are getting shorter and shorter because the LLM is getting stronger,” she says.

Delgado says building the initial model cost “a couple of hundred thousand dollars.”

The cost of generative AI projects varies widely based on how much data is involved, the quality of the data, transaction volume and other factors. But most estimates say even small projects cost tens of thousands of dollars, and larger ones can cost hundreds of thousands or millions of dollars.

Is PartPredictor hallucinating?

An important part of the pilot was ensuring that PartPredictor was correctly identifying the required part, and not “hallucinating” to use the term that describes gen AI systems like ChatGPT making stuff up. That took people with the expertise to assess PartPredictor’s answers and to correct it when it’s wrong. That training process makes AI systems increasingly accurate.

“Validating is the most important piece,” Delgado says. “You need a person with the expertise to say, ‘Yes, this is accurate, and we can move forward,’ or provide training data so the AI can continue to improve.”

While she would not provide a specific gauge of PartPredictor’s accuracy, she says she’s “very confident” it gets high marks. One indication of that, she says, is that some of the distributor’s suppliers and customers are using the system themselves, along with Parts Town’s own service department.

One way Parts Town is measuring success is by the “fix rate,” the percentage of the time a technician arriving at a customer’s location can fix the problem without returning. “We want to make sure they have the handful of parts they are most likely to need for the stated issue on their van or truck,” Delgado says. “Using it prior to going out on a service call should help improve the percentage fix rate.”

Getting ahead of inventory issues at Southern Glazer’s

While Parts Town is using gen AI to identify the right part for each equipment issue, at Southern Glazer’s Wine & Spirits the aim is to prevent problems that arise when a supplier runs short of a product the distributor ordered or does not have it in the right warehouse to ensure on-time delivery.

“Before we were always looking backwards, always running out of stock and expediting,” says Diego Fonseca, vice president of supply chain and logistics. The aim of the new AI system under development is to create a more efficient supply chain that can fix problems without human intervention, he says.

Fonseca’s team is building on the success of a demand-forecasting tool it launched last year using traditional AI features. That system incorporates three years of sales data to provide monthly forecasts of customer demand. Fonseca says the system has improved forecast accuracy from an average of 65% to 73% in recent months, an improvement worth millions of dollars to the company.

With generative AI he hopes to achieve further gains by preventing inventory problems before they occur. In the project now being scoped out the distributor will identify suppliers that frequently force SGWS to manually change orders because of inventory issues. The Gen AI system will be trained to spot potential supplier problems and take the necessary steps to ensure SGWS can fill out its customer orders and not lose sales because a supplier can’t provide the merchandise.

“Let’s say I have an out-of-stock today, or logistics delays, or overshipments, or a supplier doesn’t have a product. Generative AI can sense I have a problem in the supply chain, and give you an alert that says, ‘I think you should be looking at this, based on what I’ve learned and the variables I’m tracking, you’re likely to have an out-of-stock,’” Fonseca says.

Today, responding to supplier shortages requires manual intervention by experienced logistics managers. “Generative AI will automate all that,” Fonseca says. “It’s not hard because we have all that data. It’s just training the model to give the right answers. Now it takes people with 25 years of training.”

Fonseca hopes to launch a pilot of the new gen AI system later this year.

Zelenka calls the SGWS concept “brilliant,” in part because the company plans to start with a pilot. From her perspective as both a tech and finance leader, she says it’s best to start small with AI.

“If you’re going for the big idea and spending hundreds of thousands of dollars or more, you’re going to spend more time to build the business case and monitor results, versus, ‘Let’s try this with three people using this new tool we bought cheap and we either abandon it or advance it.’”

What’s more, Zelenka says, the SGWS concept addresses one of the best ways for distributors to increase cash flow: by generating the same sales volume without tying up cash in more inventory. “That’s a velocity of cash and sales business case,” she says. “You could easily come up with an ROI on that.”

Will gen AI replace people?

While the aim of the gen AI project is to reduce the tedious work logistics managers do today to deal with supplier shortages, Fonseca says the goal is not to reduce headcount. Rather, he wants his team to have the time to collaborate with suppliers, logistics providers and customers to create a more efficient supply chain. Delgado at Parts Town also said the company’s gen AI projects are not aimed at eliminating jobs.

But many companies do foresee generative AI reducing jobs—at least in certain departments. According to the July 2024 McKinsey survey, 31% of companies expect gen AI will reduce headcount in three years, while 19% expect it will require more employees.

Whether jobs are added or cut depends on the department. For example, 48% of respondents to the McKinsey survey predict they will be able to cut positions in service operations while 21% expect to add jobs in that area. But in IT, only 25% expect gen AI to decrease headcount while 41% think they will need more IT workers because of AI.

Those disparities reflect the differences between artificial and human intelligence, says Zelenka, who will be leading a panel on the personnel implications of generative AI at Distribution Strategy Group’s “Applied AI For Distributors” conference in June in Chicago.

“AI can replace jobs that are just repetitive and based on pattern recognition, based on a path of, ‘if this, then what’s likely to happen next?’” she says. That includes responding to common customer questions and certain coding tasks.

But, while AI can only make decisions based on what’s happened in the past, Zelenka says people can see and respond to what’s new. “The real spark of human intelligence,” she says, “is coming up with the idea nobody has thought of, approaching the problem in a novel way or seeing the big picture.”

For distributors, she says, that may mean an experienced customer service representative who’s been too busy to get away from her desk in the past now being able to go out into the field with service technicians and using her product knowledge to help them do their jobs better. Or it may be website personnel being retrained to monitor website traffic and keep up with the ever-changing ways hackers use generative AI to steal sensitive information or disrupt ecommerce operations.

“It’s an exciting time,” Zelenka says. “But you have to plan for it.” That’s the challenge posed to distribution executives by the coming upheaval in business processes fueled by generative AI.