Why This Matters to Distributors: Companies want to use AI to increase efficiency and better serve customers but worry about the risk and expense. Grainger’s chief technology officer provides tips on how to minimize the downsides of AI while getting measurable wins now and spreading AI expertise that can lead to bigger wins in the future.
A big distributor like W.W. Grainger Inc. is a complex operation. But rather than use AI to transform farflung processes, the company deconstructs a system “bit by bit,” applying AI to optimize specific tasks, in the process helping humans do their jobs better, senior vice president and chief technology officer Jonny LeRoy told attendees June 25 at Digital Strategy Group’s Applied AI for Distributors conference in Chicago.
“We’ve learned you’ve got to break down your problem into smaller chunks,” LeRoy said, and then decide where AI can help.
He gave the example of customer service. As a company that prides itself on high-touch service, Grainger is not automating responses to customer service calls. Instead, it’s giving its human agents “super-powers” by building AI-powered assistants that can quickly retrieve information, enabling team members to serve clients better, LeRoy said.
As it tested AI with tasks like this, LeRoy said it found AI to be a little like interns— “enthusiastic, energetic and fallible.” With experience, it’s fine-tuned AI assistants to better respond to customer queries, such as by asking structured questions about products to help it find an accurate answer that addresses a customer’s needs. That’s improved the performance of the AI helpers and pleased the agents who now feel they can do their jobs better, he said.
While deriving value from initiatives like that, LeRoy said Grainger also pays close attention to minimizing the cost and risks of AI, including the risk of moving too quickly at a time of rapid change in AI technology and the vendors providing it.
How AI helps Grainger enhance its competitive advantages
With AI, and with any technology, Grainger starts by asking how it will enhance what the company views as its core competitive advantage: to know its customers and products better than anyone else.
That strategy has helped build Grainger into a leading distributor of industrial supplies, with $17.9 billion in revenue last year. But it’s also created a company with a wide array of systems, both complex enterprise applications like its SAP ERP system as well as many pieces of custom-built software designed for its specific needs.
He views the company’s approach to a combination of the can-do spirit of Angus MacGyver, the secret agent hero of a TV series of the 1980s and 1990s, with the calm, joyful approach of Marie Kondo, author of a best-selling book about the pleasures of tidying up.
“We want to use the balance of these two views to understand how we can manage complexity within our organization, to keep complexity from compounding into chaos.”
That’s where the bit-by-bit approach comes in. By applying AI to discrete steps within a process, Grainger can realize benefits quickly, without disrupting processes that work well today.
As an example, he described the problem of responding to a query from a potential new customer who wants to know if Grainger can supply the often thousands of different products, it now gets from other vendors. Frequently, he said, the customer will send a large spreadsheet, often with a field that combines a product description and manufacturer brand in a way that doesn’t map easily into Grainger’s product cataloging system, making it hard to respond to such inquiries.
LeRoy says Grainger uses AI to expand that field look at all the elements it contains, helping the company more quickly determine if it has the items the customer requires.
Grainger also is starting to use AI to understand complex search queries on its ecommerce site.
“There’s quite a long tail of searches that are not well-served,” he said. “AI is much better at interpreting complex searches, multiword ones. We’re getting good signals that it’s helping to get much better search results and better metrics. But we’re rolling it out slowly.”
Grainger is also using AI to help in developing software and has seen productivity improvements in early tests, LeRoy said.
Minimizing AI’s cost and risk
LeRoy emphasized that he has encouraged his team at Grainger to experiment with AI, and they’ve responded. “I said I wanted token spend to be a problem, and now it is a problem,” he said, referring to the tokens that are the unit of AI processing, and for which users get charged by the providers of the large language models that underpin AI.
One way he keeps token spending under control is by using the “latest and greatest models, which are the most expensive” for the complex tasks, and older, less expensive systems for easier tasks.
He also warned of the risks AI creates by putting more powerful tools into the hands of hackers—as well as the cybersecurity defenders seeking to hold off the hackers.
While advising against panic over the AI security risks, he said companies must be prepared to move much more quickly in the event of attacks. Whereas once it might have been adequate to patch a vulnerable system in 30 days, today security teams must be prepared to respond in more like 30 hours, or even 30 minutes.
One way to prepare for an attack is to understand what your critical systems are and how you would keep them running during an attack.
“Have you thought about the minimal set of operations and systems you need to keep hobbling along? Do you know what those systems are and how to recover them?” he asked.
He also warned against getting swept away by the substantial number of startups offering AI-driven technology to improve specific tasks now performed by comprehensive software systems such as enterprise resource planning (ERP) and customer relationship management (CRM).
“We’re working with some of them,” LeRoy said. “But be wary if you’re thinking they’re an escape hatch from your major enterprise vendors because half of startups get acquired by legacy vendors. So, stay in good standing with your current vendor.”
In general, LeRoy advised it’s best for companies as they deploy AI to move slowly, making sure they understand their processes and the problems they face before trying to solve them with AI. Or, as he put it, “Rethink your processes before sprinkling them with AI pixie dust.”
Share this article:
