Why This Matters to Distributors: AI is immensely powerful, but distributors need practical advice on how to use it to improve their operations today. Three AI experts At Applies AI for Distributors offer common-sense suggestions based on real-world experiences on how distributors can derive value from AI, while avoiding mistakes.
Like everyone else, distributors are starting to use AI and moving quickly up the learning curve, though not without the occasional misstep. In a wide-ranging session at Distribution Strategy Group’s Applied AI for Distributors conference, three experts offered down-to-each AI advice based on actual experience.
Here are 10 tips from the speakers on today’s Tech Leaders Panel: Dan Kaminstein, senior principal product manager at ERP vendor Epicor; John Murcott, executive vice president of product and strategy at HawkSearch, a provider of site search; and Sam Bobb, o-founder and chief technology officer at Kaavio which offers product content technology.
- Look for your quickest win: Where AI will have the biggest immediate impact varies from one company to the next. For one plumbing and HVAC distributor, a Kaavio client, the greatest benefit has come from being able to onboard more suppliers quickly and sell more of their products. By having more products on its ecommerce site, the distributor boosted sales and increased conversions, Bobb said. Epicor’s Kaminstein said some clients are realizing savings quickly from using AI to check the accuracy of freight invoices. Murcott says AI-driven search can understand relationships between products to recommend upsells, and also access a customer’s history to make it easy for the buyer to reorder while conducting an online product search.
- Start small: “Pick one area where you think you can get ROI from AI and learn how AI plays,” Kaminstein suggested. As companies do that, they learn whether their data is good enough for AI to provide accurate answers in a particular application. He said AI working with bad data will give wrong answers quickly—and confidently—which can undermine trust in AI initiatives.
- How good does it have to be? Understand when data must be perfect and when pretty good is good enough. Bobb gave the example of a Kaavio client using AI to populate hazardous materials documentation: In that case, the information had to be 100% accurate. But, he said, if you’re building a chatbot to help sales reps make recommendations, “it doesn’t have to be perfect.”
- Less data can be better: It’s well known that AI chatbots like ChatGPT that draw information from all over the internet sometimes give wrong answers. An alternative is to feed an AI application only data from well-vetted sources, such as product data sheets, Murcott said. That minimizes risk, he said, because the AI is only drawing from the information on the approved PDFs “rather than all the data on the internet.”
- Don’t let AI become a security threat: Governance and security are important, Kaminstein said. There must be guardrails in place to make sure that employees and others aren’t using AI tools to gain access to information they are not authorized to see.
- Leverage the unique capabilities of AI: In some cases, AI can deliver value in unexpected ways, such as picking out details in an image. Murcott gave the example of a client company whose product descriptions lacked detail but that took great photos of each item. AI could scan each product photo and enhance each description with additional information about the product.
- Bring the truth closer to the customer: Bobb pointed out that in the past product-content systems had to work with structured data, which often meant information about a product often was transcribed from an original source, such as a printed document, into a spreadsheet and later into product software, each time “losing fidelity and adding error.” “The thing that’s revolutionary about AI compared to all previous computing systems is that it can make sense of unstructured data,” he said. That means distributors can use AI to draw data from the original source, minimizing errors from the transcribing of data from one format to another.
- Connect the dots: AI has the power to access vast amounts of data, but AI only works to its full potential when it can communicate effectively with all those sources, Kaminstein said. That means making sure from the start that you have APIs “fully capable” of interacting with the sources feeding data into your AI tools.
- Structure questions to get the answers you want: The power of AI to scan all the data that’s online can lead it to provide answers that are accurate, but not helpful in this instance. Murcott, responding to a question from the audience about how to provide proper context, said it’s important to construct the prompt—the question put to the AI system—in a way that helps it understand how you want it to answer. Give the AI guidance, such as, “You’re a helpful salesperson in this industry. A lot of context comes from the instruction,” he said.
- Don’t make the perfect the enemy of the good: Like other technology, AI sometimes makes a process better, but not perfect. “It may get them to the 10-yard line, but if they don’t score, they don’t want to use it,” Murcott said. He suggests a mindset of, “This is definitely better than before, even if it’s not perfect, let’s move on.” Bobb agreed, saying that if a project is moving in the right direction, keep the momentum going.
Do not miss any content from Distribution Strategy Group. Join our list.
Share this article:
