We’ve been amazed by the range of ways B2B eCommerce practitioners have used ChatGPT to improve their product data, write marketing copy, scale content creation and on and on and on.
But in discussions with B2B eCommerce leaders, we’ve heard one question about ChatGPT come up over and over: Will ChatGPT replace site search?
On Team Site Search:
Brooke “Lights Out” Logan, Director of Product Management at Genuine Parts Company
Subrata “ChakGPT” Chakrabarti, VP of Product Marketing at Algolia
On Team ChatGPT:
Akash “Kash-Money” Srivastava, Innovation Leader at HP
Colin “The Crusher” Cronin, Director of Digital & eCommerce Channels at Leica Biosystems
KEY TAKEAWAYS
There’s agreement that ChatGPT is remarkable, except… Contenders on both sides talked about the amazing opportunities presented by ChatGPT. The knock on ChatGPT was that (at least for now) it does not include the context and prior purchase history that site search can use to bring back more relevant results. As Brooke Logan said, “Our B2B buyers don’t want to function as ChatGPT prompt engineers to be able to get good results.”
There’s questions about whether ChatGPT can handle nuance. A number of people mentioned that B2B eCommerce search relies on nuance – the ordering part number AGD32487 instead of AGD33487 can lead to the failure of a project. ChatGPT hasn’t shown the ability to handle that level of nuance consistently (yet…).
Even after all this time, site search still isn’t providing a consistently great user experience. Colin Cronin said, “How many websites provide a search experience that’s even at Google search’s level? I’d guess your customers say that site search is worse…Generative AI can deliver a better experience because it can understand intent and interact with the user in a way that site search cannot.”
What if traditional site search was the wrong model all along? We heard the argument that site search is generally successful if you know what you need – if you know a part number and want to order that specific part. As Logan said, “I dare you to type a part number into ChatGPT and see what comes out of it. Our customers type in the part number and they get 1 result – the result they want.” But Cronin suggested that the “search term” model isn’t the optimal way to find products if you don’t know exactly what you want. “People prefer to engage in a more conversational way” to drill down to the item that best fits their needs.”
Like so much in B2B eCommerce, it comes back to the data. The contenders coalesced around the idea that traditional site search and generative AI tools (like ChatGPT) are only as good as the data that powers them. Cronin noted, “It has to do with the data you train your model on. With site search, you still need the correct attributes, index that, and train the model with that data. It’s not a function of search vs generative AI. It’s a function of ‘do you train your system on the right data.’”
It comes down to the paradigm of “conversation” vs the paradigm of “commerce.” Chakrabarti broke down the debate into two differing ways of approaching how to help a customer find something, saying one approach is about “commerce” and one is about “conversation.” He views “commerce” as building an understanding of products and the facets of those products. And once you have that foundation, you can match searches to the right product – it’s not about having a conversation with the customer. He wrapped up the event by suggesting maybe we don’t have to make a choice between the approaches: “We are morphing toward an ‘and’ approach. It’s not either/or. It’s about how we can have a human-like conversation so we can find more products to recommend so there’s a more natural discovery for customers. But we need to understand the specificity in the B2B context and what matters to you as an individual.”