Artificial intelligence is not new, but everything has changed with the launch of ChatGPT, says Nathan Decker, director of ecommerce at ski and sporting goods retailer Evo.

“The scales fell off,” Decker says about realizing the potential of using AI in its business.

ChatGPT, which became publicly known in 2023, is a chatbot developed by research organization OpenAI. The bot uses generative artificial intelligence and natural language processing to answer questions and create content.

“It seems to understand any language you throw at it,” Decker says.

And that means there are more scenarios it can handle, and retailers can be confident the chatbot could grasp the nuance and complexity of language, he says. Evo is No. 455 in the 2023 Digital Commerce 360 Top 1000.

Chatbots vs. generative AI chatbots

Traditionally, most retailers and vendors build customer service chatbots with backend architecture, similar to a decision tree, of all the scenarios of what a shopper could ask and the appropriate responses.

For example, retailers could program a bot with 200 questions a shopper could ask that could be mapped to 200 answers. And each question has 10 variations of ways it could be asked, says Christina McAllister, a senior analyst at research firm Forrester Research Inc.

But that only gets retailers part of the way there, which is why Evo doesn’t have chatbots now, Decker says.

“You’re never going to know every path; a human being can just do it better,” Decker says. “What’s amazing with this new AI is it can handle anything new, can just interpret language and understand.”

McAllister warns, however, that a generative bot could answer any question a shopper asks it, which is not necessarily a good thing. The retailer is still responsible for designing the bot not to answer questions unrelated to the business or ensure the bot has the ability to route certain questions to the correct department, such as a recall or warranty issue.

“The design barrier of entry of having all answers is lower, but the onus is still on the business to make sure the business logic is good,” McAllister says.

Evo hired artificial intelligence vendor Alby to build the bot for the brand to handle customer service questions. Evo provided the vendor with its customer service training materials and additional information about its brand to train the bot. It will cost the merchant less than $1 million to develop the bot and will cost the retailer an annual fee that is within a reasonable range of what other vendors cost, Decker says. For example, some vendors cost $20,000 a year and some are a few hundred thousand, and this is within that range, he says without disclosing the price.

“It’s not more expensive than the most expensive chat feature,” Decker says.

A customer service chatbot could help Evo during its busy season

Having a bot handle customer service inquiries would be a great help for evo, which receives an influx of traffic, sales, and customer service inquiries during the winter months. 40% of Evo’s sales are in November and December, Decker says.

“Every piece of the business has to scale up dramatically for the deluge of inquiries that will occur when we have the compounding effect of Christmas and snow,” Decker says. “The intersection of those two things creates the perfect storm.”

Plus, customer service inquiries continue after Christmas, as shoppers may need to return something or buy additional accessories. What’s more, the ski season continues until March and Evo’s sales remain elevated through the first quarter of the year.

Typically, Evo hires 40 seasonal customer service employees for a total of 80 agents during the busy season, Decker says. It starts hiring in August to have enough time to adequately train employees, which could be two to six weeks.

“Our team has to grow significantly, and we need human beings who can answer phone calls who are competent and who are able to help a customer solve their problems,” Decker says.

Evo’s AI chatbot to launch in July

A bot, however, can handle a lot of the lighter customer service questions, such as, “What is your return policy?” Or, “What are your store hours at a certain location?” The AI bot could also aid with sizing questions, such as asking a shopper her measurements and recommending a size based on a product’s size chart. The bot can also read customer reviews and on-site Q&As and quickly surface relevant information for shoppers, such as if shoppers say a shirt fits small.

“What is your warranty policy and navigating those self-help items, let the bot do that. Those are things that are solvable and don’t require our amazing staff to intervene,” Decker says.

Evo wants its agents to help shoppers with more complex questions, such as, “Which skis are right for me?”

Evo plans to soft-launch the bot in July, during its slow season, on some of its higher bounce-rate pages.

After testing the bot and seeing how it reacts to real customer inquiries, it will then add it to higher-traffic areas, such as product detail pages or search results pages.

To start, Evo does not want the bot to have access to any of its backend software, such as order management, that could see where orders are. That may be down the road, once Evo can test the bot for a while, Decker says.

Evo also plans to allow its in-training human customer service agents to use a version of this bot to look up answers to some service questions when they are first getting started.

Measuring a bot’s success

Evo plans to measure the success of the chatbot both quantitatively and qualitatively. Quantitatively, it plans to A/B test shoppers with a hold-out group that does not have the opportunity to interact with the bot compared with consumers with similar levels of engagement who do interact with the bot. It will measure bounce rates on those pages, conversion rates of the groups and dollars per session of a shopper that interacts with a bot compared with a shopper with the hold-out group.

Qualitatively, it will evaluate if it feels like the chatbot is creating a good experience for shoppers based on overseeing the interactions, shopper feedback and feedback from its staff.

Decker hopes that having that bot will mean it will not have to hire as many seasonal agents. The retailer will still plan to hire in August, but it’s hoping that once it gets enough customer interaction with the bot, it can slow its hiring. While he doesn’t have a hard goal, Decker says reducing its hiring by a quarter, or about 10 fewer agents, would be a good goal.

This sounds about right, McAllister says, saying a bot can reduce contact volumes by about 25%-30%.

Decker doubts a chatbot would ever replace human agents, who may have the depth of experience of using the products and how they feel on the slopes to help shoppers looking for nuances between different brands of equipment.

“It would take a long time to displace our knowledge or staff,” Decker says.

Human customer service agents are still among the most effective method of getting shoppers to convert, according to a Digital Commerce 360 and Bizrate Insights survey of 1,060 online shoppers in January 2023. 40% of shoppers say they are most likely to make an online purchase after a customer service interaction via email, followed by live chat with a human being at 37%. Live chat with a chatbot was among the least likely to have shoppers convert, at 14%.

Do you rank in our database?

Submit your data with this quick survey and we’ll see where you fit in our next ranking update.