In the rapidly shifting landscape of retail technology, many companies are still scrambling to define their artificial intelligence strategies. However, for home improvement giant Lowe’s, the path to generative AI started well before the public release of OpenAI’s ChatGPT in late 2022.
“We, in 2021, found OpenAI and basically saw what they were doing with transformer models and we got into a very early partnership with them,” said Chandhu Nair, senior vice president of data, artificial intelligence and innovation at Lowe’s, in an interview with The AI Innovator. “That paved the way for us to set up a team around generative AI because we started seeing the promise of what it could do.”
The company’s innovation labs had been on the lookout for emerging technologies that could disrupt retail, housing and home improvement over a three- to five-year horizon. Gen AI fit the bill.
Nair explained that the unique nature of home improvement retail means that customers not only need to have some basic knowledge of repairs or renovation, but they also typically want to visualize the project they’re building. These twin needs are a good fit for the capabilities of gen AI, which can provide DIYers know-how and also create a visual of their finished project.
“Fixing a leaky faucet could mean changing a washer or changing the battery or changing the faucet, so that requires a bit of expertise,” Nair said. “And the second problem … is everything around home improvement is visual” as DIYers want to see how their home will look after their projects are done.
“With those two in mind, we started with some very early experiments,” Nair said.
Those early experiments included Lowe’s product GPT that was trained on its product catalog. This was before custom GPTs became mainstream. After ChatGPT launched, interest in gen AI surged within Lowe’s.
“Everybody wanted to know about what we were doing,” Nair said. “We had thousands of ideas, and we had to pull back and say, ‘How do we really apply a strategy around it and tie that to our growth?’”
Anchoring AI to the ‘Total Home’ strategy
Lowe’s responded by organizing its AI efforts under a broader ‘total home strategy,’ a multi-year plan aimed at growing market share. Under that framework, Nair and his team identified three areas where gen AI could fundamentally change the business: how customers shop, how Lowe’s sells through stores and service channels, and how employees work internally.
“We broke it down to three areas, how we shop, sell, work,” Nair said. Next, they identified the top business functions under each of the three. Then, they picked six Lighthouse areas upon which to anchor their growth strategy and apply AI at scale.
That’s how Lowe’s came to develop its first AI assistant for employees, Mylow Companion, to give them needed expertise to answer customer questions. “If you have an associate working in the electrical area, he or she may not have all the expertise in lawn and garden,” Nair said. Mylow was designed to “simply help an associate get to the right answers if the customer is asking a question.”
But Mylow Companion actually started out as an analyst chatbot that could analyze the store’s sales and other business metrics. A Lowe’s business leader observed that an analyst chatbot could only help the store manager or assistant manager. How about building a knowledge chatbot for all employees?
“If it can help with product knowledge and product availability, it can help all the 250,000 associates,” Nair said.
They rolled out the renewed Mylow Companion in two stores and accuracy initially wasn’t great. “We were about 60% accurate,” Nair said. He asked for help to improve the chatbot. “The only way to make it accurate is for more people to get into it and give me feedback − thumbs up, thumbs down.”
Real-world usage also provided valuable insights. Once ‘Mylow Companion’ reached stores, Lowe’s discovered that many employees were trying to speak to their devices rather than type while they were with customers. That’s when Nair made voice activation a higher priority. “Nobody will want to type looking down when there’s a customer in front of you,” he said.
Building the platform – and governing the risk
As gen AI began to take off, so did the plethora of protocols and tools. Nair decided that Lowe’s had to build a centralized platform that organized model selection and security controls.
“We built a team, hired some specialists to build a platform through which we can scale,” Nair said. That way, “we’ll abstract all of the engineers, all the other developers, all the application builders, from having to worry about what protocols to use.”
Data readiness was equally critical. “We talk a lot about AI and the models, but the more important part is getting the data in order,” Nair said. Lowe’s migrated to the cloud with Google Cloud, investing in building knowledge graphs, context graphs and a semantic layer.
Alongside building the platform, Lowe’s also put a governance process and a ‘transformation office’ in place to vet use cases along four metrics: ROI, technology investments needed, the readiness of the business to change, and brand and reputation risk in experimentation.
“These four parameters had to be key,” Nair said. “If you have an idea, we’re willing to look at all ideas, but those have to score up for you to really be enabled as a project.”
Leading and lagging indicators
Each initiative was broken down by the ‘transformation office’ into leading and lagging indicators. Leading indicators would be the pace of AI adoption, quality of the results, cost, transaction times, system metrics, and the like. These inputs lead to lagging indicators such as productivity ROIs, growth ROIs, and net promoter score (customer loyalty) ROIs.
“Mylow Companion has driven up our LTR (likelihood to recommend), which is our net promoter score, by about 200 basis points across the chain,” Nair said.
The AI assistant has seen fast adoption at Lowe’s, thanks to the effort of the operations team and “how they embraced it, communicated it, gamified it, created incentives for using it,” Nair said.
Lowe’s has also launched an AI chatbot that interacts with customers, called Mylow. Nair said there has been “significant adoption” from customers and the retailer plans to bring in more agentic capabilities to make “commerce much more simple within that channel.”
Nair’s advice to business leaders is three-fold: First, get support from the top and have the boldness to act. “You cannot have analysis paralysis. This is not (done) by democracy. … You have to have your top leadership team help drive the prioritization and get to execution – start somewhere and move.”
Second, “it’s critical to have a platform strategy – whether you buy the platform or build the platform, that is up to you,” he continued. Third, as agentic AI takes hold, note that it comes with inherent risks.
For example, “if you want an agent to reconcile an invoice and cut a purchase order or cut a check, and it makes a mistake, who’s responsible for it? Is it the developer who built it? Is it the business team who’s managing it? Is it the tech team? How do you track it down to that single (errant) agent if you have multiple agents that are in that process?”
“Governance and having a good control and policy plan,” Nair said, “is a critical factor for success.”
Be First to Comment