How Land O’Lakes convinced farmers to use AI—’They’re entrepreneurs at heart’

At Fortune‘s Brainstorm A.I. conference in San Francisco on Monday, Teddy Bekele, CTO of agricultural cooperative Land O’Lakes, and Fiona Tan, CTO of online furniture retailer Wayfair LLC, compared and contrasted how workers at their companies have embraced—or raised an eyebrow at—efforts to introduce A.I. into the supply chain.

Land O’Lakes is using A.I. to approximate the supply and demand of different products at different times of year. The technology has become a tool used directly by the company’s farmers. Farmers see it as assisting their decisions, not replacing their expertise, Bekele says. Yet getting the farmers fully on-board takes some convincing, he says, since planting and harvesting are high-stakes decisions. “Farmers will always try things, they’re entrepreneurs at heart,” Bekele explained. “However, to fully adopt it in their operation, they want to make sure the solution really works.” 

Some A.I. models can seem counterintuitive to farmers at first. Bekele brought up the example of using A.I. models to determine the best locations to plant crops based on climate, topography, and soil. At times, the A.I. suggestion differs from where farmers have planted crops in the past. “On paper, [the A.I. model] doesn’t sound right,” Bekele says. But with some explanation, farmers come around to the idea.

A.I. can also serve as a sort of second opinion for farmers. They input their own data into the A.I. tools and use the system to confirm their own instincts.

Wayfair is a digitally-native company so its employees are fairly open to adopting new tech, yet Tan says that a tech-savvy workforce can become frustrated that A.I. doesn’t move faster. “Sometimes there’s impatience for the models to work immediately,” Tan said. “It’s not like it’s deployed today and it’s all going to work magically,” she said. 

When Wayfair adds A.I. to internal processes, it starts with low-stakes tasks to mitigate the risk of errors and ensures humans are still checking the technology’s work, Tan says. “For example, in marketing, the worst that can happen is you pay too much for a bid, so that’s something we can tolerate,” she said. “Yet other areas, like when looking at images or text for the product to ascertain the quality of the furniture, we’ll have models give a suggestion or recommendation, and humans can go back and make sure it looks good,” she said.