We’re excited to bring back Transform 2022 in person on July 19 and virtually from July 20-28. Join leaders in AI and data for in-depth discussions and exciting networking opportunities. Register today!
SAN FRANCISCO — Enterprises don’t resist AI/ML deployments overnight, and when the decision is made to do so, it involves most of the company‘s C-level leadership and a lot of analyst recruiting data and qualified scientists. It also involves an evolution that can be compared to a person learning to crawl, walk and then run.
None of this is easy or simple, but it becomes necessary in this third decade of the 21st century. Businesses are learning to crawl, walk and run when it comes to using their data to give them deeper insight into their protected business data, all the superfluous data that sits in storage vaults but that are not accounted for, and all their historical data. Do not forget all the social networks and external data (customer reviews, product reviews, etc.) that float in the gigantic universe that is the Internet and affect a company from near or far.
At VentureBeat’s Transform 2022 conference here at the Palace Hotel, a panel comprised of Fiona Tan, CTO of Wayfair; Rajat Shroff, vice president of product, DoorDash; Kevin Zielnicki, Senior Data Scientist, Stitch Fix; and moderator Sharon Goldman, Editor and Writer, VentureBeat, discussed how their automated AI/ML processes help scale and accelerate to market. Their journeys have taken them all from proof of concept to production in a sustainable way.
“At DoorDash, one of our values is that we dream big but start small,” Shroff said. “We also apply this to our AI efforts. We will start by using manual means to do non-scalable things to learn and understand how to find a suitable product for the market. Once we see the signal, that’s when we start inventing algorithms and scaling them.
“For example, when we did our analytics, we found that only about 8% of our business delivered pizza. Some of us thought maybe that was half our business. We realized we needed to be much more accurate in our assessments, so we got the team together and said to ourselves, “We need to get to 99% accuracy”. After a few months of manual data collection annotation, the team found a small sample (identifying a market, a category). Once they received a signal, they expanded the whole project. Once they hit a level of precision they liked, that’s when they passed it on to the ML team. And they started building (AI models).
After a few months of team building and deployment, DoorDash went from 60% accuracy in analyzing its business to its goal of 99%, Shroff said.
How Wayfair uses AI/ML
“We started our (IA) project by looking at the accessibility and quality of data available for the problems we were trying to solve,” Tan said, “so we wanted to make sure we had the ingredients to apply to our project. AI/ML The second consideration we wanted to know was “How much tolerance do we have for wrong predictions?” So the first place we decided to go with our project was in areas of Wayfair that could tolerate wrong predictions.
“For example, we want to use our AI deployments in marketing and advertising auctions (Wayfair). The worst thing that can happen there is that you pay too much for an advertisement, right? “This was an area where I thought we could learn and look into and get quick feedback on results. It’s a bit more difficult to use analytics to determine the quality of an item in our catalog; we wanted to that more humans do this.
Stitch Fix specializes in customization
Stitch Fix specializes in matching its customers with clothes and accessories, so its recommendation engine makes heavy use of AI and ML, Zielnicki said. “It’s really important to get it right when you send people a box of things you think they’ll like when you try them at home,” he said.
Stitch Fix has integrated AI and ML into all facets of its business, Zielnicki said.
“Issues can be as diverse as deciding which warehouse to go to, the ‘selection paths’ within those warehouses, choosing which stylist to match which customer, assembling items from item sets, etc.,” a said Zielnicki. “When we started 10 years ago, we had very little data on our items, our customers. We started with simple systems based on popularity, then moved on to standard statistical models – things like multilevel regression that work well with relatively small amounts of data. As we gathered more data about our customers and accumulated more history, we evolved into collaborative filtering approaches, matrix factorization and, more recently, a sequence-based model which is based on the sequence of interactions a customer has with us through their journey.
“All of this contributes to a more personalized experience for our customers.”
VentureBeat Transform 2022 continues virtually until July 28.