Amazon. Netflix. Recommendations are core to some of the web’s most popular platforms, and artificial intelligence helps drive them. And it’s good for more than just queuing up gift ideas and movies.
Today at VentureBeat’s Transform 2018 conference, Stewart Rogers, director of marketing technology at VentureBeat, spoke with Ryan Eberhard, vice president of product at ZipRecruiter; Armita Peymandoust, vice president of product management at Salesforce; Tim Correia, senior vice president and general manager at Trulia; and Olly Downs, founder and chief scientist of Amplero, about AI’s role in product recommendations.
ZipRecruiter, one of the largest online employment marketplace in the U.S., leverages AI to streamline the candidate selection process. San Francisco-based Salesforce uses it to deliver more than 1 billion recommendations to its customers every day, via its Einstein platform. Home-buying platform Trulia taps AI to surface useful information for home buyers. And Amplero uses it to fuel one-on-one customer interactions.
Trulia recently launched Trulia Neighborhoods, an aggregated dashboard of drone footage, photo galleries, reviews, photography, and information about neighborhoods and communities that algorithmically analyzes more than 16 million data points. With the help of an AI-informed scoring system, it surfaces the top 20 hotspots within commuting distance and automatically pulls out top reviews, photos, and safety information for home searchers.
“We know the category,” Correia said. “As we look at these things, we find ways we can drive unique recommendations and value that no one else can.”
The key, he said, isn’t just “setting and forgetting” an autonomous system. Instead, it’s being proactive about the output it’s producing.
“As you’re getting signals back from consumers, you have to understand … how to respond to the engagement levels you’re seeing and make adjustments,” Correia said.
ZipRecruiter‘s machine learning algorithms inform candidates of new job postings via email and text messages. In June, the company announced Candidate Calibration, an AI tool that has recruiters rate potential matches for a job. For every applicant they rate positively, ZipRecruiter surfaces the job posting to other, similar candidates in its database of 10 million users.
Responding to a question about the potential for algorithmic bias, Eberhard took an optimistic view. “It’s kind of a garbage in, garbage out scenario,” he said. “You need to understand the biases that are latent in your training data. The ideal solution has you weighing the pure data that you extract … and when you do that, you can remove [things like] gender so that the algorithms don’t have that at its fingertip.”
Candidate Calibration drove a 50 percent increase in the number of employers that found a candidate they liked enough to invite him or her to apply, ZipRecruiter said.
AI is a core part of Salesforce’s business. In its third annual State of Sales report this year, released in May, it forecast that AI — specifically as it relates to opportunity insights, lead prioritization, and guided selling — will become pervasive by 2020.
Nearly half of the more than 2,900 sales professionals surveyed said that AI has a role to play in guided selling capabilities, like opportunity rankings (i.e., highlighting customers with high sales potential) and suggested next steps (tips and ideas to help clinch a deal). Across the board, respondents to the Salesforce survey said they expect AI to substantially impact sales report forecasting and guided selling.
“Ten percent of clicks are 20 percent of revenue, which speaks to how important these product recommendations are,” Peymandoust said. “When you talk to actual consumers out there, they talk about trust … but they also want personalization. It’s a dilemma we have to address very carefully.”
In October, Amplero updated its software to let companies import their own machine learning models. Its platform collates insights and, in tandem with its in-house algorithms, models many different customer interactions at the same time to deliver better insights to businesses.
According to a study it conducted in partnership with researchers from the Columbia Business School and HEC Paris, it found that the ripple effect of personalized marketing campaigns on non-targeted consumers within the targeted consumer’s network caused a 28 percent lift.
“It’s very common to view short-term direct response as a proxy for long-term interaction,” Downs said. “For a business, having the capability to isolate and control groups so you can measure [something like a] KPI is invaluable.”