‘Can I help you?’: How LivePerson decides who’s worth the personal touch

Even if you haven’t heard of LivePerson (s lpsn), chances are you’ve encountered one of its products while browsing online. It’s the company behind many of the pop-up windows offering real-time chat with a representative, as well as other forms of online customer engagement. It’s also a treasure trove of consumer data that LivePerson uses to decide which visitors are worth what type of attention.

Essentially, LivePerson data scientist Vitaly Gordon told me during a recent conversation, the goal of LivePerson is to provide the same experience as shopping in a brick-and-mortar store — only better. Better for consumers, and better for the store. That means knowing who’s just there to browse, who’s there to buy and who might need a little nudge in order to pull the trigger on a purchase. And that means analyzing a lot of data.

LivePerson has approximately 9,000 customers and, Gordon said, their combined traffic is “roughly the traffic of Facebook.” LivePerson monitors much of that traffic to flavor its secret sauce — real-time models that help it decide what type of service a visitor actually receives. If a site is using the full spectrum of LivePerson services, options range from price reductions to live chat to live video. All told, Gordon said, LivePerson adds about 2 terabytes a day to its Hadoop cluster that helps build those models.

Because LivePerson gets paid for incremental sales (i.e., sales that wouldn’t have happened if not for its intervention), Gordon said one critical model determines who’s likely to buy regardless of whether they’re approached for a chat. But that’s only part of the story, because LivePerson also has to best utilize its customers’ resources, especially when there are potentially hundreds or thousands of consumers on a site at a time. If LivePerson’s models determines someone doesn’t need a personal experience, perhaps it will just offer them a slight discount on what they’re looking for. If it’s a big fish, perhaps an uber-personal, but resource-intensive, video chat is in order.

The models rank visitors on a scale from zero to 100 for every interaction the site might offer, Gordon said, and the right interaction might change by the second “because of something you did, or [because] time just passes by,” Gordon said. For example, if someone has stuff in his shopping cart, fills out his shipping information but then returns to the homepage, that’s a sign he’s probably leaving the site. LivePerson will try to reach him before he leaves, Gordon said, but even if it can’t, the site still might be able to personalize an experience if that shopper returns.

And you didn’t think all the text from those chat transcripts was going to waste, did you? LivePerson also has products for analyzing chat transcripts and doing sentiment analysis, even a dashboard for giving customers real-time information on what consumers are talking about and how they’re feeling, Gordon said. This can inform representatives’ decisions on how to respond, or maybe even help them determine that someone isn’t ready to buy and move onto someone else.

On average, he said, LivePerson ends up increasing retail customers’ incremental revenues by 20 percent by increasing the number of sales and the amount per sale. But it’s always looking to improve, which is why it bought Israel-based predictive analytics specialist Amadesa on Wednesday. At LivePerson’s volume — its customers do billions of dollars a month in transactions — Gordon said “even a 1 percent improvement in our modeling is a big deal.”

Feature image courtesy of Shutterstock user alexsalo images.