How to not get Netflix'd: Keller Williams uses AI and its vast scale to modernize

Rather than cede ground to newer real estate companies, Keller Williams' chief product officer Neil Dholakia explains how the 36-year-old firm is turning its size and trove of historical data into a technological advantage.
Written by Stephanie Condon, Senior Writer

Like just about every other business sector, the real estate industry is quickly learning that data is one of its most valuable assets. After 36 years in the business, Keller Williams has plenty of data to work with -- as well as a vast network of real estate agents it can turn to for help keeping that data fresh and useful. 

When Neil Dholakia joined Keller Williams as chief product officer in 2017, the company was already intent on modernizing. 

"We didn't want to be the Blockbuster to Zillow's Netflix, or the Blockbuster to Redfin's Netflix," Dholakia told ZDNet. "You name the company that started up in the last five years, or 10 years for that matter -- they all look at us as an old guard company, and we're the Blockbuster that they're going to Netflix."

Founder Gary Keller wanted to transform his company from being the largest real estate franchise company in the world to a real estate technology company, Dholakia said -- in effect, the goal was "to Netflix ourselves."

Dholakia gave ZDNet an in-depth look at the product strategy he's leading to leverage all of the competitive advantages that Keller Williams has as a well-established, large firm. The company is building a comprehensive set of AI-powered applications, for real estate agents and home buyers, that build on top of each other. 

Here are the highlights of that conversation: 

Leveraging the company's scale to embrace AI

Keller Williams is building a platform, Dholakia said, that will bring together its 35 years of historical data -- as well as the data continuously generated by the 180,000 Keller Williams agents around the world. 

"We use the historical data, the new data that we're collecting, analyze that, and we've created a system whereby we've identified multiple AI outcomes that leverage some off-the-shelf technologies -- like Google's AutoMLDialogflow, and things like that -- and we've built up our own data science force here to be able to come up with what machine learning model building blocks do we need, that we want to stack together to help us achieve these outcomes?"

There are a couple driving goals, Dholakia said: to help agents be a better fiduciary to their clients, and to create a delightful experience for consumers. 

"That's our strategy from a technology perspective -- get everybody onto one platform and use AI and the data advantage we have to create differentiation on that platform."

Using off-the-shelf AI tools to inventory housing

To get real estate agents on the platform -- as well as data about new housing inventory -- Keller Williams created a video app for agents. 

"We knew that we needed to be able to train a system to identify pictures of houses," Dholakia  said, "What type of house is it? What type of facade does it have?... What's an entryway, what's a kitchen, what type of appliances are in the kitchen, what type of countertop is in the kitchen?"


Keller Williams used Cloud AutoML, the software that automates the creation of machine learning models, to build the app. It lets agents record a video as they walk through a house, and it automatically identifies and tags the house's characteristics. Agents can use the app to annotate the video as well. Within just a few minutes, the app will generate a landing page or a mini-website for the real estate listing. 

"So that's instant value for an agent that they get just by using the system," Dholakia said. "They train our AI underneath and, because  we have scale of agents, we're getting more training data."

Building a cascading set of AI-based tools

Keller Williams can use that data it's collecting with this app to build even more applications and create more value for real estate agents and consumers, Dholakia explained. 

For example, now that Keller Williams can identify characteristics in houses, it can show the system pictures from MLS to catalog listings -- whether or not their agents have seen them. Meanwhile, home buyers can tag their favorite home features, letting Keller Williams create a profile of their preferences. Keller Williams can combine a buyer's preferences with the preferences of their co-buyer, such as a spouse, to produce search results superior to any filter-based results. Along with home characteristics, Keller Williams is incorporating into its applications data about neighborhoods. 

"We're facilitating a better search experience than you can get anywhere else because we've leveraged our agent base in just one use case of training our system to identify what's inside of homes," Dholakia said. 

Using Keller Williams agents as "lab audiences"

Rather than move its agent-focused applications into production in the traditional manner, Keller Williams has a "labs process" that uses its network of real estate agents to roll out apps incrementally. Real estate agents offer feedback on the apps, and with each iteration, more agents are included in the process. 

The process typically starts with as few as 10 to 15 agents and then progresses to 150 people, to 1,500 and then to 15,000. The company has roughly 30 to 40 lab streams going at any given time. Its biggest at the moment includes 60,000 agents. 

"As they progress through our system, we'll combine [labs] to create bigger and bigger pieces of functionality," Dholakia said. "So we don't like to say we've ever put anything into production and it's GA -- it's always in our lab, and we talk about lab audiences."

The process can be valuable to engineers -- as well as product managers, designers and others -- who may not be familiar with the way the real estate industry works, Dholakia said. 

"The first 15 participants in any lab... they are agents that are actually going to use the system," he said. "And before we even start writing code, before engineers touch a keyboard, we have paper in front of those agents and we tell them, 'Hey, how would you like this feature to work? What do you want to see on a screen? Draw it for us.'" 

More AI coverage: 

Editorial standards