Here's what Trulia learned from building a chatbot

Be aware of platform limitations, but building a bot is still worth doing

trulia facebook bot

Trulia's Facebook Messenger bot lets users search for properties.

Credit: Facebook

It's a competitive real-estate rental market out there, and Trulia wanted to capitalize on the interest with a new Facebook Messenger bot it launched earlier this month.

The bot lets users search for rental properties and keep up to date on new properties when they become available. 

Trulia's bot came out of a quarterly hackathon project hosted by at real estate tech firm this past May, and the company learned a lot about bot-building. The experience showed that businesses should give bot-making a shot, even if they're not tech companies, said Yardley Ip, general manager for Trulia Rentals.

"Given that the tools are so easy to use, and it's so lightweight to develop [a bot], I think businesses should try it," Ip said. "At least, at minimum, from the customer service angle. Because there are frequently asked questions that users and customers have, and why not use a bot as a way to respond to your users quickly?"

The bot was built with JavaScript, Node.JS, Facebook's Wit.Ai natural language processing tools, and Trulia's proprietary APIs, Ip said.

Trulia is far from the only company looking to take advantage of the Messenger platform. More than 18,000 bots have launched on the platform, according to Facebook. Those include interactive fiction games, travel planning bots, shopping bots, and even bots to help users playing Pokémon Go.

One of the biggest lessons Ip took away from the experience is that any company looking to get started with bots should pay attention to the capabilities available to them before they start building.

Right now, the Messenger platform doesn't support the tracking of user sessions in a way that works for what Trulia is doing. That was a major stumbling block for Trulia employees working on the bot, and one that they only encountered late in the process of building it, Ip said.

In a similar vein, companies should work on collecting telemetry about the ways people use the bot as a way to improve the system in the future, she recommended. Companies should make sure that the analytics system they use (like Omniture or Google Analytics) can be set up to track how users interact with the bot.

The company has had problems with platform instability. When Ip showed a demo of the bot, some queries took multiple tries before generating an answer. Trulia investigated that lag, but found that Facebook, not its service powering the bot, was at fault, Ip said. Facebook didn't comment on the performance issues.

The performance issues that cropped up at that time appear to be resolved, but they demonstrate one of the key issues with operating a bot on someone else's platform: Sometimes users may have a bad experience that's out of a company's control. 

Opinion is split at the company over whether the bot is a Trulia product in itself or a way to get people into the company's user acquisition funnel. But Ip expects to invest more resources in the bot as a way to better reach users. She sees potential for expanding the bot's functionality.

The march toward exascale computers
View Comments
Join the discussion
Be the first to comment on this article. Our Commenting Policies