For Airbnb, data democratization is a strategic advantage

Tools to democratize Airbnb data

Jeff Feng

Airbnb, based in San Francisco, has put a premium on data analytics from its 2008 start, consistently using data to drive its products and customer interactions, Feng said.

"We characterize data as the voice of our users at scale, so data becomes a way of understanding our user," he explained.

As such, the company prioritized investments in its data infrastructure, starting with a single-source truth first built in Hadoop and Presto and later Spark Streaming and Druid.

We characterize data as the voice of our users at scale, so data becomes a way of understanding our user.

The quality of Airbnb's data can be traced to the company's building a strong data team from its early days and continuing to nurture that team, even as it opens up data and tools to its employees throughout the various business departments.

"Data is such a critical area of expertise for our company that we decided to invest heavily in building the majority of our data infrastructure and data tools," Feng said.

In its push to democratize Airbnb's data, the company also built its own user-facing data tools, including a data portal, which Feng says is essentially a search engine for data, visualization and data resources. Workers can search for any data table, visualization or knowledge posts (something akin to research by a data scientist) then surface it through the portal and save it to a team or personal file.

The company built and used an open source tool called Superset, a data exploration and visualization platform that's now an Apache Software Foundation incubator project. The company also uses Tableau Software for data visualization.

Not all Airbnb's efforts are around granting access, however. Feng acknowledged that data leaders have put up guardrails and barriers to control who can access sensitive and regulated data.

"But as a general philosophy, we make data centralized and [make] sure the core data is clean and accessible," he said.

At Airbnb, Data University enrollment soars

Airbnb also created a data training program for its employees.

To truly get to democratized data, Feng said company leadership recognized that it was important to give employees the skills they need to actually use Airbnb data to answer questions -- something that workers today aren't universally trained in nor prepared to do, despite the growing hype around the power of data and analytics.

Data University, started in November 2016 by Feng and Airbnb senior data scientist Erin Coffman, is designed to provide data literacy from the elementary level (100) on up through more advanced 300-level lessons designed for heavy data users. The Airbnb data program runs classes for employees, with each in-person class running one to two hours. Airbnb has 30 different classes (all free) taught by some 40 Airbnb data scientists and data engineers serving as faculty members. More than 900 employees had taken classes by August 2017, with each student taking an average of four to five classes.

There's been a strong return on such efforts, with the percentage of employees using data and data tools on a weekly basis now at 45%, up from 23% prior to the data literacy program.

Feng said he recognizes that Airbnb, being born as a digital company, has some advantages when it comes to its data efforts. It did not have to contend with older tech infrastructure and built-up data siloes that create obstacles at older companies as they try to use their own data.

"More legacy companies had to make significant investments in their tooling and infrastructure to be able to realize some of the more advanced use cases we see today, especially in artificial intelligence, machine learning and experimentation," he said.

Advice for legacy companies looking to democratize data

Some companies have tried to leap-frog to data-powered, cutting-edge technologies such as artificial intelligence and deep learning only to have to "step back and realize there are so many other needs to address first before you can address those areas," Feng added.

Doug Henschen

Analyst Doug Henschen said organizations looking to democratize data should be prepared to face additional challenges.

"There's getting the right tools and capabilities in place, but also guiding users on best practices," said Henschen, vice president and principal analyst with Constellation Research Inc.

At some organizations, existing reports or analyses can be adapted or extended for use by non-experts. A dashboard built into a particular tool, for example, could help the rank-and-file start to understand how to see and use data to do their jobs.

The organization's business intelligence team can also play an important part in fostering data literacy, alerting users when a report they're trying to create already exists, or that the tools they're trying to use are inappropriate, or that the problem they're trying to solve is easily eliminated upstream by changing a process or data connection. This kind of guidance doesn't just happen, Henschen explained. IT needs to develop processes that help employees access and derive insights from data.

Business payoff for data efforts

Airbnb's efforts to put data in more hands are paying off, Feng said. One success case? Workers use Airbnb data tools for A/B testing and for experimentation on new customer-facing functions and features. And he noted that the HR team uses it to track and analyze their spending on recruiting efforts to better understand what efforts to find talent have the best ROIs.

Feng said Airbnb's data democratization efforts enable all workers, including data scientists, to "upskill," that is, to work higher up the chain of their professional capacity. He explained that by giving workers these data tools, they can focus on making decisions rather than on seeking out information. The same goes for data scientists, who can focus on more complex questions instead of running reports for business-side colleagues.

Henschen said the most immediate benefit of self-service data is a gain in productivity, but he also sees additional ROI from such initiatives.

"Data experts and power users get out of the business of running and tweaking existing reports and wrangling data and, instead, can focus on building out new analyses that might drive innovative new parts of the business," he said.

Democratizing data also means business users no longer wait in line, only to find that their needs have changed once IT delivers on their request. Instead, they are able to explore and ask new questions as they dive into the data and seek answers to their business questions.

And that self-service data exploration will pay dividends, Henschen said. "As more business users are able to answer questions and support decisions with data, there's the cumulative advantage of better decisions that are supported by data, and they can bring incalculable returns and advantages to the business."