This iteration of our interview series features a conversation with Elena Grewal, Head of Data Science at Airbnb, where she talks building a company-wide data culture and what to do before your new data science hires start.
Airbnb is one of the most-often used examples of digital disruption. But beyond simply building a marketplace, Airbnb is accumulating a huge amount of valuable data. How do you encourage a culture of data exploration throughout all parts of the business?
We think of data as the voice of our users at scale. Data scientists play the role of an interpreter — we use data and statistics to understand our users and translate it to a voice that people or machines can understand. And that’s something that we want everyone at our company to do, so that we truly have that culture of listening to our users at scale.
That’s why we decided to develop “Data University” last year, which prepares Airbnb employees to make data-informed decisions through hands-on in-person classes. Our vision is to empower every employee at Airbnb to make data-informed decisions by providing data education that scales by role and team. The program offers a full spectrum of over 30 different data education classes (more than 20 fully built out today), raising awareness and building a data-centric culture at Airbnb. We’ve had over a thousand employees take classes now. That has been huge for encouraging a culture of data exploration.
Then our data science team members are key for encouraging the culture beyond Data U, and being the ambassadors for data throughout the company.
The hardest thing about data is not knowing what questions to ask. What advice do you have for legacy business about how to start exploring data?
Start with your company’s mission — it is critical to have a clear and compelling mission to focus efforts and ensure everyone is working toward a shared goal. At Airbnb, our mission is to create a world where people can belong anywhere. This mission inspires us and guides our day-to-day decisions. It also helps push people towards the right questions; you start to ask questions about whether a given product or decision moves us towards our mission.
You must be in touch with who is using your product and what their experience is like.
Along with a clear mission, user empathy is critical. You must be in touch with who is using your product and what their experience is like. Data plays a key role here in interpreting user actions and patterns, and thus building a picture of the world. Starting with the user experience helps you ask questions about what users see, do, and believe, which is a good space to explore.
Finally, and more tactically, we teach the “5 whys” method, in Data 101, where you ask “why” five times to find the root of the question.
A mantra of the big webscale operations is to “measure everything.” But if you’re an existing organization with manual processes, measuring anything is hard. How do companies start measuring and aggregating data?
The hardest thing to do well in a standardized way is often actually the data generation itself — what we call “logging” of data. It’s easy to say “measure everything” and hard to standardize that process. And yet that can make all the difference for drawing insights from your data. The reason that it matters is this. Imagine that on your mobile app, buying an item is logged as “item_bought”, but on your website it’s logged as “ItemBought”. Then imagine that the name of that action is changed over time; it becomes exponentially difficult to track items bought when names are not uniform, and then change over time depending on the decision of the engineer writing the code. You might see a spike in items bought and it could be just because the logging name was changed and not a true effect. For any website or app you can set up systems to automatically log clicks or page views and to track the measurement using version control. Talk to your engineering team about how to make that happen. You can do this before you hire your first data scientists, so there will be data when they arrive.
So what’s your question for our Pivotal community?
If you had to test one hypothesis by looking at data with the goal of making Airbnb better, what would it be?
If you have a response to Elena’s question, add it as a response to this post. We’ll be highlighting and sharing the best ones.