Airbnb Stories about Customer Problem Solving Analysis with Technology

Problems that look easy but actually will give complexity, if not able to be solved agilely and quickly, related to customers. Although the moving average fluctuates, startups that already have trusted services will generally still record a fantastic number of users. Stories about how is the team engineer and Airbnb's data managing their customers with technology is an interesting thing to observe, as a smart strategy to accommodate customer service that goes into the system.

When there is a question regarding data handling, the answer is usually: "Of course we have a special database that is used to store consumer issues, but we don't have a way to address those issues (systematically) yet."

In general, data such as customer complaints will only be stored, or may become a file to-do list which must be managed one by one by the division customer service. But unfortunately not many are able to map these problems based on trends, especially to see the dominance of the problem in general real-time. Meanwhile, for handling problems sometimes a priority order is needed. This condition was also experienced by the Airbnb team in dealing with customers.

airbnb currently handles 80 million customers and continues to grow rapidly following the ongoing expansion. With such customers, they understand that one way to streamline customer service is to be able to understand/process large amounts of customer data (tickets) while detecting problem trends in a timely manner. real-time, even predict it. So what did the Airbnb team develop?

Content grouping, analysis and visualization system

To give effectiveness to the work, a web-based service capable of calculating trends across all customer service tickets was developed. Visualizations are presented to make it easier to read data, including mapping the type of problem, the browser used, the user's country, the subject of the problem, and several other attributes. The system by Airbnb is facilitated by a Note.js-based application with an interface built with React.

The realization of the system requires the role of various components, including the physical infrastructure to become a data store ticket storage to be analyzed. A technology opera source "Elasticsearch" was used in this development. All tickets are recorded on cluster Elasticsearch online real-time.

Incoming data is calculated in certain intervals to find out the existing trend. According to the development team, Elasticsearch makes it easy to scalable and perform Query aggregates in the incoming data set.

The illustration is as follows:

Trends found by running multi-search queries to Elasticsearch to get each ticket attribute defined. A model scoring applied to each time series, sequence of results, and returns the attribute trend at the minimum. After that, several activities are needed including adjusting the periodicity, eliminating noise data, adjust visualizations, and calculate if there is a data spike.

Describing the periodic trend of selected frequency domain use using Fourier transform, instead of using graphics. The reason is to make it easier for the system to find peak frequencies and provide sequential periodic reports of the number of incoming customer service tickets. The final result can represent changes in the maximum value and volume of a particular ticket over time. Data calculation is done in Redis, to provide extra speed when data is visualized in the web UI.

Define and prioritize problems quickly

Next a dashboards prepared for use. From there the spike trend is well defined, officers can see what are the common problems that occur. For example, there is a surge in new users who have search problems on the use of applications on certain platforms. Before this spike becomes a big problem, if the trend is exponentially increasing then the team may decide to act fast follow-up repair.

The system has been in use for at least six months now with Airbnb. One of the benefits of this system, the technical team in particular can capture more things, ranging from bugs which has the potential to be large, making it easier for the team to get priority for solving problems, especially those that require updating a copy of the code in the application.