Social Media

Project Firefly gathers public feedback data for our clients and their competitors from data sources such as review sites and social networking sites. One of the more difficult challenges involved clustering property-specific webpages for the same physical location. Using a variety of features such as property name, addresses, geolocations, and phone numbers, our custom matching algorithm allows for example a local hotel manager to compare his reviews and scores with every other hotel in a 5 mile radius. Combined with our text analytics feature, we can provide businesses with deep insight into competitor strengths and weaknesses.

Survey Engine

At Medallia, our survey engine handles transactions on millions of surveys at any time. After extensive evaluation of key-value stores, our team chose Project Voldemort, an open source NoSQL project from LinkedIn. We compared Voldemort with Cassandra and decided in favor of Voldemort due to its simplicity in API, predictability in access latency and maturity in its underlying Berkeley DB storage system. Our team has contributed back to the project by adding secondary index and range lookup capabilities.

Topic Classification

Many types of feedback contains a significant portion of plain text input, and our clients want to be able to view feedback on a particular area of their service or business. Topic classification enables us to group that feedback so businesses can understand exactly which areas require improvement. Clients often want to define their hierarchy of topics, and one of the interesting problems we face involves using clustering techniques to automatically suggest topics given feedback data.

Sentiment Analysis

Sentiment analysis combined with topic classification allows our clients to truly dig into raw text feedback data. As opposed to general sentiment solutions, our analysis of the data showed that a truly accurate sentiment analysis solution must be context-dependent on both the industry and the source of the feedback, e.g. whether it comes from Twitter or a leading question.

Actionable Items

Often customers will provide a specific piece of feedback that an employee can immediately act upon. These actionable feedback, such as 'The room was dirty' are perhaps more relevant than vague mentions of displeasure 'I dislike this hotel'. Employees can employ this type of relevancy filter to more quickly and easily respond to customer feedback via explicit actions. This project was developed from start to finish by one of our summer interns.

QA Infrastructure

As a SaaS company, many of our clients often have special requests for features. To enable rapid development without pushing unsafe code to production, we rely on our QA architecture which allows us to deploy realistic, complex data sets, up to several hundred-gigabytes of data, in mere minutes. This allows us to easily create demo environments for rapid prototyping and is extremely useful for exploring features in a safe environment, debugging production issues, and as a base for our automated testing tools.

Automated Testing

At Medallia, automated testing to prevent regressions and verify expected behavior is the responsibility of each individual developer. To maximize the coverage of our tests, we use a variety of testing frameworks including JUnit4 tests, Selenium based frameworks including our cross-browser Project Curious George and our functional testing framework, Jasmine tests, and more. Our framework runs over three thousand unit tests on every commit, and the full array of tests before each of our twice a week deployments.