Machine learning (ML) and the rise of big data have spurred advancements in predicting outcomes and suggesting solutions faster.
For example, in healthcare, data on patients’ symptoms can serve to predict future diseases. In banking, transaction anomalies for millions of users can be detected faster, leading to fewer instances of fraud. And even in the world of entertainment, what’s watched on Netflix can serve as the basis for what’s watch next (the “percentage match” for a TV show or movie).
These relatively clear-cut applications have made ML a household name outside the world of tech in the past decade. But, as will always be the case with ML, its applications are constantly growing.
For Seattle-based Qualtrics, the “experience” of customers, employees, products and brands — processed through its experience management (XM) software — plays into how it applies ML. Its Predict iQ product, for example, predicts customer churn based on signals from survey data, and other applications of ML at the company can ultimately answer questions like, “Will this customer cancel their account?” or “Will this employee quit?”
Below, Williams shared how Qualtrics applies ML, the impact it’s had on their clients and how “connecting the dots” through ML is one of the most excitable parts of the job.
What’s an interesting way Qualtrics is using machine learning?
One of the most interesting ways we’re using machine learning is our next version of Predict iQ technology. The first version of Predict iQ was crafted around helping companies using our customer experience product to predict customer churn based on signals from survey data. But, the next iteration of the feature is broader and more powerful. It pulls in data not just from surveys, but from what we call our XM Directory, which customers use as a system of record for all experience data (surveys, unsolicited feedback) as well as operational data (CRM, marketing data, ticketing data, etc.) across their customer or employee bases.
We have also generalized it to work for any binary classification outcome in any of our product lines (Will this customer cancel their account? Will this employee quit?) using the latest predictive ML techniques. And we built it so that it integrates cleanly with xflow, our actions platform, which means that the model can run inferences on new data points as they come in and trigger relevant actions: open a ticket, send an email, etc. What’s really cool about this new Predict iQ is that most ML-related investments we are making help accrue to its accuracy.
What impact has machine learning had on your customers?
The most visible, widely-used machine learning application on the Qualtrics platform is in our Text iQ suite of features, which specifically uses both supervised and unsupervised ML techniques for sentiment classification and topic recommendations, respectively. Both of those features are highly impactful for our clients, who may have anywhere from hundreds to literally tens of millions of open-ended text responses per survey or listening channel that they need to sift through and make sense of.
Our sentiment classification in particular has two modes: overall response-level and topic-level (aspect-level). Either or both can be aggregated and visualized in dashboards or reports, connected back to other structured data and correlated or otherwise statistically analyzed in Stats IQ (our user-friendly statistical analysis toolkit), and/or actioned upon on an individual-response basis via xflow. Thousands of our customers use this functionality on at least a monthly basis, and I know from talking to many of our biggest enterprise CX customers personally that our topic and sentiment analysis sit at the heart of their customer experience programs.
ML applications shine the brightest when they’re given clear, unambiguous objectives.”
What excites you most about the work you’re doing?
I love how we’re connecting the dots. Our clients are trying to make sense of all the data coming in from myriad sources and touchpoints, looking to understand what it all means and what actions to take to advance their specific goals, and the iQ organization that I lead is right at the heart of that.
We enrich customers’ unstructured text data with signals that we extract using NLP and ML (topics, sentiment, and others coming soon). We provide visualizations and deep tools for statistical analysis of their structured data, uncovering correlations, clusters, and other relationships embedded there. We enable them to use their data to predict business outcomes that are important to them, as described above. And we’re increasingly providing insight into key metrics over time, surfacing potentially time-sensitive trends and events in time-series data, such as anomalies or outliers that reflect something important for a client to take action on. It’s fertile ground for machine learning! ML applications shine the brightest when they’re able to draw from lots and lots of data and are given clear, unambiguous objectives. We have both.
