Partnering with a Financial Compliance Giant
Headquartered in Chicago, Donnelley Financial Solutions (DFIN) is a leader in risk and compliance solutions, providing insightful technology, industry expertise and data insights to clients across the globe.
Recognizing the Problem
As a global financial compliance company, DFIN receives hundreds of thousands of emails from clients all across the globe every day. With such a large volume of communications, it was easy for account managers to begin to notice various trends: some clients, it seemed, were significantly “higher-touch” than others, and required more effort to manage them; specific times of day and certain times of year yielded higher email traffic; certain topics seemed to result in more frustrated clients.
And yet, while these trends felt real, DFIN had no data to prove that they were anything more than just gut feelings. Since DFIN prides themselves on providing insightful technology to their clients, they began to wonder, why not invest in an innovative tool for themselves?
The Machine Learning Solution
Earlier last year, DFIN internally developed a proof of concept (PoC) that was intended to tag emails based on a variety of criteria so that the management teams could look for trends and sentiment across their client-bases. The goal of the PoC was to illuminate which clients needed a more white-glove service so that DFIN could staff for them appropriately, but they also hoped to identify areas where their clients were commonly growing frustrated so that DFIN could alleviate tensions before they occurred. The PoC, while simple, was a success. It allowed DFIN to see which clients sent the most emails. It also tagged those emails with what they were about.
However, it didn’t do much in the way of analyzing sentiment. Additionally, the tool had no user-friendly interface for account managers to access should they need to adjust the terms that the solution was looking for. When the original developer left the organization, DFIN turned to CURTIS Digital to bring the application up to their rigid, production-level standards and complete the job.
The solution is an elegant one: an email adapter sits in front of the DFIN mailbox and strips the emails down to text-only files, scrubbing out any sensitive data in the process. The emails are then passed through a variety of pipelines to analyze different metrics: sentiment (was the client happy? upset? neutral?), Line of Business, Context (what types of work do these emails relate to?), Company, Urgency, and Action. After the emails are processed, the data is sent to Power BI so the DFIN team can review.
The most impressive part of the solution is the Power BI reporting. DFIN can now see which companies are high-touch vs. low-touch, how many emails are sent during a variety of intervals, what their sentiments are, etc. One interesting discovery: negative emails tend to be sent right around the end of day, while happier emails arrive between 2-4 PM. The system process over 3,000 emails and hour.