IBM Watson aiding in casework

By | 3 minute read | November 18, 2016


In Salt Lake County, Utah, people were struggling with being re-arrested because they wouldn’t show up for court appearances. The county government teamed up with a Code for America team lead by 2016 Fellow Kuan Butts. They created a technological tool leveraging IBM Watson and Twilio to solve this problem. Code for America identified a lack of communication between individuals in the justice system and their caseworkers as the pain point to solve with new technology.

Individuals in the justice system receive a caseworker to aid in the court process. Caseworkers help their clients remember appointments and sift through personal issues that get in the way of solving court issues. Caseworkers can have upwards of 100 clients in the justice system at a single time to support. They would try to reach clients via phone to schedule a one on one meeting. Due to the high number of clients, caseworks only saw clients once a month. It can be difficult to create a real relationship with clients when only meeting once per month. Communication between meetings is critical to cultivating a relationship based off of trust and openness, which is crucial to enabling a case manager help a client be successful. Additionally, other problems arose from phone calls being the primary method of reaching clients. Most clients are under-resourced, so they may run out of minutes on their cell phone. Some people would borrow phones in order to contact caseworkers. Phone tag was a constant problem, making communication difficult in general.


To solve these problems, Code for America developed ClientComm, a web based application that allows caseworkers to send and receive SMS messages at scale to clients’ cell phones. The caseworker messages through the web app, and the client texts back through their phone or a phone they borrow. It enables caseworkers to communicate with multiple clients who are hard to reach. Switching from phone calls to SMS increased communication by becoming asynchronous, where clients can reply whenever instead of playing phone tag. Even though most of the clients are under-resourced, they generally have cell phones with unlimited texting given to them from the government. The app allows users to have secure data, set notifications for clients (appointments, court dates), and store transcripts of their conversations. ClientComm uses the Twilio SMS API to send SMS messages from their web application.

Watson enters in as a method to enrich the SMS messages to further improve communication. The IBM Watson Message Sentiment add-on, offered through the Twilio Marketplace, analyzes SMS messages and returns a score from -1 (most negative) to 1 (most positive) based off the sentiment of the individual message. Code for America built into ClientComm a visualization showing the tonality of each client using the sentiment of their inbound messages. The idea is that when caseworkers are about to message one of their clients, they can pull up this dashboard to see if they are generally negative or generally positive. They may use this visualization to aid the way they frame their message. Or, if a client is usually positive, a very negative message would be a red flag compared to a negative message from a client who is usually negative. Additionally, this capability allows caseworkers to see how a client’s sentiment has changed overtime.


What Code for America found through regression analysis is that for low risk clients, success correlates with having a more positive attitude. But, for a high risk client, positive tonality actually correlates with a higher tendency towards failure. A potential explanation is individuals with a more negative tonality that are higher risk have come to terms with the severity of their situation which might be quite grave. That “realism” might be necessary to that client being able to understand and face what needs to be done to become successful.

Clientcomm has saved case managers on average over 5 hours per week, increased communication from monthly to twice a week, and increased the success of pretrial clients from 50% to 82% over a 5-month period. The increase in communication has already proven success in improving the lives of caseworkers and their clients with not even a year under its belt.

IBM Watson Message Sentiment has further use cases in finding the voice of the customer and flagging negative or positive messages for further review. We’d love to hear what you’re building with Watson.

ClientComm is currently in Pilot stage.

For more information about ClientComm, visit the website here 

For more information about the IBM Watson Add-ons for Twilio, visit the Twilio Marketplace here



Most Popular Articles