AI Message Routing UX/UI

Rhinogram

Project Background

Rhinogram is a HIPAA-compliant patient engagement platform designed to streamline communication between patients and healthcare providers. The platform facilitates secure virtual visits without requiring patients to download an app, offering a seamless and intuitive user-experience. Serving over 13.5 million patients nationwide, Rhinogram provides a secure care solution tailored to modern healthcare needs.

Healthcare practices often face challenges such as managing inbound calls, scheduling appointments, handling cancellations, and ensuring efficient patient care. Rhinogram effectively addresses these issues by enabling more convenient, precise, and efficient communication. Practices leveraging Rhinogram’s platform experience, on average, a 30% increase in the number of patients treated, driving both operational efficiency and practice growth.

Problem

Phone and text receptionists at large hospitals are inundated with a high volume of inbound calls and patient initiated text messages. They spend time reading inbounds texts, interpreting them and routing them to the appropriate team member, providers or group. Often additional staff is required to handle the inbound texts during peak hours. The question we were tasked with answering was: Is there a way to make this process more efficient and eliminate the need to increase staff and give receptionists back time to handle other administrative tasks?

My Role

As the UX product lead I wanted to understanding the problem and determine the best solution. I needed to analyze available data to understand, for a large hospital how many inbound texts a receptionist currently reads, routes and answers. I wanted to determine the average amount of time it takes for a receptionist to review an inbound text message and route it to the proper destination. This would provide me with a baseline and a metric to measure success. I also wanted to talk with receptionists to understand any other frustration or needs associated with the routing process.

Limitations

To establish baseline metrics, I conducted interviews and gathered data from a focused group consisting of three large hospitals and practices. While the sample size was limited, it provided valuable insights into key trends and operational benchmarks.

UNDERSTANDING THE PROBLEM

  • A single receptionist may handle between 50 to 200 texts per day, depending on their shift length and the hospital’s size.
  • By analyzing the data of 3 of our large hospitals we were able to determine that a receptionist during peak hours handles on average 54 text messages.
  • Further analysis determined that 34 seconds is the average amount of time needed to read and accurately route or personally answer and inbound text message.
  • Average response time to an inbound text message during those peak times was 27 seconds

By the numbers

200

messages per day

54

messages/hr

34

seconds/message

Question Ladder used for automated routing

early sketches for automated routing

BRAINSTORMING & WHITEBOARDING

Question Ladder

  • How to streamline the process?
    • Can we automate?
    • What could we automate?
    • How would automation reduce the amount of inbound messages?
    • What is amount of effort needed to automate?
  • How would we classify a Patient Initiated Message?
    • New conversation
    • No current open conversation with patient
    • Time between messages?
  • AI – Natural Language Processing
    • Engineer Spike

CATEGORIZE & SORT

To leverage Google’s Natural Language Processing (NLP) for interpreting inbound messages, it was necessary to establish a set of defined categories to classify each message accurately. By establishing a workflow (Fig 1) and analyzing and reviewing thousands of patient-initiated messages from past interactions, we identified 13 distinct categories that comprehensively encompass the variety of message types received.

This data-driven approach allowed us to assess category usage, from the most frequently occurring to the least utilized, providing valuable insights for optimizing message handling and response strategies.

Fig 1: automated routing workflow

Front End Design for Category Setting

1
2
3
4
5
1

Multi-select inbound message category

2

Single Select for the type of routing.

3
  • Feature Development: Introduced the Suggested Routing feature to allow users to test the routing process in a controlled environment.
  • User Familiarity: Enabled users to become more comfortable with the new functionality by providing a hands-on testing experience.
  • Machine Learning Refinement: Collected data during user testing to train and enhance the machine learning model.
  • Accuracy Improvement: Improved the precision of category routing by leveraging real-world usage insights.
  • Enhanced System Reliability: Combined user feedback and model optimization to ensure a reliable and effective routing process.
4

The user selects the group or specific member where the inbound message category will be routed.

5

The user has the option to create an auto response to a message that is being routed. ie “Thank you for your message, we have forwarded your question to our billing department and someone will be responding to you with in the next 10 minutes.”

This message can include variables to make it more personal as well as attachments if any forms would be needs to speed up the workflow.

Suggested Routing

We developed the Suggested Routing feature to enable users to test the routing process in a controlled environment. This approach not only helped users become familiar with the new functionality but also allowed us to refine the machine learning model by gathering data to improve the accuracy of category routing. This dual-purpose strategy ensured both user confidence and enhanced system reliability.

routing analytics dashboard

Automated Routing Simple Analytics

Once organizations gained confidence in the Suggested Routing process, transitioning to automated routing required only a simple adjustment in the administrative settings. Organizations were provided with an intuitive dashboard to monitor the number of messages automatically routed each month, along with an estimated time savings calculation.

Based on our research, the average time required to manually route a message was determined to be 34 seconds, which served as the benchmark for calculating time savings. This streamlined approach not only enhanced operational efficiency but also offered actionable insights into the impact of automated routing.

PROMOTION & ADOPTION

To promote this new feature we created an in-app, new feature guide using Pendo (Fig 2) and segmented the guide to display to organizations that had the highest number of patient initiated messages and messages sent. These are the organizations that benefit the most from automated routing. We asked these organizations to be part of a beta group. 23 organizations signed up for our beta group.

Fig 2: Pendo in-app guide to promote beta group

UNEXPECTED BENEFITS

One of the the constant challenges we’ve had with our application is getting organizations to close conversations. Many larger organizations can have thousands of open conversations which can impact load times and clog up groups.

In order for Automated Routing to work optimally, conversations must be closed so a message from a patient will be categorized as a Patient Initiated Message. One of the criteria for a PIM is no current open conversations for that patient.

We saw a 76% reduction in open conversations that were errantly left open for the organizations that are using automated routing.

RESULTS

Following the conclusion of our three-month beta program, we refined two categories and identified potential phase 2 enhancements for Route Tracking. Among the 23 participating organizations, 19 adopted the feature, achieving an impressive 82% adoption rate. Post-beta, an additional 10 organizations implemented automated routing, further expanding its utilization.

For large organizations handling over 160 patient-initiated messages daily, automated routing has demonstrated significant efficiency gains, saving approximately 90 minutes per day. Additionally, organizations utilizing automated responses in conjunction with auto-routing have experienced a 47% reduction in overall response times, enhancing both operational efficiency and patient satisfaction.

“Automated routing is a gamechanger. We were able to cut the time in half that is spent getting messages to the right person. This saves us roughly an hour a day, valuable time we are able to spend with patients”