Mental wellness - A real world problem

A large fictitious insurance company in the US noticed that their customer ratings are going down gradually and CLV diminishing. Upon some research they found out that their help desk received lot of complaints from their employees calling in sick very frequently. These unplanned days off from the employees impacted their daily routine causing delays to customer commitments leading to poor customer experience.

Upon further research, the company found out that stress, anxiety, depression, and insomnia were main reasons for their employees calling sick and the pandemic made it even worse. They hired IQZ Systems ("IQZ") to build a solution that allows the employees to express their thoughts, feelings, current conditions and guide them with right recommendations to overcome these issues. Company named the solution "Heal & Hale". Company also requested their marketing team to send out campaigns to their employees about the initiative and how they are planning to help their employees.

How Heal & Hale helps

Connecting employees and volunteers

Idea is to build an easy-to-use platform for the affected employees and for the volunteers who provide consulting and counseling to the affected employees. IQZ built the following components in 4 weeks.

  • A Pega campaign within the organization to promote Heal & Hale initiative
  • A React based website integrated with Pega Web Chatbot for employees to express their feelings. Enabled Sentiment and Topic Detection on Web Chatbot.
  • A Cosmos React Web embed website for volunteer registration
  • Pega Cosmos React user portal for volunteers to address, track and monitor employee requests.
  • Pega process AI for registration case management
  • Pega Decisioning to provide Next Best Recommendations


Address employee concerns

With the mind set of talk-to-me-first, initial step is for a Pega chat bot to interact with the users who seek guidance and help. This allows the users to be anonymous and feel comfortable sharing their current condition. The bot asks a maximum of 5 questions along with user demographic information to determine the depth of the condition. These two sets of information are used as predictors and fed into Pega strategy to obtain next-best-recommendations (NBR) of therapy or treatment or visit-a-subject-matter-expert. If the user choose to visit a SME, then the user is asked to provide personal information such as name, email ID to connect them with the SMEs. Upon providing the information, the users are given a Pega appointment ID as confirmation to establish contact with the SMEs.

The next-best-recommendations list is an output from Pega Adaptive Decisioning Model (ADM). This ADM model uses the above predictors to provide the recommendation.

This ADM model receives feedback on every user touch point. This constant feedback allows the model to self-train and improve prediction accuracy to provide the next best recommendation.

Inbound channel flow

Outbound channel flow

Appointment flow

Platform for Volunteer SMEs

A Pega Cosmos Web embed Smart form allows doctors, subject matter experts, peer group to register themselves as volunteers to provide consulting and counselling to the affected employees. As part of registration process, the volunteers provide their real-world work experience, mastery or specialty skills which is fed into the ADM model as one of the predictors to choose the right skilled SME for each counselling request. Prior to accepting them as volunteers, the system performs a fraud check using Pega Process AI to eliminate fake volunteers. Upon confirmation they land in Pega user portal where they can see their worklist and the employee list seeking counseling.

Registration flow

Volunteer feedback mechanism - Training the ADM model

Though counseling and consulting happens outside the system, the volunteers update the Pega case with the suggestions and recommendations provided to the employees. These updates are fed back into the ADM model for further prediction accuracy to provide more accurate next best recommendations in future.

How IQZ built it

Using Pega's Express Methodology, during Discover phase, we determined the needed Personas, Channels, Data Elements, Interfaces and External Touch points. Once the initial End-to-end application design was put in, the self-sufficient Scrum team worked parallelly on multiple components.


Pega BOT - Inbound Channel

To help the affected employees, the recommendation was to keep their communication as discrete and as anonymous as possible. We used Pega Digital Messaging Services and Pega Web Chat Bot to interact with the employees to gather information about their condition. We applied Pega Natural Language Processing (NLP), Sentiment and Topic detection capabilities to determine the bot flow. Pega Web bot is integrated in react based external website.

Blue bird Bot flow

Pega Predictions - Next Best Recommendations (NBR)

Pega decisioning is a very integral part of the solution. Multiple ADM models are used at various interaction touch points to provide accurate Next Best Recommendations of therapy or counselling services. Constant feedback is fed back into these models to improve prediction accuracy. In BOT interactions, custom strategy is used to suggest the best SME to schedule an appointment for consultation. Issue type, Issue severity score, Location, Expertise is used as the predictors to find best SME.

ADM Model

ADM Model

ADM Learning - Response feedback

Continuous response feedback is fed back to the ADM for better performance. We collected feedback, for the BOT suggested recommendations, also an outbound campaign is sent to the user to collect the rating of the suggestion. IH

Monte Carlo Data Sets - Initial training of ADM & NLP Models

Instead of control group training for ADM (Which may require few months), we did initial training using Monte Carlo data set and custom strategy. We compiled well known scenarios into ADM training business strategy and fed/ trained the ADM with minimum 20K records.

ADM Training

ADM Training

Pega Process AI - PMML model generation

Company wanted to ensure that the volunteers performing the service are thoroughly vetted before they are added to the system. We used Pega Process AI capability to perform fraud checks using experience, number of cases handled, expertise, age as indicators. Process AI prediction is based on the PMML model. Since the company does not have historical data we used Monte-Carlo data set to generate the historical data. Data set generates different sets of combination and their probability of fraud value is calculated based on the logic provided by Business. Generated records are used to create PMML model.

PMML Model

Pega Cosmos react Web embed - Registration case management

Registration form for SME is included as Cosmos React Web-embed in an external react based IQZ website. Registration cases are handled in Pega case management with Pega Process AI. Upon the lower threshold of the propensity, the cases will be rejected automatically by the system. Genuine cases will be sent for approval. On approval users will be notified with the login credentials for Pega Cosmos react portal.

Web embed

Social channel and real time event for ADM feedback

Twitter integration is done via Pega REST integration connector, to monitor the social signals of affected users. The twitter post from the user is used to get their sentiment. On negative sentiment we call real time event, which will get the Next Best Recommendations for the user and send via outbound channel. Social Channel integration

Pega Cosmos Web portal - SMEs feedback

Application provides Cosmos React portal for SMEs to handle the consultation with the users. After consultation SME provides recommendation comments on the case which is again used for ADM learning. Other suggested actions are captured in a Data table. In future the suggestions are analyzed and added as new recommendations in the CDH.

Cosmos portal

Collect User Feedback

Users are requested to provide their Feedback on the system. This feedback is collected and stored in IH for future use. Two areas where Feedback are requested:

  • Feedback is collected once the CDH Offers are presented and selected by the user. Feedback
  • Feedback is collected from User for their Appointment with Experts/Volunteer so it can be stored and used for Rating Calculations Feedback

What we learned

  • For SME Rating Calculation, external assignment for user to provide their feedback was not working, due to some security challenges with Pega Cosmos React. We implemented via an Anonymous API.
  • Unable to use CDHResponse DataFlow OOTB, hence initiated and assigned Real Time Data Flow Execution.

Future Scope

  • Public User Registration Process for the User to track their activities and how the mental wellbeing changed over time by using Heal and Hale application.
  • Reports and charts with more analysis for measurable results and meaningful insights from the users.
  • Collect SME Comments (Other Recommendation) over time and create new CDH Actions through batch processing.
  • Capabilities like tracking user’s food routine, lifestyle routine, voice interactions with Bot and organizing social activities

MANDATORY Application Config

Before validation of the application. Please follow the steps mentioned in HnHConfig-MANDATORY.pdf file (part of the file) as we implemented in Pega Enablement instances and in case of non usage the instances will be suspended and a new instance will be initiated on login, so URL needs to be updated in the mentioned DSS rule.

Built With

  • adm
  • pega
  • pega-cdh
  • pega-chatbot
  • pega-cosmos-react
  • pega-decisioning
  • pega-text-analyzer
  • pmml
  • process-ai
  • react
+ 56 more
Share this project: