LOCATION: Rio de Janeiro, Brazil
DATE: 2019/2020
DURATION: 6 months
ROLE: Design Lead & Project Manager
TEAM: Cosmobots co-workers (Cofounder Henrique Carvalho & Frontend Developer Daniel Martins) and GSK Innovation Squad (Marketing Manager & Tech Lead)
TOOLS: Figma, Adobe Illustrator, Adobe Premiere Pro, Postman, Twilio API for WhatsApp
GSK is a global biopharma company with focus on innovation in vaccines and specialty medicines development. In April 2019, GSK Brazil launched "Together - Connecting for Life" - a program created to connect GSK with startups to bring a new look at critical industry challenges. More than 200 national and international startups signed up to participate in the pitch competition and, in October 2019, 4 startups were approved to run pilot projects at GSK - Cosmobots being one of them. 
Our chosen challenge was called "Artificial Intelligence to support decision making in risk management" and the goal was to provide a solution to streamline the risk analysis process, facilitating the identification of problems and increasing the efficiency of the process. Our proposal was a Sales Intelligence AI-powered virtual assistant designed to optimize Sales Team strategies and performance by offering real-time contextual information on prospects.
PROBLEM: Poor quality prospecting data has a negative impact on Sales performance 
GSK Sales Representatives connect with pharmacists, physicians and other medication-prescribing professionals in order to educate them about new developments in the pharmaceutical industry. They schedule and attend sales meetings with health care providers, following leads and cultivating new customers. 
Before visiting a medical professional, a Sales Representative gathers all information available regarding that specific client and which products and treatment plans they could be interested in. The main problem was how to gather this information, since it was distributed across several channels: email, 4 different desktop dashboard apps, a non-user friendly prospect database, a tablet app, a podcast and a message board. 
The main issue was not the lack of information, but the overflow of information and the inefficacy of accessing it from multiple sources. 
SOLUTION: Implement a data integration strategy through an AI-powered virtual assistant 
Our goal was to develop a solution that could integrate all accurate and relevant data from multiple sources into a single spot, available to be conveniently accessed anywhere and anytime, while also pointing out the best approach for each visit. To achieve this goal, we developed "Gessika", an Artificially Intelligent Virtual Assistant designed for WhatsApp that offers real-time data insights and guides Sales Representatives on their visits.

Concept video presented at the Pitch Event 

The Design Approach: 
I followed the Double-Diamond model of the Design Thinking process for this project. The first stage (Discover & Define) is the "problem space", where the problem is identified and then explored until clearly described and understood. The second stage (Design & Deliver) is the "solution space",  where the solution is created, tested and delivered for feedback and further iteration. It is a non-linear iterative cycle that promotes an ongoing workflow where the best ideas are brought into action rapidly, tested and adjusted. 

The Double Diamond Design Process: Discover, Define, Design, Deliver.  

The Discovery:
The starting point was in the description of the challenge proposed by GSK program "Together - Connecting for Life": to incorporate Artificial Intelligence into the overall digital strategy of the Sales department, facilitating the identification of problems and enhancing decision-making processes. I started the process by researching all applications and channels already available to the users (Sales Representatives). 
User Interviews: 
In order to understand which problems were faced by the Sales team on a daily basis, I had to talk to them. I prepared a questionnaire and interviewed 12 Sales Representatives on a short talk with each one of them separately. Since there was a national GSK Sales event, I had the chance to choose participants from different regions of the country and get an overall view of what they all have in common, regardless of being from different demographic groups. The questionnaire below served as an interview structure for reference but there was room to redirect the conversation as well. The User Interview stage helped me to learn more about our users routines, backgrounds, pain points and expectations regarding what they considered to be the ideal technology to aid them to achieve their goals.

User Interview Questionnaire 

Key Findings:
• Users are often on the move, unable to access desktop applications.
• They all use two mobile phone devices: one for personal matters and another for work. 
• WhatsApp is the main app used for work-related private chats and group messages.
• Although they have a tablet with internet connection, just a limited amount of resources and information are available there.
• Users get confused with so many apps and devices.
• Data presentation is clumsy and sometimes outdated.
• Mixed opinions about interacting with a virtual assistant.
These in-depth interviews helped me uncover pain points regarding their frustration of not being able to access the necessary information when they actually need it, their disappointment regarding data format and quality and their doubts about the implementation of a chatbot solution. 
Redefining the Problem: 
The insights from the discovery phase helped me to rethink the problem definition:
"Users need a real-time mobile solution that merge sales intelligence data from several sources into a single solution."  

Next, I defined my key user personas: 

Persona 01

Persona 02

The Development Process: 
Using the MoSCoW method, I designed a prioritization  matrix of the components that will build our Gessika Virtual Assistant App.  
Designing the Solution:
At this point, based on the results and findings of the previous stages, I was able to start designing the solution. The solution architecture​​​​​​​ was designed considering all the business requirements and it was developed with the Cosmobots chatbot consuming (engaging with) a custom-made GSK API. 
This API allows the chatbot to retrieve real-time data while interacting with the user via text. The chatbot requests the data using the fetch method and the API returns a response object in json format, that is then translated into text in a reply to the user. The graph below displays an overview of the conversation flow involving these 3 parts (Chatbot, User, API): 

Conversation Architecture: Chatbot, User & API.

Prototyping, Testing, Repeating: 
One of the positive characteristics of the chosen design approach is that feedback can be gathered when changes can still be implemented easily, since the design and test stages start early. Another positive characteristic is that less resources and manpower are required in a process as lean as this one: designing for the Minimum Viable Product (MVP) with Mid-fi prototyping lead to testing while the work is in progress, so the iterative design process can be repeated as often as necessary. 
For this project, I designed a mockup API for this project using Excel and Google spreadsheets on Sheetlabs (a platform that allows you to turn spreadsheets into APIs) and then tested it through Postman (a platform to build, test and iterate APIs), which saved time and freed the dev team to work on other time-sensitive tasks regarding the API network stability and security. Once my mockup API was up and running, I connected it to the chatbot, coding their interactions with the programming language Node.JS. Since the technical foundation of the project was established, I started testing the solution and using the feedback collected to adjust and improve it accordingly. 
At the same time that i developed the technical elements of the project, I worked on the conversational experience and visual design elements as well. I created illustrations, videos and other multimedia material used to present the solution to stakeholders at different development stages, from the initial pitch to the final presentation in a national conference. 
Designing and Testing the Conversation Experience: 
Building great UI/UX for a Conversational User Interface (CUI) on WhatsApp was a challenge due to some limitations imposed by the platform. WhatsApp is mainly a text-based interface, so I could not rely on one of the most essential elements of interaction design: buttons. 
WhatsApp Messenger is the world's largest messaging platform and it allows enterprise use of its API, so companies are able to build their chatbots and connect them to their WhatsApp Business account. WhatsApp Business API has recently introduced buttons, but they were not available yet when Gessika was designed.  
The first important task was to allow users to choose from a list of options without resorting to buttons. The initial solution for an effective text-based navigation was to use numbered menus: chatbot displays one numbered option per line and users reply by entering the corresponding number of the selected choice instead of typing words or whole sentences. This interaction model provides a more dynamic conversation flow and avoids chatbot fallbacks, which is when the chatbot doesn't understand the user's message. 
It would require considerably more time and effort if users were supposed to interact with a non-numbered menu, specially considering the limitations of finger typing on a virtual keyboard of a mobile phone and how often auto-corrector suggestions turn into misunderstandings. 
Before proceeding any further, I did remote usability A/B testing with 10 participants equally divided in 2 groups of 5 each. One group tested the version A, with numbered menus, while the other group tested the version B, without numbered menus.
Testing our prototype allowed me to validate my design solutions and collect actionable insights for further refinement. The testing outcome was analyzed focused mainly on these aspects: efficiency of use, fluidity and error rates. The version with numbered menus (A) performed significantly better than the version without numbered menus (B) in every aspect considered. 
Regarding efficiency of use and fluidity, "A" presented higher rate of task completion and faster user response times to prompts. Regarding error rates, "A" showed almost no chatbot fallbacks.
Measuring User Experience with UX KPIs
I relied on both quantitative and qualitative UX KPIs (Key Performance Indicator) to measure user experience. User Experience metrics can provide valuable insight on usability issues and help tracking progress toward goals. The following UX KPIs helped me to evaluate chatbot performance and assess user experience:
➜ Task success rate (percentage of completed tasks and conversations).
➜ Time on task (time spent on each task and on the whole conversation).
➜ User error rate (average error occurrence calculated according to the number of  occurred errors in relation to error opportunities per task).
➜ Customer Satisfaction (CSAT survey at the end of the interaction). 
➜ User Feedback Message at the end of the interaction.

Satisfaction Survey results after 1350 interactions (130 unique users)

Product Launch and Final Thoughts   
Once we had our fist "Gessika" MVP up and running, we used the results gathered from our UX KPIs to improve conversation prompts, data format and quality and overall technical architecture. The project was launched successfully on the annual GSK event aimed to introduce their best new products and features. "Gessika" had a huge positive impact on the digital transformation strategy of the company and it is still evolving with new features and updates.
This project was a milestone for me, since it involved different aspects of user experience design, conversational design, project management and digital transformation architecture. I gained valuable experience as a project leader and was able to learn more about how users interact with a chat application. It was also a great challenge to expand my knowledge learning the guiding principles of data architecture and API management systems. Every new project gives me the opportunity to use the skills gained from previous projects as I learn new skills on the process as well, making me able to design better interfaces each time.