AskDoki: Healthcare WhatsApp Chatbot
How might we use technology to make sexuality education accessible to young people?
In Kenya, improving adolescent sexual and reproductive health and rights (SRHR) remains a high priority. Teen pregnancy and motherhood rates in Kenya stand at 18% meaning about one in every five adolescent girls has either had a live birth, or is pregnant with her first child ¹. At the same time, 49% of unmarried girls from 15-19 have unmet needs for family planning, which means that they would use modern contraception if they had access to it, but they don’t ². 13,000 girls drop out of school annually due to early and unintended pregnancy ³ and almost 30% of all new HIV infections are among adolescents and youth ⁴.
While there is empirical evidence that Comprehensive Sexuality Education (CSE) plays a major role in improving SRHR outcomes, there is little data indicating which channels e.g. radio, digital, edu-sport, are the best way to deliver LSE (Life Skills Education) to Kenyan adolescents. So our team at YLabs spent the last several months explored the pressing question — how might we use technology to make sexuality education engaging and accessible for Kenyan adolescents? This is one focus of a larger project led by Amref, Faya, which aims to discover the most cost effective channels for delivering LSE and SRHR content in Kenya.
After multiple rounds of co-designing and testing with users, using a youth-driven design approach, our team designed and developed a WhatsApp chatbot called AskDoki. AskDoki, means “Ask a Doctor” in Kiswahili. AskDoki, is a natural language processing (NLP) ⁵AI-powered counselor for young people seeking advice and services to protect their health. Doki, our doctor chatbot persona, delivers comprehensive sexual and reproductive health education for adolescents based on eight topics ranging from healthy relationships, gender, STI and HIV/AIDS, to future livelihoods in Kiswahili and English. She also connects them with nearby youth-friendly service providers for free, and answers questions that are popular among young people. Our team chose WhatsApp since it is one of the most used social media platforms, and 96% of Internet users 16-64 reported using WhatsApp the previous month¹¹.
What are some of the best practices in designing chatbots for adolescent sexual and reproductive health?
Designing digital health interventions for youth in low- and middle-income contexts requires careful attention to young people’s needs, literacy, economic resources, and digital access. Here are some guidelines that our team incorporated into our chatbot design process in order to create on-demand SRH information and services that are accessible, fun and safe for young people.
1. Align content with desired project outcomes and users’ needs
Before creating content for a project, make sure there is alignment on the key indicators that define the success of the project between all stakeholders. It is crucial to agree on what success looks like for the project before focussing on the key messages included in the chatbot’s educational content.
At the beginning of our design phase, based on the Faya project outcome indicators, our team created a content outline and the corresponding key learning objectives. We then developed a creative brief for our team that included designers, a copywriter, illustrator, videographer, and photographer to ensure that all content and assets were aligned with the project goals. In addition, we learned from rough prototyping testing that youth prefer information provided by an authoritative figure like a doctor rather than a near-peer or adult with unknown credentials. As a result, the educational chatbot content was shared from the perspective of a positive, engaging, and hopeful Doki/doctor persona.
Access to WiFi and mobile data bundles has been consistently ranked as the top barrier from accessing the AskDoki chatbot service. Mobile users in Kenya have access to “free WhatsApp”, meaning users can access WhatsApp text and audio for free without depleting any of their purchased data bundles. This presented a design opportunity. The main drivers behind the creation of AskDoki were the familiarity young people have with using Whatsapp, the free Whatsapp service, and the ability to develop a chatbot on the Whatsapp platform.
2. Privacy is key
Privacy and confidentiality is often the highest priority for young people in SRH services ⁶ ⁷. When designing any digital or physical healthcare intervention for young people, reassure them that you value their privacy, and prioritize building and communicating the privacy protection measures to build trust and confidence in the chatbot. During youth’s first interaction, AskDoki reassures youth that "I am a bot, so this entire conversation stays between us”, and offers the option of reviewing the privacy policy.
Building on Kenyan youth’s perception that WhatsApp is a safe, secure platform, AskDoki seeks to create a sense of privacy for the young audiences. All of our interviewees reported that they felt safe when using AskDoki, and one of our interviewees said, “yes [I feel safe], because I know I am talking to her and she had assured me it was between the two of us.”
When designing a healthcare chatbot, it is important to limit requests for personal information and ask questions sparingly and sensitively. Users will often exit the service when they are being asked too much personal information online. If you have to ask questions, aim to generalize the answers e.g. “below 15”, “15-19”, “above 19”, instead of asking users to disclose their exact age.
Lastly, the backend design should ensure users’ privacy. In our project, all identifiable data is stored securely in a separate database from any sensitive information e.g. demographic questions and analytics. This minimizes the risks of compromising our users’ privacy. All of our backend frameworks are GDPR compliant. For example, Amazon Web Services (AWS) is used to store conversation logs securely, and Google Cloud Firestore (GCF) is used to store user data with separate collections for user identifiable data which can only be accessed by the admin.
3. Create youth-friendly and locally relevant content
CSE curricula are often created using technical language, heavy on facts and biology. However, most adolescents don’t have the patience for dry, boring material ¹⁰. One key learning from our prototyping phase is that youth prefer to-the-point content that doesn’t beat around the bush, and feels like a day-to-day conversation with peers. This might be difficult for experienced copywriters to adapt, but youth-friendly content isn’t always grammatically correct! Consider incorporating slang, visuals, and emojis to help youth digest the content. For AskDoki, our local designer Nzisa Kioko tested and iterated our content based on feedback from local youth to ensure we are using language that resonates with them.
We consistently hear from young people that to keep them engaged, they need a combination of audio and visual content, in addition to written information that answers their highest priority questions. However, access to WiFi and mobile data bundles has been consistently ranked as the top barrier from accessing the AskDoki chatbot service, which brings us to the next guideline below.
4. Ensure that youth have data-frugal options
During the rough prototyping phase, we learned that mobile users in Kenya have access to “free WhatsApp”, which means users can access WhatsApp text and audio for free without depleting any of their purchased data bundles. However, downloading files like highly desired illustrations and videos is not part of the free WhatsApp offer, and we anticipate that many young people would not have mobile data bundles to view the photos, illustrations, and videos. Therefore, as we were creating the content, we ensured the most critical information was conveyed in the free WhatsApp text messages, whilst also including more visually engaging materials.
5. Incorporate touch points to motivate, prompt reflection, and delays to support user engagement
We all get bored, or frustrated, or overwhelmed by the amount of information we are receiving. To prevent young people from dropping off, keep checking in with your users to make sure they are following the content and are engaged. We make sure the AskDoki chatbot is interacting with young people at a comfortable pace, and AskDoki checks in with our users every 2-3 messages to confirm comprehension. AskDoki either asks them e.g. “does this make sense?”, or introduces a mini reflection prompt to engage youth in the learning process.
In addition, we incorporate some micro delays (0.1-0.5 second) between the messages to help youth feel more comfortable with the pace of interaction. It’s well-known in web design that if an animation is too fast, it’s hard for our eyes to see, and it can cause users to feel dizzy. Young people gave us feedback that in our early prototypes, the chatbot was talking too fast, since AskDoki was replying without any delay. This immediate response felt very different from interacting with a human! Since then, we intentionally incorporated a 0.25 second delay between each message to ensure AskDoki interacts with youth at a more comfortable and life-like pace.
6. Decide what kind of chatbot your project needs, and don’t assume you actually need one
By now, you might be asking whether or not a natural language processing (NLP) chatbot is suitable for your project ⁸. If you expect your users are able to complete their task with several steps, the more traditional rule-based bots is likely a better option. Rule-based bots communicate through pre-set rules, and user input must conform to your pre-defined rules in order to get an answer. An open conversation with this type of bot is often not possible.
Our team decided to build AskDoki as a NLP chatbot despite the additional design, development, and training effort because a NLP chatbot can learn from real-life experience and adapt. We want AskDoki to be shaped by the questions and responses by our young people beyond our co-design process, so that the chatbot becomes truly relevant and helpful to them.
Currently, our most pressing challenge is linkage to child protection services. Despite having incorporated sample training data so that the chatbot is designed to provide appropriate resources for users in distress, AskDoki is far from perfect. It is important for a reviewer to monitor the content and ensure cases of potential harm and abuse are reported. Furthermore, it is important to consider reach — a chatbot can only reach adolescents who have access to a smartphone, and this may exclude the young people who are the most vulnerable e.g. girls in rural areas.
7. Set up a post-launch chatbot monitoring and iteration plan to make sure users needs are being met
If you decide that a NLP chatbot is indeed the best option for your project, you will need to hire an experienced developer to build the classification system, find and/or create sample data to train the bot to understand intents, defined as the goals the chatbot users have in mind when typing in a question or a comment, and generate human-friendly, medically-accurate responses. Consider hiring representatives of your target audiences to team up with your technical copywriter to generate those sample data (and later, ask them to help test the data), to ensure that your chatbot’s comprehension meets user needs! After your chatbot is ready for launch, you will need to allocate dedicated staff to review the interactions between your users and the chatbot regularly, and train the chatbot on additional real data.
Our AskDoki reviewers monitor the anonymous conversation logs frequently, and they flag the cases when AskDoki is unable to correctly respond to users' questions and needs. If the content already exists, but AskDoki misinterprets the users’ intent, our development team corrects it. As we track new requests or unanswered questions from users in the current version of AskDoki, these become a great source of user-driven content! In summary, careful monitoring and iterating are critical in improving your chatbot’s comprehensive capability.
WhatsApp is a well-known platform increasing in popularity globally, whereas WhatsApp chatbots are a new technological solution that few NGOs and healthcare organizations have leveraged, so our team is very excited to share the finding from the co-design of this new intervention with youth. These are only several of the guidelines that our design team has developed during this healthcare chatbot project. To find out more about our project work or to share your own experiences - get in touch @YLabsGlobal or sign up for our email newsletter to stay in the loop
References
Fact Sheet Teenage Pregnancy - Kenya https://www.rutgers.international/sites/rutgersorg/files/PDF/RHRN-HLPF_A4leaflet_Kenya.pdf
Kenya MOH. National Guidelines for provision of adolescent and youth friendly Services in Kenya, second edition. Nairobi: MOH; 2016.
Muganda-Onyando R, Omondi M. Down the drain: counting the costs of teenage pregnancy and school drop out in Kenya: Centre for the Study of adolescence; 2008.
National AIDS Control Council, Kenya’s Fast-track Plan To End HIV and AIDS Among Adolescents and Young People
What is NLP? https://landbot.io/blog/natural-language-processing-chatbot
Natural language processing is the linguistically oriented discipline in computer science that is concerned with the capacity of software to understand natural human language - written as well as spoken.World Health Organization. WHO Consolidated Guideline on Self-Care Interventions for Health [Internet]. https://www.who.int/reproductivehealth/publications/self-care-interventions/en/
Digital Self-Care, A Framework for Design, Implementation & Evaluation. Self-Care Trailblazer Group. https://www.psi.org/project/self-care/digital-self-care/
Do you need a NLP chatbot? https://landbot.io/blog/natural-language-processing-chatbot#section-5
Executing UX Animations: Duration and Motion Characteristics https://www.nngroup.com/articles/animation-duration/
Youth-centred digital health interventions: a framework for planning, developing and implementing solutions with and for young people. World Health Organization. https://apps.who.int/iris/handle/10665/336223. Accessed July 21, 2021.
Digital 2021 Global Overview Report by We are social and Hootsuite
https://wearesocial.com/digital-2021. Accesses Sep 20, 2021.
