Piloting a Chatbot for Sexual & Reproductive Health

I have been fortunate enough to have spent the majority of my adult life exploring many different design disciplines. Having studied interior design and pivoting towards digital design I now find myself humbled as a Service Designer at Praekelt.org where we use technology for social impact.

Always eager to embrace the unknown, I began working closely with our partner, GirlEffect.org, to research, design and prototype a chatbot capable of providing information for vulnerable adolescent girls on sexual and reproductive health. The design questions I faced were very different to the questions in the marketing and media sector that I had been familiar with: how might we create a safe digital space for girls to engage? What will build their trust? What are their challenges in life? How will we truly drive impact within a conversational environment? Who are we leaving behind? I had to approach this through the lens of our users.

My initial desktop research around chatbots (and specifically, why they fail) led me down endless rabbit holes, an integral part of the Discovery phase. But there was light at the end of the tunnel in the form of a voice — to be more specific, a voice about Voice. Oren Jacobs, Cofounder and CEO at PullString, Inc. said at Google’s I/O ’17, “If you don’t spend the time crafting that character and motivation carefully, you run the risk of people projecting motivations, personality traits, and other qualities onto your App and brand that you may not want associated with them.” Paramount to our interface’s success was developing characteristics and qualities that made the persona distinctive and trustworthy .

Together with the project’s partners, we embarked on a journey of user research. South Africa and Philippines were selected as regions as their societies were markedly different in culture and values. Apart from providing a research design framework for co-creation workshops, a prototype would also be required to assess interest, uptake and potential for impact. Our workshops with a cohort of potential end-users and gatekeepers revealed valuable insights around attributes associated with trusted figures which the team were able to refine in order to give our prototype the most relevant personality.

For instance, we saw that our users sought information from familiar members in their personal networks rather than an expert, life experiences trumping specialist knowledge. This moved us towards a persona that embodied an older sister whose relatability could be trusted.

RapidPro, an open source messaging tool, was chosen as the development platform to build our prototype. Despite having no prior experience with RapidPro, I was able to quickly master it thanks to many available resources and a community eager to share learnings. The fast-paced sprint environment used to develop our prototype made RapidPro an effective platform choice. UX decisions and content management could be undertaken in real-time with minimal impact to existing users’ experiences.

From the onset, I had a vision to approach the chatbot as a service that would be channel agnostic. Preference for IP Messaging applications vary globally ,and we had to keep in mind that the chatbot’s journey starting in WhatsApp then moving into Facebook Messenger had (and still does have) an undefined future. With its multi-channel deployment capabilities, RapidPro allowed us to iterate the prototype throughout the piloting period, taking learnings and adapting flows, logic and custom metrics where needed.

Working with existing large service providers provided both an opportunity and a challenge. While they brought large audiences and immediate reach that enabled us to scale, Facebook also has a lengthy and arduous technical approval phase. Facebook is also (rightfully) cautious about new features, and denied a request for a feature to test a desired behaviour that required push notifications. Likewise, we had to work around existing advertising policy restrictions to recruit users.

While we still have a long way to go, we have made countless learnings and continue to iterate with plans to test demand and uptake in new markets.