In this post we highlight factors that might influence how conversational interfaces (CIs) are viewed, used, and interacted with across emerging markets in the future.
Risks in using messaging APIs:
Stemming from the recent GDPR legislation on data sovereignty, privacy, and security, companies like Facebook and Telegram quickly had to adjust their messaging applications to, ultimately, restrict the amount of information CIs can extract from users. Given these new rules, and potential threats to these companies’ core business, messaging application companies may choose to restrict and close access to their messaging API, making it considerably more difficult for CI providers to build on these platforms.
To hedge against this risk, CI providers should plan to build across different channels (SMS, different messaging applications, web). They should also limit dependency on these applications for critical business functions and instead focus on using CIs as acquisition channels rather than for business functions. So, hypothetically, Messenger can be used to acquire customers through Facebook ads; once acquired and customers have bought into the tool, users could potentially be migrated from Messenger to an independent application.
Considerations around disclosure:
Throughout our research, we heard questions and concerns around disclosure : Is the conversational interface obligated to let the customer know they are talking with a bot instead of a human?
Whether or not CI’s are ethically obligated to do so is unclear, but currently, none of the conversational interfaces we’ve investigated volunteer that information. While some insinuated that the interaction is automated through button pushes and stock responses, many prefer to not explicitly disclose that a bot is in play.
Regardless, some users—particularly those in rural and peri-urban settings with limited digital experience—are willing to try the interface if it helps answer their questions. As they become more familiar and comfortable with the interface, it is likely that s/he will trust whatever information the interface provides. If the content is accurate and well sourced, the user will may well benefit; however, if the content is inaccurate or fake, the user may suffer negative consequences.
There is no immediate solution, but in time, new digital users will become savvy enough to understand that they are talking with a machine and, hopefully, savvy enough to discriminate between flagrantly false information and accurate information. Meanwhile, it is unclear how to address this cost-effectively. On one hand, if the conversational interface openly informs the user that they are talking with a machine versus a human, it could discourage use of the interface. On the other hand, by not letting the user know they are talking with a machine, it may be unethical to mislead the user into thinking they are talking to a person when they are in fact talking to a machine.
Emergence of voice as alternate interaction tool for conversational interfaces:
Text-based tools to interact with customers will continue to grow, evolve, and reach new populations globally, while the illiterate will continue to face barriers. This demographic, who likely cannot fully understand messages received and/or struggles to type out their own messages, will presumably be locked out of these new services and further marginalized.
However, some companies, such as Hishab and Maya Apa in Bangladesh,, are developing voice solutions to address literacy challenges. Unlike interactive voice response (IVR), these solutions interact with users through voice rather than number prompts, creating a more authentic and lifelike experience. As this technology matures, it should help address literacy considerations and make for a more meaningful user experience that can be easily accessed over airtime rather than through mobile data.
However, it is our understanding that building an emerging market language voice solution like the one Maya Apa built currently requires an extensive amount of time, technological capability, and data storage. According to these two providers, building the language database in Bangla took several years and required painstaking recordings of different pronunciations of words and scripts to produce a minimum viable language database capable of responding to basic conversation prompts. Building a similar oral language database from scratch in sub-Saharan Africa—given the dearth of local language content —would probably take just as long .
Nevertheless, it is encouraging to see that companies are making the effort to solve for this often slighted demographic. Applied to financial services, this technology could make real headway in addressing issues relating to education and capacity building that would otherwise be ignored due to literacy challenges.
Conversation interfaces are starting to emerge as a viable business resource for financial service providers across the emerging market. Their ability to cost- effectively interact with users and extend an array of different services, from the transactional to the educational, may help businesses connect with new customers while providing consistent and reliable support to existing ones. And, as the technology barriers to building a conversational interface continue to fall, we expect that more and more businesses will look to these tools as critical revenue generating and cost reduction strategies.
Nevertheless, these tools will only succeed if they consistently solve for customer needs. To best calibrate expectations with what customers challenges they can address, businesses need to not only understand the technology that powers CIs but also users’ digital behavior . Accordingly, businesses need to source content that is accurate, present it in a way that is easily understood by the user, and construct the application so that it is intuitive and reliable. If they are able to accomplish this, the CI will likely earn the user’s trust, which would ensure the tools value. However, if the CI either begins to provide inaccurate information or operates inconsistently, the user is unlikely to trust these tools, potentially rendering them obsolete. It is our understanding that, because these tools are so new, the emerging market user is often unwilling to retry them after a failure Thus, it is imperative that, once deployed, these CIs work and work effectively.