Conversational interfaces (CIs) are currently being deployed to help extend access to financial education and services across emerging markets with strong results. In one pilot in Tanzania, CGAP found that farmers who used Arifu’s interactive learning platform saved at rates five times that of farmers who did not use the platform. Although these findings are preliminary, they suggest these tools have a significant potential to change behavior.
While many partner organizations are excited by such findings, they often struggle to deploy a conversational interface because of their inability to answer two questions:
- Where to find evidence behind the value of using conversational interfaces?
- How to decide whether to build a conversational interface in-house or outsource to a third party?
Below are our thoughts on these two questions:
Where to find evidence behind the value of using conversational interfaces?
Multiple organizations told us that despite their curiosity about what conversational interfaces could do for their businesses, they were unsure of how CIs could be integrated and with what results. Many did not understand the inner mechanics of how a conversational interface worked, the design considerations needed when building for different demographics, and, more broadly, the value proposition they offered in practice. The lack of publicly available primers, case studies, and evidence exacerbated their confusion and misunderstanding, which could be addressed if more research reports were made available.
Reports such as this, as well as FIBR’s Artificial Intelligence: Practical Superpowers; GSMA’s Messaging as a Platform; Accion Venture Lab’s Can Chatbots Promote Financial Inclusion; and CGAP’s Interactive SMS Drives Digital Savings and Borrowing in Tanzania are important first steps in addressing the learning curve. The next step is to supplement these reports with in-depth case studies that capture the challenges to be addressed, the value proposition offered by a CI provider, and results observed thus far. This deep dive would not only familiarize partners with the different companies working in this space but also better orient the companies around how best to build, design, and implement CI for different customer demographics.
How to decide whether to build a conversational interface in-house or outsource to a third party?
Below are a few of the considerations organizations should factor into their choice:
- Security considerations: As described in the previous blog posts, conversational interfaces can provide customers access to information, transaction services, and tailored advice. For the first category—information—the conversational interface can likely provide answers to targeted questions without needing access to personal information, such as transaction records or bank account information. But the CI will probably need access to the user’s personal information to offer transaction services or tailored advice. This would require the user to submit the personal information over the messaging application, a potential security concern when managed by an outsourced entity or conducted over a third-party messaging application like Messenger or Telegram. To work around this, some use cases will support using a one time password for authentication—similar to PIA when using the interface to access M-PESA. But for use cases that require more in-depth information about the user, organization will most likely need to develop their CI application in-house and in adherence with the company’s security protocols.
- Experience of the technology team: Through the many different open source tools (natural language processing engine; cloud storage; analytic software), non-technical staff can effectively build CIs. However, in some cases, the open source tools either do not work or do not exist for emerging markets. For instance, one company we spoke with built a customized two-way SMS gateway for one of their emerging market clients because the service did not exist in that country. In situations like this—particularly when the technology has not been used before—it may be appropriate to outsource the work to an organization with experience building and working with conversational interfaces rather than trying to build in-house.
As the field of conversational interfaces evolves and their value proposition becomes more clear, organizations will probably have less of a learning curve than those pioneering the technology have now. But to address and minimize the learning curve, reports such as this, along with access to grant support and technical mentorship, can help minimize potential risks while building capacity for organizations looking to trial conversational interfaces.
