When brand-new, revolutionary technology first emerged, the ICT for Development sector was abuzz with anticipation—and fear—about how it might transform our ability to connect, influence, and reach people worldwide.
No, we’re not referring about November 2022, when ChatGPT was made available to the general public. Rather, we’re talking about April 2016, the month when Facebook Messenger API was introduced, giving developers the ability to create chatbots that could communicate with page messages.
A few years later, in 2018, WhatsApp did the same, granting access to a small number of Business Service Providers, including Turn (formerly known as “Engage”) and infobip, via which companies may create WhatsApp-based services.
The Advocacy Journey of ONE Chatbot
The mission of the advocacy and campaigning group ONE is to combat extreme poverty and avoidable diseases, especially in Africa. Since 2012, ONE has expanded its reach by utilizing mobile technologies to reach new audiences. The organization uses SMS, USSD, IVR, and mobile-friendly websites to make it simple and straightforward for inhabitants of the African continent to take action.
By now, ONE had amassed a mobile supporter base of two million users. Realizing that it would be too costly to continually re-engage these customers through SMS or IVR, ONE saw chatbots as a potential avenue for more fruitful long-term relationship building. Additionally, ONE believed that chatbots would provide a workaround for Facebook’s progressively cryptic newsfeed algorithm, delivering relevant content in a friendly, conversational style straight to consumers’ chat inboxes.
Five Things To Be Learned About Chatbots:
First, start small
It has taken ONE a few years to develop a chatbot that, in our opinion, accomplishes the goals we had set out to accomplish. Therefore, it is first advised to start modest and move slowly if that is comfortable for you.
Delivering amazing results within months is generally emphasized because of the way hype-cycles and finance operate. Additionally, there is the desire to “move fast and break things,” which is often equivalent to “move fast and break things, at the expense of your budget and the safety of users.” The chatbot work of ONE was conducted in an exploratory and pragmatic manner, with no initial focus on goals other than learning.
ONE discovered that a chatbot targeted at young Nigerians that could be accessed on Messenger and WhatsApp had the most chance of succeeding through a cycle of user research, building, launching, and testing prototypes centered around one particular campaign and market at a time. Launched in tandem with the Covid-19 pandemic, ONE quickly updated the bot with information about reliable Nigerian Covid resources, including information and support services, as well as case-tracking information.
Building on the lessons learned from this version, ONE eventually developed the ONE Africa chatbot, which provides comparable functionality along with more actions for supporters and bilingual information in English and French for different markets.
Even with a targeted offering that had little in the way of content and functionality, there was still a lot to learn about designing chatbot flows, creating, adapting, and uploading content, integrating with third-party platforms, developing and testing flows, integrating data-tracking, interpreting data, marketing, and calculating costs and scalability. Make sure you have enough room to properly learn and iterate because this all took time.
Give user feedback and testing top priority
Few businesses were utilizing chatbots to reach their intended audience when ONE began experimenting with them. Only a few “chatbots for good” were found by ONE’s design research, including UNICEF’s U-Report and an inventive story-telling chatbot by WaterAid.
Simultaneously, the conversational interface of a chatbot was more recognizable and approachable for those in the global south, who were more used to utilizing USSD and IVR services rather than mobile websites and applications. That’s why ONE was more interested in knowing how people would perceive them than in mastering the technical aspects of developing and constructing a bot.
Through the use of ONE’s network of young activists in Nigeria and other markets, we routinely consulted with members of our target demographic in order to comprehend this and begin evaluating the usability of our chatbots (encompassing efficacy, efficiency, contentment, and error tolerance). In addition to monitoring comments through polls and open-ended questions in the bot itself, we used WhatsApp groups to collect qualitative insights and continuously learnt what was and wasn’t working.
Words like “interactive,” “funny,” and “easy” were used by users after our first chatbot was tested. They especially liked the conversational tone and localized language. Additionally, they were thrilled to be utilizing cutting-edge technology, yelling, “Naija, no dey carry last, in fact we be the first man!” Even the employment of buttons at each “fork” in the conversational branch to advance the discussion was considered an efficiency rather than a barrier.
Overall, through early and frequent testing, we discovered that a certain subset of ONE’s audience thought the chatbot was a fantastic method to get young activists interested in using cutting-edge technology and to make it simple and quick for them to present what ONE had to offer. By contrast, they claimed that the ONE website—which has since undergone a redesign—felt overly complicated and was primarily intended for desktop or smartphone users.
Through user testing, we were also able to pinpoint issues and suggestions for enhancements, such as message length, the requirement to respond to a minimal quantity of small talk (e.g., “hello,” “thank you,” “OK”), and the addition of further personalization (e.g., being greeted by name). We were also able to improve the way we used language, since users expressed a preference for various language alternatives over the pidgin-English combination we had first used. These possibilities included English, Pidgin, and Hausa. While backend data analysis would provide the “what,” receiving user input on a regular basis provided the “why” and, eventually, the assurance that we could keep spending money on chatbots.
Budget for marketing expenses
Regretfully, when it comes to digital tools, the well-known phrase “if you build it, they will come” from the movie Field of Dreams does not hold true. Despite being hailed as a potentially cost-effective method, chatbots still need marketing spending to attract users.
In the absence of an integrated onboarding mechanism in your use-case (e.g., direct in-person referrals through Community Health Workers), you will have to depend on third-party advertisements, mostly on Meta platforms, with call-to-action buttons that direct users to your chatbot.
ONE conducted tests using a range of internal and external marketing strategies in order to learn more about the expenses and consequences associated with expanding its chatbot user base. For instance, ONE obtained an 8.5% CTR in a trial aimed at re-engaging current Nigerian email subscribers, and 3.4% of those emailed proceeded to join a petition through the chatbot. It’s difficult to assess this performance because there is little to no publicly available, comparable data; that being said, this chatbot-focused campaign outperformed earlier, comparable email efforts that ONE had conducted.
An additional key challenge associated with attrition was brought to light by this exercise: ONE observed a notable decline in the number of link clicks and conversation starts, most likely due to cross-app compatibility problems (users clicking on links who might not have Messenger or WhatsApp on the device used to open the email).
In 2020, ONE experimented with Meta ads to drive Nigerian users, ages 18 to 36, to both Messenger and WhatsApp chatbots. What they discovered quickly was that driving users to Messenger was less expensive and easier than driving them to WhatsApp: one of the top-performing campaigns cost $0.17 for each new Messenger user, while new WhatsApp users cost $1.77.
Because of this, even with the extra expense needed to operate a WhatsApp-based business, just 14% of our customer base uses the app. In keeping with the gendered pattern of internet usage, it was also less expensive and simpler to convert male users, while acquiring female users was 2.5 times more expensive.
You must also take into consideration frequent modifications to Meta’s advertising platform and significant variations in the user experience between an advertisement promoting a Messenger chatbot and one promoting a WhatsApp chatbot, as any digital marketer reading this will all too well know. These factors can have a significant influence on cost and conversion rates.
The overall amount spent on acquiring new users was more competitive when compared to other lead generation initiatives that sent traffic to the ONE website. This suggested that a chatbot would be even more successful in cultivating active supporters. However, because to Meta’s tight grip on the digital advertising ecosystem, all of these initiatives were expensive and time-consuming.
Make a plan for ongoing involvement
The majority of goods rely on offering features and information that will satiate consumers’ requirements or interests to the point where they are driven to return time and time again.
However, as anyone who has created a digital product—or really any product—knows, this is not representative of how people actually behave: users’ attention is sharply divided, and businesses must contend with a plethora of virtual and physical distractions in order to win back users.
Like many, ONE did not factor re-engagement into their design from the beginning, and as a result, they have only lately begun to struggle with their user retention constraints.
The guidelines established by Meta for re-engaging users outside of a “free” 24-hour window that a user opens by texting your chatbot make this task of re-engagement much more difficult. Outside of this window, companies and brands are limited to sending only extremely precise use-case-matched unsolicited messages at a cost, or paying for sponsored messages that have significant restrictions. Regarding WhatsApp, every message needs to be approved by WhatsApp, and broadcast messages are controlled by a mechanism that limits the quantity of (uninvited) messages that can be delivered to users in groups of up to 100,000 messages.
In order to learn from its mistakes, ONE conducted a number of internal tests before developing a rather sophisticated automated re-engagement system that encouraged users to communicate with the bot once more in the weeks and months following their initial interaction.
ONE drastically reduced the re-engagement strategy as a result of what we discovered during these experiments, namely how difficult it was to not just understand the generated re-engagement data but also to get over the broadcast message rules.
It is limited to one re-engagement push within less than 24 hours of the initial chat, which is when users are most likely to want to re-engage, if they do at all. Additionally, a broadcast message system to re-engage Messenger users outside of the 24-hour window has been implemented (after it was determined that, for the time being, WhatsApp restrictions made it prohibitive as a broadcast channel). As of this writing, statistics indicates that 25% of users who are urged by the bot to use it do at least one more action within it; ONE expects to increase this percentage in the months to come.
Ultimately, developing a larger audience engagement plan is the key to successfully re-engaging users rather than focusing solely on technical or financial constraints. In the instance of ONE, we discovered that we had not planned a chatbot supporter journey past the initial interactions, thus we were not always sure where to direct our current users. When creating a chatbot, you also need to take this into consideration.
Make sure chatbots benefit all users
Because ONE’s chatbot work was experimental for a considerable amount of time, it was developed mostly outside of regular business operations and depended on a small team of employees as well as outside specialists with the necessary technological know-how.
For the most part, the chatbot work functioned in a silo, even though the team collaborated with both local and international colleagues to comprehend campaign objectives, create content, arrange user testing, and share learnings. One major problem was understanding performance data on a regular basis and creating accountability and motivation for the constant development that all digital goods need.
In hindsight, we ought to have worked harder right from the outset to better integrate work across teams and guarantee that employees were appropriately engaged throughout the entire business. Additionally, we suggest designating an internal “champion” for the chatbot, whose job it is to encourage the chatbot’s incorporation into regular thought processes.
This situation is not unique to chatbots, though; rather, it is a reflection of how digital platforms are frequently (erroneously) viewed: as a simple communication channel or discrete output, as opposed to a tool with a comprehensive, cross-cutting role to play across many workstreams.
For instance, ONE was able to better understand its African audience thanks to the chatbot work, which benefited both the campaign and policy teams. In a similar vein, the chatbot posed inquiries concerning supporter journeys, contributing to broader corporate conversations on the subject.
The last issue is capacity; when teams are already overworked, the addition of a new platform and its attendant maintenance chores, which range from content upgrades to performance monitoring, might drive them away. The chatbot could be pushed to the back burner in favor of other priorities if it cannot prove its worth to both users and team members, that is, how it can make tasks simpler or more effective.
