Lessons Learned in Building Sustainable and Accessible Chatbots

Chatbot Logistics and Development

Before starting a chatbot project, it’s important to sort out logistics to prevent delays later in the process.

Step 1: Choose a business solutions provider (BSP) to help you connect to the WhatsApp Business Platform and build your chatbot.

CARE spent several months reviewing different service providers and assessing feasibility and fit before choosing Turn.io.

Key learning:

• Before reviewing partners, a good step is to align with the team about functionality requirements and desires, internal resources and budget. For example, if the chatbot is meant for 1:1 support and inquiries, a solution that has strong customer service type functionality would be important. Alternatively, if your team has strong developer support a solution like Twilio might be more cost effective.

Step 2: Set a realistic timeline for your project.

CARE assumed building or programming the chatbot would be the most difficult and lengthy step, but learned that planning the chatbot content and communications flow was where we should’ve budgeted the most time. Writing the chatbot content and its desired flow, building the chatbot, testing the content and chatbot with target users, running quality assurance, and developing a go-to-market plan can take quite a while. Your timeline should also account for any extra time needed if you are collaborating with multiple teams on the project or if you need to produce the chatbot in multiple languages. Some teams we worked with included communications, programming and IT.

Key learnings:

• Verification of your chatbot line through WhatsApp can only be attempted once a month. To prevent a rejection you need a certain number of users already interacting with the chatbot. Make sure you provide early access to a group that can test the chatbot and help establish verification.
• Account for the possibility that your chatbot does not get verified the first time and build that potential delay time into your project timeline.
• Using a landline number for your chatbot can be harder to set up as landlines can’t receive a code to link the WhatsApp line to the Facebook page.
• Ensure that your chatbot line is connected to the Business Manager of the local country office’s Facebook page. This ensures that your chatbot is linked to your Facebook page when it comes time to launch and advertise the service.

Step 3: Develop your content and communications flow

Researching, developing, and finalizing chatbot content was the step that took up the majority of the project timeline–even more so than we had initially anticipated and planned for.

Key learnings:

• Chatbot flow built out in a Powerpoint presentation format is much easier to digest when in the development and editing stage. Check out this example from CARE Ghana!
• Ensure that the language you choose to use is the best fit for those you are trying to reach. No matter how amazing your chatbot is, if your audience cannot understand the language it’s in, the potential impact could be hampered.
• Content needs to be designed to encourage “deep flows”. Ask yourself, how are you creating content that encourages the user to continue interacting? The average user interacts for 2-3 messages. Consider how to provide valuable and impactful content within the first 2-3 interactions as well as how to drive deeper engagement​​—for example, by including a quiz.
• In content design, consider the format that the content will be uploaded–as regular messages, using Threads (which allow you to create simple, yet high performing message sequences like surveys or quizzes), or using custom-coded Stacks (which requires more technical savvy, but also has powerful uses like connecting to live content sources, creating list messages, using personalized variables in messages, and more).
• Consider using characters native to SMS and texting such as emojis, concise language, and keywords to make your copy more effective.

Step 4: Build, test, and iterate your content flow

CARE found that even the communications flow which we had initially finalized needed iteration once we started uploading content to Turn.io’s interface and testing the chatbot with prospective members of our target audience. This part of the process yielded several key learnings that built on those established in our previous step.

Key learnings:

• It’s important to account for localized content development and running it by country level participants since content developed by subject matter experts may not be accessible or relatable to the target audience.
• There is a learning curve when asking people to engage with an automated chatbot as most people using WhatsApp are used to personalized 1:1 conversations. Consider how best to build your chatbot to suit this kind of conversation flow or include a disclaimer about the automated nature of the chatbot.
• Ensure that any message flows that results in a dead end have a footer message that directs the user back to the original or main menu.
• Include a ‘help’ catch-all message that directs people to contact information or a number to call in case of an emergency.
• Consider data usage. If you are linking to online content from the chatbot, will your audience be able to access it or want to use their data to do so? Ideally, contain as much content within the chatbot as possible.
• Include a call to action to subscribe or opt-in to notifications. This way, you can re-engage users or notify them when new content is available. It also allows you to conduct user surveys in the future.

Step 5: Staffing and troubleshooting the chatbot

CARE learned that building an organizational structure around the chatbot is crucial for the service to be sustainable and scalable.

Key learnings:

• Adapting to new technologies and incorporating them into projects and workflow requires a team, even if it’s just one individual, that is engaged in the project and trained in using the chatbot development platform.
• When the time came to actually build and debug the chatbots, CARE found Turn.io’s platform to be user-friendly, even for non-technical staff. At any point where we faced a technical hurdle, CARE turned to Turn’s extensive documentation on how to use their service.
• CARE was also able to get support from Turn by reaching out via email or better yet, their WhatsApp helpline (+1 650-600-1008). This additional support was one reason CARE chose Turn.io as its solution.

Marketing and launch

Just because a chatbot has been built doesn’t mean people are going to automatically engage with it. CARE strongly suggests building in a marketing budget and plan that includes a promotional campaign to accompany any chatbot launch.

CARE learned this lesson by working with CARE Ghana on a chatbot project. After eight months, the chatbot had just 200 users. However, when CARE Ghana included the chatbot details in an online ad campaign promoting positive nutrition, adoption of the chatbot skyrocketed.

Key learnings:

• An ad campaign via Facebook is an ideal solution. Facebook has ads that a user can tap to automatically be connected to the WhatsApp chatbot.
• Utilizing Facebook targeting, ads can be targeted to specific populations that are more likely to find the information relevant and start a conversation.
• To drive a higher conversion rate from ads to chat, consider retargeting an audience with a specific call-to-action. Or to encourage existing users to opt-in to notifications. CARE Nigeria and CARE Bangladesh did this in their campaigns to urge people to take their feedback surveys.
• In addition to a Facebook ad campaign, there are other ways to encourage adoption, such as:
  • Add to /where-we-work About page for the regional or country office
  • Add to email signatures
  • Add to the About Us section on Facebook pages
  • Promote through organic social media posts
  • Add to relevant offline materials like program flyers or relevant educational materials.

CARE’s final insights and thoughts

Using chat for impact opens up so many possibilities, but it’s crucial to approach each project with a practical mindset. Consider if a chatbot is really the best platform to drive impact. Will you be reaching your target population? Do you have the resources, both financial and human, to carry out a chatbot project and to sustain it? It’s also useful to see what peer organizations have done in the field. Have they used chatbots for a similar purpose? Is what you’re building going to overlap or add to other projects/products that are in use?

CARE’s continued partnership with Meta, and all its platforms, provides unique opportunities to learn and experiment with technology for development and program impact. Programs like the immunization chatbots on WhatsApp have helped broaden CARE’s perspective about the ways social media can be used. We hope to continue building on the learnings from this program to develop new chatbot use cases, and sustain and improve upon our active chatbots.