As with many new promising technologies, chatbots are getting hyped as a new innovative tool that can aide humanitarian response and international development practitioners. There are many reasons to be excited, but suitability is key to success.
What follows are some insights and areas of consideration that our technology and design teams, which has worked on technology deployments extensively in remote and low resource settings, have gathered on how to think about good use cases for chatbots.
1. Chatbot Limitations
The first question to ask yourself: is a chatbot the right match for my project needs? The usefulness and value of a chatbot is dependent on the tasks being assigned to them.
If a chatbot were looking for a job, their résumé might list the following as strengths and weaknesses:
- Perfect math abilities
- Excellent data retrieval skills
- Great time and scheduling management:
- Limited conversational capacity
- Narrow domain expertise
A chatbot will work well for you if the respondents answers are quantitative or easily recognizable word responses, such as numbers or specific words and phrases like “Monday” or “High fever”. Chatbots also excel when implementing surveys with complex skip logic, jumping to particular questions quickly based on previous answers of the respondent. For example, a chatbot can quickly be able to direct responders in survey in the case of answer to question A is greater than 1 and the answer to question B was lower than 5, then send them to question E.
Where a chatbot doesn’t work so well is dealing with open-ended questions and descriptive qualitative answers, for example, “What was your secondary school experience like?” or “How are you feeling today?”
Matching chatbots abilities with the needs of your program is an important initial consideration to determine if it is an appropriate technology choice.
2. The Impact of an Invisible Interface
When users open a standard mobile application or a website, they see blocks of text content, images, and buttons that invite them to interact. This is the interface. In a chatbot, the only interface is a text-based conversation.
That has some advantages – a clean, no frills focus on text – but it also means that your interface is invisible. However, when engaging with the chatbot a user cannot see all the chatbot offers – they will mostly have to guess what the bot can and cannot answer.
As you contemplate a chatbot for your program, ask yourself, how would the user interface look if the information was visually presented in place of a chatbot? Would you end up with a complicated interface involving lots of options and user interface elements? Then a chatbot could provide a better interaction.
If your user interface would be something simple like a search-box with a couple of checkboxes, having to communicate with the chatbot may only increase complexity for the user unnecessarily.
3. Channel for Chatbot Communication
Is your target audience using a communication channel suited for the chatbot technology?
You’ll have greater success if you deploy your chatbot through channels that are already widely used by your target audience and allow for chatbots. For example, if your users are on Facebook Messenger, WeChat, Telegram or any other platform that allows deployment of a chatbot, you will benefit from that existing channel.
Other communication channels like WhatsApp do not currently support chatbot deployment. Having to download an app or registering on a website to use your chatbot creates additional friction for adoption and reduces the response rate.
4. Stress Level of Interactions
In a stressful situation, the last thing we want is to deal with a cognitively impaired machine with no empathy. Many of us have experienced this type of stress when calling to customer service or a medical insurance company and only dealing with a series of pre-recorded messages.
If your chatbot might be providing responses to complex issues that involve emotional or stressful situations, especially if it’s a situation you would usually express empathy or sympathy, chatbots are a terrible match.
5. Use of Notifications
When your users need a nudge to start an interaction, chatbots come in handy.
Mobile apps can trigger notifications and websites can send reminder emails, but for frequent daily pings, users are more likely to respond to a message than to click on a notification, go to an app and enter information.
Warning – do not abuse this ability: if the user never reacts to your nudges your bot has probably become just another annoying attention drag in their daily life. Consider putting in place some mechanisms to detect this situation and adapt by reducing the frequency of the nudge. More importantly, never send a nudge for a task that has already been done.
Examples of this push notification benefit are food diaries and time or mood trackers. Triggers for these nudges go beyond a particular time of the day: bots could ‘listen’ to calendar invites, stock values or Twitter hashtags.
6. Natural Language Processing Requirements
Natural language processing techniques are what chatbots use to extract meaning from written text. Platforms like wit.ai can help you match intentions with what is written, but that’s far from a chatbot actually being able to derive complex understanding from the sentence.
Regardless of all the advancements in the area of natural language processing, there is still a long way before chatbots can understand and have fully human-like conversations.
If the kind of interaction you want to provide requires an intelligent conversation where understanding complex sentences by the user is needed, then you should wait some more time for the field of chatbots to improve.
7. Adding Human-in-the-loop Support
With any well designed software, developers get some type of notification when things are not going well with the system: errors that are reaching users, apps crashing, etc.
Since it’s common practice to have a “Sorry, I did not quite get that, try using simple sentences” as a default response with chatbots, user experiences can easily create frustrations and anger.
Make sure you include a way to identify such interactions and learn from what your users are trying to do with your chatbot. Even better, if you can have a human-override in those situations, you get the best of both worlds: chatbots allow you to scale and take care of the grunt work while your humans attend to more complex needs of the users.
Does your organization have an idea for a chatbot? We want to hear from you – leave a comment on this post and let’s start a discussion on the potential for chatbots in development.