I remember when I first started hearing the buzz about bots. My first thought? 'Here we go again...' - a reaction to the endless cycles of hype followed by business-as-usual that typifies the digital sector.

However, over the past few months I've had the opportunity to design a few 'bots 4 good', and I'd like to share what I've learned: how they work, what they could be useful for, and where to start if you'd like to get one. I believe that done well, they could be really useful add-ons to your digital strategy as they provide a rich 'in-between' space for mobile users who aren't fully digitally literate.

1. What Is A Chatbot And Where Can I Find One?

In theory, a chatbot is an A.I. system which mimics human conversation and can help you accomplish tasks. They're often called 'virtual assistants' which give you an idea of the role they're supposed to play in our lives. There are voice bots (like Siri or Alexa), and text ones: we're talking here about text-based bots, specifically ones that work on Facebook Messenger, although there are other mobile platforms like Telegram or WeChat which support them. In the same way that you chat to your bestie on Messenger, you can also add a 'bot' as a contact, and 'chat' to them until you get what you need.

For organisations battling with the Facebook algorithm to get seen in their audience's newsfeed, bots are one-to-one communication channel which bypasses the newsfeed entirely.The fact that text bots are optimised for use on some of the most popular mobile messaging apps is significant in terms of their utility in social development, as they bring together some of the benefits of both social media and mobile websites and apps - and go some way to overcoming their downsides.

In practice, however, Natural Language Processing (NLP), which allows a computer to not only understand you, but to also answer you back coherently, is in its infancy - it works best for specific use cases with a limited range of options: ordering pizza, for example.  It takes time for the bot to become good at talking back, a bit like a child. Realistically, few of us have the patience to wait for our bot-child to fully grasp the English language, and so we give up, frustrated. This is especially true in developing contexts, where local languages, slang, and lower literacy might slow down this process even further.

This is why, in March 2017, Facebook announced that they would be allowing developers to switch off the keyboard in bots... that's right, to stop people from *actually* chatting . Before you get put off, in my opinion this means bots have a much better chance of being useful to your audience (and you). What you're left with is a good old fashioned decision tree, rendered more lively through a 'chat-like' interface.

2. Components Of A Messenger Bot

When you design a bot, you're helped by the fact that you can only operate within a limited set of parameters. The basic building blocks are 'message cards' - literally a message from the bot to your user, which might be informative, inconsequential banter, or a question. A message card might just be an image or a gif too. Imagine you're sitting down with a member of your target audience. What conversation do you want to have with them? What conversation do you think they want to have with you?

You can also create carousels of message cards, and add headings, subheadings, images and links to provide users with a menu of content to choose from. Each message card can also include  buttons underneath to prompt further action, for example, a link to an external webpage that opens within the Messenger window, or a 'Dial' button that triggers a phone call.

Finally, you can create 'quick replies' - little buttons with which your user can answer questions or navigate through the decision tree you've set up for them with one tap.

BBC Earth's bot, with message cards and emoji-based quick reply buttons.

These options might seem limited but they allow your audience to:

• view and browse snippets of information both written and audio-visual

• make choices about the kind of information they want to see

• answer questions about themselves to inform the information you serve them

• view longer pieces of content pulled in from a website, without leaving the familiar environment of Messenger

• share content with you - from photos, to videos, to stories

• share content with their friends

In this sense, they're a fantastic transition space for those members of your audience who are using a web enabled phone but have low digital literacy - done right, they can basically provide a stripped down web-like experience.

3. Examples And Use Cases For Social Good

A good example of a bot which uses open-input successfully is my new best friend Poncho the weather cat, who sends me personalised weather updates and sassy gifs. It works because it's specific, and it's fun. 

Another bot, which I worked on, is a bot allowing African media organisations to serve the latest news stories to their audience, as well as gather eyewitness accounts to help journalists cover stories like corruption or service delivery failures. Again, it works because it's simple, and also sticky, because users can keep requesting more news stories from the bot until the stories run out. U Report also has a pretty good bot, pinging users periodically based on their demographic information to take part in surveys. Where it falls short is a complete lack of personality - it has no warmth to it, and therefore the messages start to feel intrusive after a while. A good bot will mimic a real human interaction just enough to put users at ease, whilst signposting them to a real person when the bot can no longer reply.

One use case which I haven't yet come across, but which would be valuable for many projects, is a simple on-boarding bot, for when users message you on Facebook for the first time. I'm sure you have lots you want to tell them, and now's your chance: who you are, what you do, what services you offer - and crucially, what you want them to do next, whether it's sign up for educational SMS or answer a quick poll. This is a process that would typically be manual, relying on a good social media strategy and an assiduous social media manager having to repeat the same interaction over and over again. With a bot, you can automate this process, saving you time, improving your brand's user experience, and hopefully, pushing more users to where you want them to be.

My experience of launching bots in countries like South Africa and Kenya has been interesting in terms of user uptake. Some users instinctively 'get it' - they can tell they're talking to something automated, and roll with it. Others however, clearly believe they are talking to a human, and are frustrated by this bozo's apparent inability to help them. Practitioners should therefore think carefully about how they introduce their bot - remember that bots plug straight into the 'message' function of your page, so when users message you, they will expect to be talking to a page manager. 

Bots are also not appropriate in situations where there is anything like a grey area, requiring sophisticated human decision making skills and empathy to answer users - especially in cases like sexual violence. This ties in with the question of open-input vs closed decision tree style bots. If you open the door, your user will step through, and pour their heart out to you. You need to be sure your bot has the skills to answer properly, or run the risk of causing harm to someone who has taken the difficult step of reaching out for help.

If you'd like some help thinking through how a bot could benefit your social impact programme, or designing a prototype to test, get in touch on Twitter @isabelleamazon or mail isabelle@panoplydigital.com. I promise we can have a real conversation, grey areas and all.