One of the key aims is to use technology in the appropriate way to let teenagers know there is somewhere safe they can ask questions and get help. The chatbot is conversational and friendly, asking direct questions from the start to gather feedback.

It lets users know they are talking to an automated chatbot (this is important — in this situation, being very transparent about who/what the user is talking to could make a difference in trusting and using the bot again) and offer assurances about privacy and removing data.

The assistant starts with asking what you like to be called with free text input, before moving on to a guided navigation tree structure where users can select input buttons and find out about the service and get more information. There is also an opportunity for a live chat handover to an appropriate operator if required — though this depend on availability (it’s a charity after all). 

Tonally, it’s consistent accross all the touchpoints — microsite, Youtube videos, the chat itself and feels friendly and simple to understand- making sure the important message stands out.

Key takeouts for charities thinking about Conversational AI initiatives

• Keep an appropriate tone for the target audience — in this case they are likely well versed in YouTube, Instagram, TikTok etc, so it needs to feel relevant to them.

• Conversational style, how questions are framed, look and feel are all important.

• Being upfront about whether the service is a bot or a human from the start answers a key question ‘who am I talking to, can I trust them?’ make sure any live agent handover is flagged up clearly. 

• Don’t overcomplicate the bot. The key design principles of conversational AI should apply, the design should be invisible, helpful and relevant at all times. 

• Being rules based (no NLP) is fine, as long as the conversation answers the questions is needs to. The downside is that having no NLP elements can make the bot feel more ‘artificial’ or single track — this can limit re-engagement.

• Make it clear how you can get support outside of office hours, and if using NLP do homework on trigger words and phrases, You can base these on existing call centre scripts, social media listening or customer service frequently asked/requested questions and answers.

• Co-create with target service users if possible to ensure tone and the questions being asked are relevant and useful — ultimately leading to a better service and usage. 

• Personalisation can really help to up engagement. Sketching out user personas and matching flows to their needs is a part of this process. 

• Answer questions upfront about human/bot interface privacy questions. 

What to do next — 4 things we’d tackle

• Well, as its on webchat only, all the CTAs are directed there, I’d look at extending to Facebook Messenger, SMSand possibly WhatsApp and drive users from there to the full experience.

• Having more conversational interactions could work well, it currently a little restrictive to have just one free text input (what’s your name?) at the start perhaps remove the free text option after the brief onboarding for clarity. 

• It would be great to see a little more personality within the bot itself, it doesnt necessessarily need a name/persona, but would have been interesting to see how it might encourage longer conversations. 

• Make a little clearer personalisation and conversation tree flows — between users/worried friends. Giving each a richer flow of content can help to keep users coming back for more relevant content. 

Obviously its a great initiative, and hopefully the trial produces useful feedback on data and usage to bring forward into the next version and other initiatives channels and enagement campaigns.