With this framework in mind, this helps build out a bank of user responses, that would be recognised by a chat bot to generate a meaningful response.

To achieve this, I pre-determined variations of the different types of questions users are most likely to ask in the given scenario within a library of 'training phrases'. This would allow AdaBot to extract important information to help service the request and generate a meaningful response.

To control the flow of the 'conversation', answers were kept short with an awareness not to include too much technical jargon within each message. Where appropriate, personalisation was also used to create a tone for the chatbot and connection with the end user.

User Questions:

Pre-empting variations of questions for example:

"I need help with formatting in Microsoft Word"

"How can I quickly format a document in Word?"

"How can I make the text the same in my document?"

Chatbot Responses:

Greetings variations:

"Hi, I'm Ada. How can I help you today?"

"Hello, nice to meet you. I'm Ada. What's your name?"

"Hi! The name's Ada, what are you looking for help with today?"

Clarification/confirmation responses:

"So I can get you to the right place, can you give me a few more details of your issue?"

"I think I've found some suggestions to get you started" [user selects from choice buttons]

"Sorry, I didn't quite get that. Can you help me out a little with some more information?"

"Let me know if what I found is not quite right and I'll look again for you"

"Sorry we didn't get there first time, name. Can you tell me a little more?"

"Tell me in a few words how else I can help you"

Suggestion responses:

"Beep boop...I've found this guide {link} in my memory banks which you might find useful"

"Before you go, I also found this video {link} which might also be useful to you."

"Sorry, {name} for not finding quite what you're after. Let me try again!

Feedback bot responses:

"Beep boop! Great! I'm glad you found today's session useful, {name}."

"I'm so glad {name}! Thank you for your feedback!"

"Sorry to hear that, {name}. Leave me with your email address and a member of my human team will be in touch."

Closing bot responses:

"Before you get on with your day {name}, do you mind telling me how I did today?" [user rates chat]

"It was great helping you today {name}. If you need anything else, you know where to find me!"

"It was a pleasure helping you today {name}. Over and out!"

"Great! I'm glad you found the guide useful"

See: Example guidance document

During the chat, the support documentation was linked to the user to access, as well as a short video guide to act as an alternate resource. This resource also linked out to other related resources. The following aspects were taken in to account when piecing together the written guidance:

• Easy to digest, concise information

• Avoidance of too much technical jargon to overwhelm the user

• Written to build skills outside of the guide (eg. adding tips)

• Fully accessible guidance (addition of alt text to images, video captions, use of headings etc.)