I’ve seen a lot of chatbots, made quite a few chatbots myself, and have seen others build chatbots. I’ve learned a few things and here are some key moments to look out for
Set up your bot for success by letting people know to interact with your bot and what it can do for them.
This is fine. *gif* it’s polite, friendly. However…(click) Personalize/customize by greeting the user by their name (click)
What does the bot do? Why should someone use it? Communicate the value your bot provides (click)
always listen. Anticipate everything so you can teach your bot how to behave in any situation and minimize blocks in the conversation. By any situation, I mean variations of each situation.
Train your bot to recognize variations of inputs that map to the same intent(click click)
train your bot to understand variations in message with the same intent, anything on your menu may trigger the task (click)
QA: Prepare for non-verbal and non-text inputs like emojis and images, weird ways of saying the same thing, etc. (click)
Prepare for chit chat and profanity, existential questions, etc.
find data points to echo back to the user
show that it listened
gives chance for user to correct misunderstanding instead of asking them to explicitly confirm.
Instead of this, (click) acknowledge what the bot heard before offering an answer so people know what the answer is based off of
So, if this was wrong, the user will be able to correct it
Give reasons / data for people to trust the bot’s answer and to make decisions easily
(click) (click)
Give detailed information so people can make informed decisions
Give reasons / data for people to trust the bot’s answer and to make decisions easily
taxi? What’s that? This is a room service and housekeeping bot.
(click) generic fallback vs specific guidance
(click) reiterate what your bot can do to set the right expectation
(click) handoff to a human
Autopilot, bot building platform, using ML and NLU
Build once, deploy on any comms channel. deploy as IVR. Then deploy same as Chatbot. Same as Alexa app. No code changes needed.
Why? for bots to be useful, they need to be on the channels people prefer.
Autopilot is three things.
It is a NLU engine. And that is where we started when we announced Understand.
Autopilot is a conversational application platform that lets you build sophisticated workflows.
And it is an omnichannel hub that helps extend those workflows to whatsapp, line, sms, chat or any other supported channel from the next gen like Alexa or google assistant.
24/7 uptime automated bot, easy to handoff to human
If you look at what you need to build an intelligent IVR, or Bot.
You’ll need NLU, Dialog state management, Business logic programmability and channel integrations to be where your customers are.
Today with what you have, you only have NLU part, you’ll be left to do everything else.
Or you have vertical based bot builders focused on sales/support but very limited customization.
Autopilot gives you the NLU, the application platform and the omnichannel hub to give you all the connectivity.
When would you use Autopilot? When would you automate an interaction?
Think of booking an appointment. It’s a well defined workflow, you know the questions and you have an idea what the answers will be. Its well suited for something like a self service bot or virtual assistant.
Now think of buying life insurance. This is a task that is partially automatable, and partially not. You want to automate the first portion of the task (the data collection of the person’s age, income, etc.) and then pass it along to an agent to complete the sale by making it a more consultative transaction.
Lastly think of something like losing your wire transfer. Finding your lost wire transfer is rocket science. You cannot solve this with self service or with an agent. You need to go all the way to an international FXS to find your wire transfer.
Some forms of customer service are suited for self-service, some are hybrid, and some are for specialists or agents.The problem lies when the self-service workload is sent to an agent. Sometimes customers actually prefer self service.
Or if you send a lost wire transfer case to self service, the customer is going to be nervous and they’re going to try to break through your IVR/bot to try to get to a real human.
Tasks are what you want to do
Actions are how you make up Tasks, can be combined together
How are tasks triggered? Samples trigger each task: different ways of saying the same thing (saw in Moment 2:anticipate
Autopilot Actions are JSON commands or instructions that tell Autopilot what to do. With Autopilot actions, your virtual assistants can say or show something to the user if the device has a visual display, remember something for context, collect data, or redirect so you can control the flow of your dialogue.
Actions are how you program the Tasks that your Autopilot virtual assistant performs and are executed when a user invokes a Task during the dialogue. A user can also land on a task by being routed by the natural language router or by being redirected from another task.
Autopilot uses a request/response model like TwiML, Twilio’s XML language. When a request comes in Autopilot performs the natural language analysis for you, abstracting it away so you don’t need to worry about it, to route to the user to a task which is configured with URLs that point to your app. Your app then responds using the Actions, and Autopilot does the work to make sure they work across all channels.
Data collection is one of the most common tasks that bots perform. The { Collect } action enables you to ask questions to users and efficiently collect answers in one flow, grouping questions together
The Autopilot { Hand Off } action enables you to transfer a user from an Autopilot session to a human agent, passing along all the context of the conversation.
Click for gif
DEMO of Autopilot. Here we have DocBot, the world’s best medical bot. We’re going to automate this flow: you either want to get today’s on-call staff, make an appointment, or talk to the hospital. So we’re going to build a virtual assistant that does those 3 things.