Decisions, Choices, Chatbots — When to Utilize Their Power
The big vision for an omnicompetent chat bot allows for open questions with open answers. Such a truly intelligent chatbot would be able to respond to any query with useful information.
To understand how complex this is, imagine a google search, which doesn’t only have to understand what it is you’re looking for, but get it right on the first try every time. Then, it has to extract exactly the information you’re looking for and phrase the answer in a way that feels natural to you, the reader. So far, chatbots are built differently. They mostly utilize complex Decision Trees.
Decision Trees and Pizza Pies
To illustrate my point, I’m going to describe a fairly simple decision tree for a pizza restaurant.
Top Articles on How Businesses are using Bots:
1. WhatsApp for Business: A Beginners Guide
2. Revenue models for bots and chatbots
3. Series of stories on AI, chatbots and how can they help businesses
Initiation — What does the Customer Want?
If the chatbot is supposed to be omnipotent and human-like, it has to start the conversation like any human would:
“Hi, what can I do for you?”
We will then assume that customers can only want one of three things:
- Talk to a human being
- Order food
- Book a table.
Not too bad, right? Three options should be quick to figure out.
The first challenge is that even those three simple tasks aren’t always phrased in the same way.
There are three main ways of dealing with this:
- Manual training in which any possibility of asking for those options that you can think of is entered into a database. Then, more are added whenever human aid is necessary to sift through requests that the machine still doesn’t understand. Even for the first request in our example (“Talk to a human being”), there are hundreds of ways of asking for human assistance, so it takes a lot of creativity, paired with trial-and-error to get this one right.
- AI-based learning in which large “training sets” are used to train a neural net that might use keywords (“human”, “assistance”, “help”, “speak”) to figure out what the client might want. If this seems a bit like overengineering, that’s because it often is. There are tools and startups that help with this, but for a simple pizza joint, this seems a little much.
- Give clear options. If a customer’s behavior can be boiled down to three or four choices, just offer those. Make them clickable if possible. While this reminds us of the bad-old-days of automated phone help lines, it is a little quicker and a lot less frustrating for users than the most common chatbot response: “Sorry, I don’t understand what you’re saying. Please try to phrase differently”.
We Know the Problem — What’s Next?
On websites, giving the answers necessary for those three options is a bit easier. With a click on “Contact”, “Menu”, or “Reservations” on a restaurant’s website, customers are already where they needs to be.
The next step is mapping out the response to each individual customer problem.
Talk to a Human Please
In the first case (Talk to a human please), a chatbot can either connect to an employee via chat, or offer a “call now” button. This branch of the tree is pretty short and can be solved quickly.
Order Food
The second case is much more complex. The order process is different for each customer. For regulars, who already know what they want, the chatbot needs to understand 100% of the order to proceed to delivery and payment. It might even have to save a database of old orders to ask: “Hi John, Would you like to order the same pizza as last time?”
For new customers, it might need to show the complete menu or offer options (Starters/ Pasta/ Pizza/ Desert/ Drinks) to choose from. From this point on it becomes more and more complex and customers are quickly wondering why they chose to order pizza via a chatbot instead of just checking. out. the. website.
What now? Are Chatbots Trashbots?
It turns out that even in the fairly straight-forward problem of ordering pizza, chatbots are not the omnicompetent gods that were promised.
What Chatbots can do however is support customer service. If a customer starts his journey off by asking simple questions, chatbots can provide simple answers.
They can also suggest answers for the most common questions to service agents. Chatbots can give information concerning opening hours, phone numbers, order or tracking information and many more things that are routine inquiries for customer service reps.
They can also be used for fun quizzes or marketing efforts that require precise, or multiple choice answers.
Chatbots could be great teachers assistants for homework. They could grade, explain, and figure out problems for children.
As is so often the case, it is all about the use case. Even the most complex hammers are still not great tools for turning screws.
If you have any questions, or anything to add (or correct), feel free to do so. I’m currently looking into anything AI-related, so if you have any must-reads, interesting companies, or topics to discuss, please share. And if you liked what you read, sharing and clapping is much appreciated.
