Your browser is not supported.
The latest version of Safari, Chrome, Firefox, Internet Explorer or Microsoft Edge is required to use this website.
Click the button below to update and we look forward to seeing you soon.Update now
A whole lot of what’s being billed as ‘conversational AI’ today is not so much about artificial intelligence as it is about logic, meaning scalable AI agents are few and far between.
Take the following example of a customer booking a table at a restaurant.
Simple, right? Chatbot asks question, customer gives answer. It’s like filling out a form, but with a conversational interface.
But customers like to ask questions.
In reality, conversation doesn’t follow a linear path. For example, when you ask someone a question, they’ll often reply with a question of their own. It’s not always the evasive type; often people need clarification in order to provide more useful or interesting responses.
Clarification is especially important in customer service transactions. Let’s zoom out on that restaurant booking example.
The answers a customer gives are often based on variables that you wouldn’t necessarily have pre-empted. If you ask for an order number, the customer might ask how they find that number, they might say, ‘is that the one that starts with ABC?’ They might just say they don’t know what it is.
A customer service agent needs to be able to answer questions like this on the fly, or at least have the ability to find out the answer.
Conversations move back and forth.
It’s not just unexpected questions that throw conversations off linear paths. Let’s zoom out one more time on that restaurant booking example.
After they’ve already asked a bunch of questions, a customer might decide to go ahead and make a booking. But what if there’s no availability? Depending on how keen they are to actually visit this restaurant, they may be willing to compromise with another time or date, or they may have more questions that will help them to decide if the compromise is worth it.
If we think about the order number example again, the same logic holds up. A customer may ask a bunch of questions about how to find their order number, and may still give an incorrect answer, in which case, they need to come back to the beginning of the conversation and try again.
While the idea of an ‘if-this-then-that’ conversation sounds simple, the reality is that they are virtually impossible to pre-plan.
You might start off with a nice little decision tree, but once you factor in the myriad ways in which customers veer off topic, you’ll soon find you have something that looks like this.
But a messy diagram doesn’t affect your business. Or does it?
The real trouble happens a little further down the line. You’ve listened to hundreds of hours of previous customer calls, scripted out every possible path a customer could take, and you’ve launched your lovely linear chatbot or voicebot to deal with these complex conversations.
And then something changes. Your rotate your seasonal menu, offer a new subscription plan, change the price of something or amend your refund policy.
So you go back to your script to make the necessary changes. And this is where you see the knock-on effect.
You’ve only updated one piece of information, but it could affect an unlimited number of other nodes (or turning points) in your pre-planned script. Now instead of making one change to your technology, you end up making dozens. Not only is this a complete pain, but it’s time consuming and it’s expensive, especially if your technology provider is going to charge you a nice hefty sum to rewrite your entire bot.
As humans, we don’t need to pre-script the conversations we’re going to have throughout the day. We’ve developed a pretty decent understanding of language, and we’ve acquired a ton of knowledge that means we can make conversation up as we go along.
And that’s exactly what artificial intelligence is about.
At PolyAI, our scalable AI agents use your knowledge bases, alongside their own unbeatable understanding of natural language, to understand what customers want, and choose appropriate responses.
This means you don’t need to anticipate every possible turn a conversation could take. You simply give the AI agent the knowledge it needs, and it matches the customer query with the relevant response, much in the same way that a human agent understands what a customer is looking for, and uses their training to provide assistance.
The very nature of a predetermined conversational flow makes it virtually impossible to scale. The more twists and turns a conversation requires, the more complex the flow becomes; the more complex the flow becomes, the more expensive it is to update and maintain it.
Using a non-linear architecture, supported by AI, it is possible to craft virtual agents with the capacity to learn and grow in the same way your human agents do.
If you’d like to discuss the possibility of using scalable AI agents in your contact center, get in touch with PolyAI today.