Epileptic Trees: Why Decision Tree Scaling Is Causing Chatbots to Fail

by Impactful Ai

In the Black Mirror episode, Be Right Back, a company creates virtual “copies” of real people based on past conversations. It’s an experimental service, and one that’s not without myriad flaws. And it’s also worlds beyond anything currently in the real world where conversational AI is concerned. 

For all the lofty promises made by chatbot vendors, the technology has fallen well short of expectations — so much so that you could very nearly call them snake oil. 

So, what went wrong? Today, we’ll cover decision trees and why traditional chatbots struggle to scale.

The Trees Have Too Many Branches

The problem is largely one of design. Most chatbots leverage decision trees that are incapable of effective scaling. While such a mapping model might be suitable for basic automation, human conversations are complex. 

Conversations rarely progress in the logical fashion a basic decision tree requires to function. They almost always involve multiple subjects and intents. Why? Because no one speaks in an “A to B” conversational style. 

As a result, single mapping compounds exponentially as you string paths together, quickly becoming confusing and unmanageable. There’s also the fact that you cannot possibly predict every angle a lead might pursue and every question they might ask. Nor does single mapping account for the fact that leads often ask multiple questions at once, expressing multiple intents. 

The first step to building better chatbots, then, is designing better decision trees, right? Brainstorming a broader mapping model that allows you to consider multiple intents, and augmenting it with the capacity to collect contextual information you can draw on later. With multi-intent mapping and machine learning, your bots can jump between intents with relative ease, since there’s an internal, fundamental logic driving this mapping. 

So how exactly do you deploy this kind of mapping? 

Building Better Intent Decision Mapping

The first step is to review your internal scripts. Every single team has sales scripts, talk tracks, calls that they’ve ‘recorded for quality assurance.’ This is going to be one of your most valuable resources in building out your multi-intent decision tree.

Look at the questions your customers ask, the information they want you to provide them. Look at how frequently they combine multiple intents into single statements, and what intents frequently go hand-in-hand. More importantly, think about what you might want to ask your customers, and how they might respond. 

It goes without saying as well that if you’re using a third-party AI provider, you’ll want to make absolutely certain  that it can handle multiple intents 

From here, you want to think about your script itself. How do you want it to evolve? If a customer heads down path A, what happens if they want to switch? That’s where most chatbots fall apart.

They’re a train on a train track, and can only go forward. If they go off the path, they crash. You need to design yours to be more like an ATV — it follows an established trail, but can easily go off the beaten path if it needs to. 

Lastly and most importantly, you need to focus on your customer, not your chatbot. Let the humans drive the conversation. Deploy a sophisticated solution that can adapt to the way they actually talk and ask questions. 

Sending Bots Down the Road Less Traveled

As always, there’s a grain of truth to Black Mirror. In Be Right Back, the experimental androids developed from the company’s “virtual humans” ultimately fail. The reason? 

Because they remind people that they’re bots. They behave in ways we don’t expect, or that are detrimental to the user experience. If your own chatbots are to avoid this, they need to be more personal, more sophisticated, and less reliant on set questions and answers. 

Better decision mapping represents the first, arguably most important step in accomplishing this. 


Learn the Latest in AI: Follow Impactful AI Today

Impactful AI is on a mission to redefine the AI narrative by showing what’s possible when we embrace the collaborative potential of this transformative technology. 

Are you looking to stay up to date on the latest AI developments, trends, and insights? Sign up for our newsletter to gain access to specialized content created by some of the industries biggest names.