WHY EXACTLY DOES YOUR CHATBOT SUCK?

They are popping up everywhere — today’s business landscape is indeed becoming inundated with chatbots. Many people are interacting with bots far more than they realize as companies scramble to install them, but are bots delivering a better, more efficient customer experience?

The basic premise of streamlining user needs into a conversational platform with vast amounts of useful information, available around the clock is excellent in theory. However, as countless investors in this platform have discovered, implementation can prove quite challenging — and they end up struggling with what appears to be a simple interaction.

Why is this? Why are so many chatbots falling short of expectations?

THEY AREN’T SMART ENOUGH — YET

There is still a significant divide between the current technology and the expectations fueled by the often overblown hype. Many bots operate with simple decision tree logic behind the interface that parses the input to the bot which responds based on a predetermined conversational path. Problems arise when the user leaves the designer’s predicted path and the bot fails to adapt. Building on these simple flows requires the business to correctly anticipate all possible user utterances — quite challenging to say the least.

Maintaining context and revisiting earlier parts of a conversation, while second nature for a human, is still hard for a bot. The following dialogue illustrates this challenge.

Me: Where are we going on vacation?
Wife: Aruba.
Me: When do we leave?
Wife: Friday.
Me: What is the weather like next week?

My wife understands the conversation is about Aruba. She also understands when I ask about the weather next week I am referring to Aruba and not my current location. Bots can stumble when the context of the conversation changes as many designers expect the user to drive the bot by providing all relevant information in every utterance. Designers need to create mechanisms allowing the bot to maintain essential pieces of context to better mimic human communication.

The conversation above, while it assumed the context, stayed on track. Many times, humans interject different thoughts, the focus changes, and the bot falls off. Below is an example of this type of interaction.

Me: How was your day at work?
Wife: Terrible.
Wife: Can you pass me the salt?
Me: Why was it terrible?
Wife: My day or the salt?

Bots still need significant improvement in understanding and adapting to context or switching focus during a conversational flow.

LEARNING A LANGUAGE TAKES TIME

The process of learning to communicate in a new language takes time — even for super bots. Learning human conversation is based on a continuous feedback loop of multiple trials, and many are expecting too fast a learning curve for their bots out of the gate. Companies need to treat bots like a child learning their first language — and give them lots of patient guidance. When we as children correctly mimicked a familiar sound we were showered with praise, eventually discovering the meaning behind the words. It’s unfortunate many users don’t give their bots this same luxury.

Further slowing development, is the fact many companies don’t know how to interface with their learning children. Many users throw acronyms and oversimplified abbreviations commonly used in business at their bots which were designed to understand a formal language. Bot users need to speak in simple language patterns and patiently rephrase when a bot doesn’t understand.

Fortunately, bots can learn from each conversation and improve over time by combining analytics with a feedback loop to identify and match common questions with the correct responses. Also, the power of Natural Language Understanding (NLU) is allowing bots to begin to understand context even when the input is vastly different from the training.

COMPANIES HARBOR UNREALISTIC EXPECTATIONS

Many companies want to go to market with a bot that can handle all user queries from day one but this is a poor model and a recipe for bot disaster. A much more realistic approach is to train your bot to solve for one use case, allow it to gain intelligence, and judge its success based on it’s original purpose.

A good example is the Google bot AlphaGo, which was specifically designed to master the ancient strategy game of Go. The AlphaGo bot excelled in its original mission — easily defeating the world’s best human players. However, if asked for tomorrow’s weather report AlphaGo would undoubtedly disappoint.

Companies today are expecting far too much from their fledgling deployments — far better to start slow, temper expectations, and allow your bots to intelligently grow into the roles for which they were designed.

COMPANIES OVERLOOK THE USER EXPERIENCE

Many companies forget the importance of designing their bots to maximize the user experience. It’s critical to integrate a user experience design expert early and often into all phases of bot development. These experts will be expected to interact and handle the bot like the end user.

To perfect the experience, these designers must fully understand the end user’s journey, challenges, and goals. They must also analyze and polish the conversational flow to keep the dialogues friendly and as natural as possible without degenerating into deception. Finally, its essential designers employ some form of sentiment analysis to identify customer frustration — orchestrating a seamless agent transfer to maintain flow and optimize the user experience.

DATA FAILURE

The hard truth is bots will only be as good as the data — and their growth and development depend on the successful collection and analysis of this precious resource. Many companies are failing to collect and effectively analyze the data necessary to fuel their bots — and are missing out on an immense opportunity to expedite their success.

LOOKING FORWARD

Not all bots today are bad — but unfortunately few are good, with a vast majority falling short of expectations. This result is often due to unrealistic visions, impatience or poor design but also to the growing pains of a nascent technology.

In the future, I will be writing more about bot building and how to predict your bot’s future success. For now, I urge bot parents to be patient with your children — taking full advantage of every opportunity to experiment and improve function. If your bot platform is currently disappointing, or you’re thinking of deploying bots, please remember to tolerantly trust in the learning process, keep expectations realistic, design for user experience, and effectively utilize data when creating your next great AI interface.

Possibly the best news of all — these children are honor students capable of improving with every interaction and the more users patiently embrace the technology the better the bot experience will become for everyone.