We’ve been hearing a lot about chatbots and voice recognition interfaces lately. In fact, at the end of November, Information Development World is hosting an entire conference dedicated to the topic. My customers have been asking lots of questions about chatbots these days, too. So I’ve been doing a lot of research, thinking, and prototyping this next step in user interaction.

Using chatbot technology to create a virtual agent makes a lot of sense. A chatbot can offer a way for users to get help from your company without interacting with humans. When programmed well, chatbots can also hold your user’s hand as they locate the information they need. Rather than just searching for a needle in a haystack, a chatbot can interact with the user to locate the appropriate haystack to start searching. Or maybe even take the user directly to the needle.

There are a number of challenges that we face as we start to employ more chatbots. Here are a few that come to mind.

Challenge #1: How to Speak Human

I have been told by people who are tasked with creating corporate chatbots that the most difficult part of programming a chatbot is making it “speak human.” Let’s think about this. The primary reason that companies are deploying chatbots is to save money by needing fewer people to answer the phone. People are expensive. Chatbots are not. This means that the first challenge is to make sure the user doesn’t give up on the chatbot in frustration and simply pick up the phone. While many people want to be ‘self-service’ these days, if the chatbot frustrates them, it’s a non-starter. The phone prevails.

How do we get a computer to interact with a human in a more human way? The chatbot needs to have a “voice”. It needs a personality. It needs to be able to phrase questions and provide responses in a very believable way. Programming an interface to be human is very challenging. First of all, we are all different and we speak differently. Heck, my son is at a university in Texas and he keeps saying “y’all” to me. Y’all? I don’t speak Texas. But a person in Texas might expect a chatbot to do so.

You need to decide on your voice, tone, personality, emotion. Have you noticed that Siri can be pretty snarky? Some people like that. Others, not so much.

Speaking human is the number one challenge for chatbots.

Challenge #2: Understanding What the User is Asking

People are inconsistent. We speak inconsistently. We write inconsistently. And we search inconsistently. You and I may not use the same words when we look for an answer to a problem. I may not use the same words twice if I need help for the same problem. We are just unpredictable.

The first task of a chatbot is to figure out what you are asking. Perhaps the chatbot says, “How can I help you?” and you put in some question. It may or may not be grammatically correct. It may contain spelling errors. In fact, there is a chance that the question won’t make sense at all. What’s a poor chatbot to do?

Often, the first response from a chatbot is a question to your question. “I see you are asking about unidentified flying objects. Is that correct?” And often, the user gets frustrated right then and there. Hand me the phone!

Unfortunately, a chatbot cannot help you unless it understands what you want to know.

Challenge #3: Too Much of the Wrong Information

The back end of a chatbot can be a number of things:

  • A content management system
  • A cognitive system
  • A set of files stored on a hard drive

Chatbot technology doesn’t make your content better or easier to read. It doesn’t make your content more searchable or easier to find. In other words, those same problems you have with your content do not go away because you’ve now got Siri searching it. If you haven’t cleaned up your content first, then your chatbot gets to deliver the problematic content in a nice user interface. Get me the phone!

In addition, if you don’t have a clean, detailed taxonomy that has been well thought-out and makes sense, a chatbot is not going to do a better job of finding that needle in your content haystack that you would if you were searching with a browser.

I think we call this lipstick on a pig. A pretty interface hiding the ugly state of content beneath.

Once the chatbot locates the information, it needs to be worded in a human way. The chatbot is not going to reword the content. It is going to present your content. And that takes us full circle back to challenge #1.

Bottom Line

In order to have a successful content strategy that includes chatbot technology, you need to:

  • Understand the meaning and intent of the user’s request.
  • Be able to locate the information requested.
  • Be able to present the information (and all interactions) in a way that is human.

And this is no easy feat.


Val Swisher
Latest posts by Val Swisher (see all)