Can you read between the lines of a chatbot?

When humans engage in discussions with others, we often are able interpret the subtext behind what is being said.

We read between the lines to determine others true motivations, biases or underlying beliefs. We might look for clues in the words they choose, their body language, or the context in which they are speaking to help us interpret the true meaning behind their message, or what is their needs and intentions. For example, a study published in the journal "Science" mentions that people were able to accurately guess a speaker's emotional state based on brief vocal cues with an accuracy rate of over 70%. Other studies have shown that people can accurately interpret deception, sarcasm, and other forms of subtext in communication. Also equally important to conversation to be successful: we expect that the other side of conversation captures OUR intents and needs on the conversation, and answer on that.

Overall, our ability to interpret and convey subtext is an important part of human communication, and it can help us better understand the intentions and feelings of those around us. When it comes to digital communication, it is important to consider how we can design interfaces and AI solutions that can effectively utilise this subtext and provide a more human-like interaction. Or other way around, how to make sure our personalised experience on digital channels capture the right intents, motivations and expectations?

Can you find a human from segments?

Usually, user needs are described in simple structured ways to describe their behaviours or clustering for "personalisation". Contents and messages are written keeping those segments in mind. Naturally this type of user segments are rough averages, archetypes based on demographic factors such as age, gender, income, and location. While these factors can provide some insights, they may not capture the full complexity of users' behaviors and true needs.

Several studies and industry experts have argued that traditional customer segments have limitations in accurately representing users for various reasons.

The Impact of conversation on User Experience

As argued we are skilled at interpreting meanings from communication with humans. However, we often neglect most of the information from non-human channels. Our brains work differently when we talk to other people, or when we read or hear non-interactive “static” content. Despite the constant flow of information from web pages, ads, billboards, we tend to ignore most of it. On websites, users can ignore what they do not need and actively search or browse for the information or products or tools they want. Look at any given heatmap of any given website use, and you see how we skip and hop over info, but focus on what we need. If the message directed to me based on the segment I happen to belong is not meeting my needs, I just... skip. I skip it. That’s it. Static looking wall of text interpreted my intents wrong, but no biggie! I found my button, pressed it and statistics look that I am a happy customer who got to use the service.

However, in conversational interaction, the situation is different. As the context becomes conversational, our brain wirings switch. We come in loaded with expectations and subtext interpretations that we do with humans. “This thing conversates with me, so I expect now the rules of human-to-human conversation apply”.

If website performs well, one might think users were “happy” with the content and messages. But, it might be they were just ignoring the false assumptions of the needs and found on their own what they needed.

How conversative context can display the disconnection between segment and intent?

I have a real life example of how message is changed depending how it is conducted. We lately moved to new apartment, and I read the internet page of a household security service. I considerend the servicce was OK, I almost bought.

Due happenstance on following week a representative from company came in and representative (young person) explained the service to me. It was scripted pitch that the person talked trough. I was appalled. The message was not based on my needs with the service. Message was annoying to me, felt like obscure fear-sales. I refused to buy that service.

Later I re-read the website, and the content was same. Why was I not appalled by the message from website? Because I was able to skip most of the info (fearmongering) when reading, but I (over)interpreted that from human-to-human communication.

I don't blame the salesperson. I think he did a good job, but I was just difficult customer, and my intents and needs were not as the most common ones. Therefore, I was annoyed. I got into conversation where other side assumed what I want, and did not leave that assumption even after I stated, “that is not what I need”. A bit more experienced salesperson might have ignored the pre-scripted speech, comments, and hooks, and talked to me more of the stuff I had in mind.

I can sense the same also when I am chatting with pre-scripted chatbots. If I do not find the info I need from static knowledge bases, I get frustrated. But If I do not get the same info from chatbot, and it does not understand me, I get ANGRY. On my real life case, the website's content was the same as that of the scripted pitch, but I could ignore irrelevant information and search for what I wanted on the website. However, when interacting with a chatbot, I am loaded with expectations and subtext interpretations, making it vital for other side of conversation understand needs and expectations.

Life centricity is the key for AI powered services