Tomorrow’s Conversations: ChatGPT-4o and the Future of AI with Hans van Dam

Have you ever put on too much makeup or spritzed on too much cologne? There is a fine line between the right amount and an overkill application for both — when they lose their core purpose, which is to enhance what is already there. 

Chatbots have a similar rule. When a chatbot is too complex and focused on solving every problem imaginable, it becomes counterproductive. 

A good chatbot doesn’t need to do it all — it should simply solve the specific problem it was designed for and enhance the user experience. 

The future of AI isn’t that it can do everything — even if that could be true and it’s what we’ve been made to believe — it’s that we can train AI to solve specific pain points and enhance what’s already there without becoming a one-to-one replacement. 

Hans van Dam, CEO of the Conversation Design Institute, is on the field trying these solutions in real-time, with real results.

In his conversation with Rachel Ann Kreis, VP of Marketing at Landbot, and Jiaqi Pan, CEO & Co-Founder at Landbot, Hans shares his view on developing chatbots and leveraging human behavioral models, and discusses the future of AI and conversation design. 

Human Behavioral Models in AI Chatbots: Motivation, Ability, and Prompts

A sizable part of AI systems becoming more advanced is keeping the fundamentals at the center of those developments — how do you structure a conversation? How do you ensure the tone and voice sway towards building trust? 

Once the answers to those questions are accounted for, you can move deeper into the process in the form of behavioral psychology.

“When you think about driving conversion on a landing page, how do you apply all those copywriting techniques into that conversational interface? That's pretty advanced copywriting if you do that well,” Hans says. “If you don't know what conversation design is and you only understand the foundation — and want customers to do a certain thing — you're better off doing it yourself, a lot of times.”

Generative AI does not have the best track record for crafting sincere-sounding conversations that resonate with customers. If the AI lacks a grasp of the fundamentals of conversation design and you need increasingly detailed prompts to achieve the correct outcomes, it may make more sense to take a manual approach.

The other side of the coin concerns effort — how much of it must the customer exert to complete the relevant task?

Human behavior in this context falls back on three key aspects: 

• Motivation;
• Ability; 
• Prompt.

“People need to have a certain level of motivation and ability to perform the action. There needs to be a little trigger. Basically, all behavior can be explained through those three elements,” Hans says. 

If you want to stop watching so much television, move the remotes so they’re less accessible. To start eating healthier, replace the chips and chocolate candies on the counter with fresh fruit and raw nuts. 

People form habits based on ease of access—the less friction in a task, the more likely it is to be completed. 

“Let's say I'm in the kitchen doing dishes, my hands are wet, and my phone rings, and I see it's my girlfriend. I want to pick it up. I have motivation, but I can’t answer it. That's the prompt,” Hans says. “I am motivated, but I'm prohibited because my hands are wet. I don’t have the ability.” 

This psychological factor of human behavior ties into how likely a customer is to sign up for promotional emails, demos, more information, etc. The more friction you remove from the process, the more people are likely to follow through on the prompt.

How convoluted is your prompt? What is the motivator? Are there barriers hindering the desired outcome? 

Remove the confusion and favor clear, familiar processes as much as possible while implementing new AI features and enhancements.

Quality Control in High-Impact Customer Interactions

Quality control means more than watching for bugs—it involves including different perspectives and allowing high-impact scenarios to play out dynamically as needed.

“Where did you source that data from? Was it just the WhatsApp conversation you had with your two best friends, or was it everybody having these conversations, and are all these cultural backgrounds and linguistic preferences included?” Hans says. 

Are multiple perspectives included in the training set? The integrity of managing the data must also be considered.

“You put four white guys in a room, and they're going to make something that works very well for four white guys, and that's really not where you want to be,” Hans says. 

Having diversity on your team — preferably a good mirror representation of your audience — does wonders and gives them the freedom and time to talk through different conversation models.

If your team is backlogged and has tight deadlines, they don’t have time to ask questions and figure out the best path forward. 

Even though there are 100s of steps and checklists you can make to get a proper model built, it comes down to a few main questions: 

• What is the intention behind the model?
• Who is it for?
• What are those intended users trying to accomplish?
• What is the desired outcome?  

“Being aware and mindful of that helps — it's going to make it so much better already. And then you can get into all the detailed steps and governance around it,” Hans says. “If everybody started doing that, you’re already making a lot of progress as an industry.”

The Nuances of Measuring Chatbot Success 

Success looks different depending on the software's goal — we all know this. Every project has different KPIs and OKRs, depending on the department and industry.

If you’re using a lead gen for a webinar, the marker could be the number of signups you earn. But it’s not that simple — you want qualified signups. Measuring success based on the reality of the project can help reduce fluff and confusing metrics. 

If you don’t know what you’re looking for in terms of performance, everything looks important, which can slow down the review process. 

“For every bot we create, we look at that BJ Fogg behavioral model — what do we want people to do? And then you write the thesis of what you want the system to do, and then assign the proper metrics to it,” Hans says.

Another example of differing metrics is in customer support. There, you’ll often measure automation rates, customer satisfaction, and cost to serve. 

“The cost thing is what many people leave out,” Hans says. “Because sometimes these new technologies are way more expensive than our traditional ways, and your business case doesn't work with it.” 

Looking at the total cost to serve is a good indicator of whether a chatbot makes sense for your specific use case, financially. That means measuring data points such as the labor it would take to build a bot and the cost of the entire customer journey compared to traditional systems. If it doesn’t add up, a chatbot may not be the best choice. 

“Sometimes what happens is, when you build a bot, and you put it on your homepage, you actually get more handoffs to the call center or the contact center because your total numbers of traffic, all of a sudden, have gone up,” Hans says. “With chatbots, there’s never just one isolated thing that you're measuring. You need to look at the whole thing.”

Performance metrics and business metrics are two separate categories that should be reviewed with a different lens. 

• Performance metrics include customer satisfaction, ease of use, response rates, etc.
• Business metrics look towards finances, revenue, turnover, and so on.

Determining if a chatbot is in the best interest of your business and your customers is just as essential as understanding its purpose in the company's overall structure. 

From there, you can determine the bot's semantics, reflected in the design based on intentional research and experimentation. 

Listen on Apple Podcasts, Spotify, or your favorite podcast player.