Let’s delve into the intricate arch of designing a decision tree or a rule-based chatbot starting with some context. We want to develop a rule-based chatbot that recommends golf balls best suited to the visitor’s purpose. Now, if you’re thinking golf balls, are they really that different? Oh, you’d be surprised. Like a fine line, the right golf ball can genuinely enhance your golfing experience.
So, we’re setting the stage as usual by resetting ChatGPT4 and assigning it the role of chatbot developer. This assignment, in this case, is to decide which questions to ask and which options to offer. One of the secrets to great prompts is to give the chatbot an example of what you mean that clarifies a lot. So, we provide an example in the form of the Golf Rangefinder chatbot we developed earlier.
The initial request is expressed following the example with ‘Write a question and three to four user choices for the product golf balls. Then add three alternatives to this question.’ Chad GPT suggests three different questions to initiate a chat with, but I’m not terribly impressed with any of them. So, I ask Chad GPT to offer me more options, and it offers three additional options.
Now, whereas I recognize that every question it suggests could help me narrow down the suggested golf ball, I don’t say anything that really wows me. I decided to improve my prompt by creating three more questions and user choices, this time targeting the usage of the golf ball.
ChatGPT’s response to this prompt is much more to my liking, particularly question one alternative eight: ‘What type of golf do you usually play?’ I think this question is non-threatening and generic enough that it should encourage anyone to provide an honest answer. My first rule for a rule-based chatbot is to get the visitor to start.
For the drill-down process, I confirm my choice of question one alternative eight and ask my new AI friend to create the next question the bot should ask, followed by the next user choices. Now, for each potential choice the visitor makes from the offered options, it asks questions that are appropriate for that type of golfer. For instance, a recreational golfer needs to choose between just having fun, improving their scoring, or something else.
Based on that success, I decided to push harder in Proc ChatGPT. For each question two, create the next level of questions and user choices for each possible user choice. This time, it does great on the first option, recreational play, by offering great follow-on questions for each choice. However, for the other options, it doesn’t offer specialized questions. Not a problem, just means I have to slow down and ask it for each individual level 2 question, and I’ll be good to go.
Now, given that a rule-based chatbot is really a dialogue consisting of questions and user choices, I think I’m well on my way. However, being a visual person, I’d like to see all of the options presented in a decision tree. ChatGPT is not currently able to produce drawings, so I prompt it to present this decision tree format in Mermaid. ChatGPT beautifully responds by explaining what Mermaid is in this context, with a disclaimer that it cannot directly create or render Mermaid diagrams, but it can produce Mermaid.js code.
When I do that, I can see the decision tree and immediately recognize that every outcome eventually leads to an end, which is not satisfactory. So, I go back to ChatGPT and ask if it can recommend golf ball brands for all endpoints. It can and does recommend specific golf balls for every endpoint in the decision tree.
Now, I decide to go whole hog and ask ChatGPT to incorporate that into the existing decision tree. And here’s the final outcome of this exchange. You know what I love about this process? It’s how we navigate the delicate balance of being precise without overwhelming users. We’re like detectives, picking up clues, gathering data, and using that to provide a tailored golf ball suggestion. But we’re also like friends, making sure the user feels heard and supported throughout the process. After all, customer experience is everything.
In the end, designing a golf bot is a bit like playing golf. It requires strategy, precision, and sometimes a little bit of luck. But when you get it right, it’s nothing short of magical. So, pick up your metaphorical clubs, my friends. It’s time to play the game of bot development. And with that, it’s time for me to wrap up. I hope you enjoyed this journey as much as I did. Remember, whether you’re a chatbot developer, a golfer, or even a golf ball, be the best you could be and, most importantly, enjoy the game.
