Conversational Prototypes make Gen-AI feel real
Image by Midjourney, prompted by Matias Vaara
How can we make our Gen-AI experience ideas real?
In the coming months, companies are scrambling to develop ways to use generative AI in their products and services. Traditional prototyping tools like Figma fail to capture the psychological nuance and interactive nature of conversational experiences.
I’ve been experimenting on a few recent projects on how to prototype AI experiences with Gen-AI tools.
Conversational Prototypes are fast and cheap test versions of AI experiences built with readily available tools like ChatGPT.
Conversational Prototypes can help you:
make the fuzzy Gen-AI product opportunities tangible and understandable
develop an intuitive feel of what’s the experience of interacting with the AI
quickly validate and iterate on conversational AI ideas with users and internal stakeholders
explore questions around ethics, privacy, and law before committing to development
How can you build a Conversational Prototype to test your idea?
1. Design a ChatGPT prompt (GPT-4) that instructs the AI how to lead the conversation.
For example, you can ask it to be a fitness coach that helps people design a fitness plan through a conversation.
In your initial prompt, instruct the AI to act as a conversational agent in your given context
2. Include all the needed context in your prompt
To ensure the experience aligns with your goals, add all the needed context for the AI.
Instructions for the AI to stick to can include:
- tone of the conversation (ie. friendly and encouraging)
- length (ie. 5 questions)
- desired results at the end (ie. weekly fitness plan, product recommendations)
- any topics to avoid (ie. politics, things outside your business)
- any needed background information
3. Try the conversation
Start having the conversation yourself with the AI. Based on the experience, you can iterate on the initial prompt.
For example, I once felt to AI was asking too many questions before getting to the recommendations.
You can even create several prototype threads to try varying approaches.
I started a conversation with my AI fitness coach that I “trained” with my initial prompt
4. Present your product idea internally
Next, you can present your idea with a concrete prototype. It’s easier to elicit feedback and excitement with a tangible prototype compared to just pie-in-the-sky ideas.
5. Validate the prototype with stakeholders
At this point, validating your thinking about conversational AI using your prototype is essential.
Ask them to try the prototype and share their feelings about the experience. I’ve noticed people respond in different and sometimes unexpected ways to conversations with AI’s.
You can also share only your initial prompt with relevant internal stakeholders to try. With ChatGPT, they can start having their conversation with the chatbot you’ve instructed.
6. Iterate on the prototype and mitigate any risks
Based on your feedback, you can iterate on your initial prototype. You might want to tweak some conversation details or pivot on the overall concept.
Prototyping is also an apt opportunity to expose any risks related to the legal, privacy, safety, or ethical dimensions of the AI experience. It’s easier to engage legal and other stakeholders with a concrete prototype.
7. Start building a bespoke solution
If the prototype brings value, start looking into building a bespoke solution using your data. Once you’ve validated the overall idea with a prototype, securing resources for the much more costly unique AI implementation is less of an uphill battle.
If you have workshops about using gen-AI in your business this year, build a Conversational Prototype to move the ball further. It’s the fastest way to move from up in the clouds ideas to shipped experiences.
