Building AI Products (Part 1): How to Pick the Right AI Use Case and Model

The 3P framework for evaluating AI use cases with examples of AI products that solve real customer problems

This post is co-authored with my colleague, Jamie White.

---

Dear subscribers,

Today, I want to share part 1 of a 3-part guide on building a generative AI product at your company.

Many companies are building generative AI products only to realize that nobody wants another AI chatbot.

Instead, the best AI products are improvements to your core customer experience. So in part 1, let’s cover:

1. The 3P framework for evaluating AI use cases

2. Examples of AI products that solve real customer problems

3. How to pick the right AI model and development framework

Part 2 will explain prompt engineering, retrieval augmented generation (RAG), fine-tuning, and pre-training. Finally, part 3 will focus on evaluation and deployment.

Let’s dive in.

---

The 3P framework for evaluating AI use cases

When building an AI product, start with the 3Ps —play, prototype, and protect.

Credit to Scott Belsky (CPO Adobe) for inspiring this framework

Play

I've always advocated for starting with the customer problem, but AI is moving so fast that you must first understand what’s possible.

So before brainstorming AI use cases, your team should spend time to:

1. Understand how AI works. Check out my curious beginner’s guide to LLMs.

2. Use AI tools daily in their job. See my favorite AI tools, prompts, and workflows.

3. Share learnings. Start a Slack channel for your company's AI experts to ramp up everyone else.

Encouraging your team to become “AI native” first will help them separate the hype from the real use cases.

Prototype

So, what is generative AI 10x better at than the alternatives? From my experience:

AI is 10x better at summarizing info, generating content, and extracting insights.

Keep the above in mind as you follow the steps below:

1. Map out your customer journey. Don’t start by asking: “What can we use AI for?” Instead, put the tech aside and focus on mapping your customer journey and the pain points for each step.

2. Find pain points that AI can solve 10x better. Look at each pain point and ask:

   1. Is this a deep or frequent customer pain?

   2. Will solving it significantly grow the business?

   3. Can AI solve it 10x better than non-AI solutions? Any use case that requires summarizing info, generating content, and extracting insights is a good bet.

   4. Do we have proprietary data or UX that gives us an edge over other AI products (e.g., just using ChatGPT)?

   5. Will our product get better or become redundant as AI models advance?

3. Prototype to validate demand. AI has made prototyping so easy that there’s no excuse for not validating your product with real customers as soon as possible. Here are three ways to do so: