Introduction

GenAI & LLM have become the latest hot topics, and they are widely discussed. Nowadays, many people are using them in various ways. You might be among those few considering integrating them into your product or developing something new. It's true that LLMs have simplified AI product development. With a lower barrier from idea to product, everyone now has a fair chance at product development. However, challenges in product development persist. Just because it's easier to build a product doesn't mean you can automatically create the right product, and no AI can solve this issue for you.

The challenges include:

1. User adoption of the product.🏃🏻

2. Higher standards for product capabilities and user expectations.

3. Lower competitive advantage and no clear moat.

4. Standard challenges with LLMs like reliability and stability.

5. Increased risk exposure due to external, unstable dependencies.

While developing Dinnerfy.com using this new approach to product building, I realized these challenges and developed a 5-step framework to overcome them. This framework will not only aid in identifying the right AI ideas but also ensure that you have a moat for long-term sustainability.

TL;DR - Avoiding AI Wrappers

Many AI product developers are essentially creating AI Wrappers. While these may solve the right problem, they often lack a real moat, making them vulnerable to competition or feature replication by more comprehensive products.

To overcome these challenges, a dual parallel approach should be followed:

1. Quick Turnaround - Use the 5-step approach to identify the right problem and employ AI wrappers to build and test prototypes faster.

2. Long-term Moat - Incrementally, find a clear market niche and moat to develop in-house models and custom datasets that are difficult to replicate.

This strategy offers both speed to market and a sustainable long-term advantage.

The 5-Step Approach

Let's dive into the 5-step approach that will guide you to the right AI product faster. Speed is crucial. The quicker you navigate the ideation maze, the closer you get to the right product.

1. Leverage Domain Expertise

Instead of inventing a problem, explore issues you have experienced firsthand. If you're new to the industry (like me 😉), spend time with those who have lived the problem. Without these insights, you won't be building the right product because you won't understand its current workings. People rarely change existing habits, so your best approach is to find ways to shortcut existing habit loops by creating more efficient processes. For a successful AI product, you need:

✅ A well-defined problem scope with bounded steps.

✅ Digital interaction steps in the current process.

✅ Sufficient, consistent, clean data.

If you're unsure where to start, first remove your AI cap and seek natural motivation alignment. Engage with people in that domain (through online forums or in-person meetups/conferences) and see where your interests lead. Starting somewhere is better than waiting for the right moment.

2. Break down the problem

Once you've identified a potential domain problem, break it down into smaller steps. AI isn't a magic wand that can instantly solve complex issues. Decomposing a large problem into smaller, more manageable parts can provide clearer solutions. It's akin to addressing root causes rather than symptoms. Often, in our eagerness to solve a problem with a novel solution, we jump hoops and make assumptions that don't align with reality. The MOM test is an excellent framework for interviewing potential users and then reaffirming your understanding of the problem. Think like a scientist, keeping the problem space separate from the solution space. If the problem is too broad, narrow your scope. It's better to start small with a specific market segment while still considering a larger potential market. It's also beneficial to experience the problem firsthand through consulting. This can deepen your understanding of the issue and provide some revenue for bootstrapping.

3. Prioritize Features

With these broken-down problems, you might have already envisioned some features or solutions. The next step involves prioritizing these features since you cannot build everything at once. List all these features, then identify the following for each to verify the solution and gain some traction before attempting to build everything.

To prioritize these features, you'll need to:

1. Use the Kano model to categorize features as Must-Have, Performance, or Delighters. Maintain a balanced set of features to solve first.

2. Identify the Importance and Satisfaction with the current approach. Calculate the Gap, where:

   Gap = Importance - Current Satisfaction

   Prioritize features with the highest Gap first.

3. Rate the building effort as High, Medium, or Low, and use it as a tiebreaker.

4. Prototype and Test

With prioritized features, build a quick and basic version of your product. Don't focus too much on design or user experience initially. It's important to test these prototypes/concepts as soon as possible with customers to avoid analysis paralysis or over-optimizing the solution. Aim to solve the problem at least three times better than the current method. You can use the following approach to build it faster:

1. Use an LLM wrapper (or RAG retrieval over LLM). See if Lamatic.ai can help.

2. Employ a No-Code approach on top of the LLM to build custom workflows and UIs.

3. Build upon existing OpenSource projects.

Explore more tools in this newsletter edition. 👇🏻

5. Build

If your features resonate well with customers, it's time to properly build your product. This requires having the right data to train models or perform retrievals for augmented generation (RAGs) and implementing them efficiently in your app flow. When implementing any AI solution, consider AI safety, performance, reliability, and user feedback. All these points will ensure that you're not just checking off an AI requirement but building a cohesive, self-improving AI product.

Conclusion

AI, then GenAI, and ultimately AGI, are set to drastically change the technology landscape. They have certainly made development more productive, but they've also rendered many products obsolete. It's crucial to have a clear pathway that solves actual problems, with or without AI. AI can definitely improve solutions, but don't use it just to ride the AI wave. If you're currently working on an AI use case, I'd be happy to chat and assist in your strategy. Email me at [email protected].

🙏🏻 Thanks for reading,

Aman Sharma