Portfolio of Experiments: A Framework for AI Builders
Portfolio of Experiments: A Framework for AI Builders
How can AI founders build and run an efficient portfolio of experiments
Welcome to Infinite Curiosity, a weekly newsletter that explores the intersection of Artificial Intelligence and Startups. Tech enthusiasts across 200 countries have been reading what I write. Subscribe to this newsletter for free to receive it in your inbox every week:
Hello friends,
Running experiments is an integral part of life for early stage AI founders! I've been thinking about a useful framework that can be used to navigate this part of the journey.
A good analogy here is that of a financial manager running a power-law-adhering portfolio.
Most will fail in this portfolio.
Some will be okay.
And a small number of them will pay out big time.
Here's how AI founders can think about it as they build their portfolio:
1️⃣ Define your super specific customer: You're solving a specific problem for a specific person. Who's that person? Learn everything there is to know about that person.
2️⃣ Narrow down the product scope: Don't over-engineer things. Start with smaller AI models that already work. Make sure they don't require too much data and compute power.
3️⃣ Minimize initial investment: Run experiments that don't require millions of dollars and months of work. Find ways to access low-cost (or free) AI compute power. You need to prove/disprove your AI hypotheses by spending the least amount of resources possible.
4️⃣ Build your data advantage: Understand the data you have access to. Is it unique to you? Can anyone access it with some time and effort? And no, your Fortune 500 pilot customer telling you that you're the ONLY startup they will give the data to doesn't count.
5️⃣ Use existing AI tools: Don't reinvent the wheel on anything. Capitalize on available tools, datasets, APIs, and pre-trained models. These tools will accelerate your AI experimentation timeline.
6️⃣ Get access to AI talent: Having access to AI talent is a critical component here. You need to find ways to get great AI people to work on your product.
7️⃣ Embrace uncertainty: There are no certainties in the world of AI. Only probabilities! Building a product that a specific group of people love is like catching lightning in a bottle. Most AI experiments will fail and that's okay. Be prepared to iterate, refine, and validate your hypotheses continuously.
Efficient and effective AI experimentation is a great way to accelerate your path to product-market fit.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 friend who’s curious about AI: