The Lean AI Methodology

How Founders Can Build AI Products This Year

This Week: We had the pleasure of interviewing our friend Ash Fontana, Managing Partner at Zetta Venture Partners and author of The AI-First Company.

This book provides a detailed framework for companies, large and small, to easily start becoming more data-driven in their decisions. Buy the book here!
Let’s dig into it now...

Every founder, venture capitalist, and Fortune 500 is talking about artificial intelligence (and machine learning) today. We can’t go more than a few days without another tweet by Elon Musk or Sam Altman about the promises (and dangers) of AI.

That said, we are still only at the tip of the iceberg of what AI could become for business. At this point, software development has gone through many major iterations (waterfall, agile, etc.) and best practices have emerged. To date, the same can’t be said about artificial intelligence.

So, how should startups think about building AI products? What is the right approach for a startup founder who wants data-driven feedback loops (ship product, collect data, improve product)?

We interviewed Ash Fontana, author of The AI-First Company, to explore this concept. The book gets its title because startups should be putting AI “at the start of every conversation in your business… It affects what products you’re building, the people you’re hiring, where you’re allocating capital, what initiatives you’re going to take, and so on.”

What is Lean AI Methodology?

Lean software development has taken off because it aligns engineering goals with business goals. The key is quick iterations, frequent releases, and talking to customers throughout the entire process.

Ash shares a similar framework for building AI products:

“The key is to narrow the constraints so you can get to the point where you can ask one question, solve it with one model, run it over one dataset, and get one result.”

The key is to not try and do everything at once. “Don't try to turn everything into an AI product. Don’t start with multiple datasets or complex algorithms,” shares Ash. “The key is to get something small over the finish line and assess from there.”

Pick the Right Problem and Process

It’s crucial to solve the right problem. It can be tempting to start throwing around “artificial intelligence” and “machine learning” without truly understanding why it’s needed.

You want to solve a specific, useful, measurable problem. The problem statement should be simple to write out and address all of these key points.

Moreover, you have to pay particular attention to the scope of the problem. You want a narrow problem that your team can build a great process around.

For instance, it’s not just building a more complex model that will increase the scope of your project. You have to think about the data infrastructure, data cleaning, data labeling, building pipelines, validating results, and so on.

You want to standardize parts of your workflow (like the tools being used) and give people the freedom to try different approaches to achieve the business outcome.

The best way to do so, according to Ash, is to get a single task completed. That way you can evaluate the results with customer data and get their feedback, instead of continually improving a process that may not be what solves their need.

Use the Resources You Have Today

There is also a big misconception today about what is needed to build an AI product.

“Don’t start by hiring a CIO and 20 engineers,” said Ash. “That’s not the first step. The first step is giving your team the latitude to experiment and test.”

Your existing engineers have plenty of technical ability. There are some that are very comfortable working with data and statistics. Give them the freedom to try things out, to make these tasks part of stretch goals or continued education.

The key is to encourage training, online courses, and off-the-shelf solutions from vendors to develop your current team.

The wrong approach almost always hinges on investments of millions of dollars and months of time without a clear understanding of the returns.

Conclusion

There is hardly an industry or process that artificial intelligence won’t influence. From how packages are routed across the country to forecasting demand for a manufacturing plant and helping software developers make fewer errors, AI will shape our world.

The key is for startup founders to build a product and then think deeply about where AI could be used to create a sustained competitive advantage. 

“You don’t need to be an AI company from day one. But any existing company could become an AI-first company,” Ash stated.

We encourage you to check out Ash’s book and think about how you can implement AI-first thinking at your company to solve a single problem this year.