(Photo by Christopher John on Unsplash )

"Anyone can have an idea. Execution is what matters."

This is considered conventional wisdom in business, but is often forgotten when it comes to AI. And I don't just mean new-age genAI – this also applies to the predictive ML/AI we've been working on for years.

How do companies miss this? It's because they're on autopilot. Since AI falls under the tech umbrella, and building models requires writing code, they apply their standard technology measures:

Code is everything.

Code is the secret sauce.

Code is our competitive moat.

(Experienced software developers would soften those statements. And I would agree with them. But that's a story for another day.)

The problem is that the vast majority of code in ML/AI is not even close to being secret sauce. It's boilerplate. You can build models for different companies, in different industries, on vastly different datasets, yet the code will mostly look the same. Everyone is using the same toolkits – TensorFlow, Torch, and HuggingFace, to name a few – and the code they write is a twist on the examples in the documentation. This common ground is the very reason that industry-neutral automated machine learning (autoML) tools exist.

What makes a model yours is the training data behind it. Companies that have finally caught on now say:

Data is everything.

Data is the secret sauce.

Data is our competitive moat.

Yet even that is not entirely true. Your customers don't care how you built that AI model; they're interested in how the functionality and outputs – the predictions, the classifications, and all that – impact their work. They're paying you to make their life easier, and how you get there is rarely of concern. (For now we'll set aside discussions of data privacy and copyright infringement.) A competitor could build a model that provides roughly the same outputs as yours, and with roughly the same level of performance, using a completely different dataset.

That brings me back to my point:

Execution matters.

In AI you need to figure out when and how to embed a model into your products and workflows, actually train a suitable model, get it running in production, and then sell sell sell. You then need to get those products in front of customers who will pay you for the privilege of using them. That's how you make money with AI.

You can have the best code, the most proprietary dataset, and a model that performs to near-perfection … and yet, a competitor can still come along and eat your lunch because they did a better job on execution. This is a bitter pill to swallow. But it's the truth. And the sooner you accept it, the better your chances of beating the competition.

So as you rush to hire data scientists and data engineers, remember to round out that roster with product managers (to guide the vision), software engineering (to integrate the model into your products), IT operations (to get everything running in production), and sales/marketing (to convince customers to sign up and pay for what you've built). You'll also want to loop in risk management (to make sure you are placing smart bets), the legal team (to keep your activities on the straight and narrow), and PR (for the inevitable problems that will crop up).

That entire group, operating as a well-oiled machine? That is your moat in the AI world.