If you are looking for the right place to experiment with artificial intelligence (AI) in your company, I’ll save you some time: ignore Alexa and Siri. They are very well-executed plays, but they will lead you straight into an AI abyss.

To be sure, AI, or more broadly, learning algorithms, are transforming industry after industry. This is happening relatively quietly, while the mania around robots, chatbots, and deep neural networks has reached fever pitch. And it’s leading an amazing number of companies astray.

So let me get this off my chest: natural language processing, chatbots, and humanoid robots are the most expensive, least value-added, most technically sketchy places any company can start to apply AI to their core operations.

OK. </rant>.

The core distinction you should watch for is the type of data any AI provider uses to create value.

There are two fundamental types of data — structured and unstructured. Structured data is numbers. Unstructured data is everything else — text, images, audio, video. While it is true most of the world’s data is unstructured (emails, blogs, photos, YouTube videos, etc.), these forms of data contain very weak signals to help you serve your customers better, determine what prices you should set, figure out where to store inventory, decide which products to produce, or help you optimize your logistics. Numbers largely hold the secrets to these decisions.

The leader among companies creating real value from AI, and the leader of the AI head fake, is Amazon. Alexa is cool. She can do amazing things with a simple command (unless you actually try to have a conversation with her). But… she’s a distraction. Alexa is not what Amazon is using to obliterate industry after industry. The key to Amazon’s dominance is the hundreds of embedded learning algorithms throughout most parts of the company’s operations.

Amazon gets the right products, in the right amounts, in the right places, with the right promotions, the right bundles, at the right price, for the right customers — every single time. And these decisions are driven by sophisticated learning algorithms trained largely on structured (numeric) data like transaction history, economic data, weather data, competitor behavior, events, demographics, psychographics, customer data, and more.

Unstructured data is the icing on the cake — but not the cake. Sure, each of these decisions can improve incrementally by natural language processing (customer sentiment) or image classification, but the vast majority of the value is derived from the structured data rather than the unstructured data.

So, where should you start with AI, if not by installing spooky robots to speak with your customers?

Probably demand forecasting. The first step to serve customers, create profits, reduce working capital, and general operational awesomeness is to predict accurately what products your customers want — and then how many, where, when, at what price, and bundled with what other products. AI is great at that.

Put your company data and outside data to work. Crunch it on a supercomputer with algorithms that detect patterns you cannot see, predict the future, and make recommendations that seem magical. Once you understand demand, you can shape it. Then once you shape demand, you can make sure you meet it. AI is also great at that.

Just, please, keep it real and ignore the head fakes.

Hey Siri, where should I start with AI?

Ask Noodle