AI knows us better than we know ourselves. Have you ever wondered how different services accurately predict what tv shows, or books, or songs you like? How retail sites find clothes or items that you want to buy even before you know they exist? It’s not magic. It’s just well-executed data analysis. Consumer data reveal preferences and predict likely future outcomes, and when it comes to making sales, there is nothing better than knowing what customers want to buy and putting it in front of them.

There is no question that AI personalization increases B2C sales. Personalization has already transformed the way consumer goods are sold, a change that will create $2.95 trillion for those able to effectively adopt new technology. In this post, however, I want to explain how personalization works, and why it is so effective at increasing sales in B2C and B2B settings. I’ll also sketch out a few unique ways that distributors can use data-driven personalization to increase multi-channel sales.

While countless businesses rely on personalization to drive sales and to please customers, a study of the music industry is especially demonstrative for how transformative AI can be. Music taste is extremely personal, unpredictable and individualized. For years, however, studios and record companies took an impersonal approach to music by peddling the same songs and albums to all of their customers.

Radio channels made some attempt to segment listeners with different DJs and shows. But for the most part, Billboard charts and services like iTunes just promoted the top songs. These approaches played the top song for any individual, but not the top song for each individual.

With AI, however, music vendors were finally able to make things personal. Modern streaming services and music vendors use AI to present customers with playlists and songs digitally curated just for them. Features like Spotify’s “Discover Weekly,” “Recommended for You,” and “Play It Again,” are all examples of data-driven personalization in action. Listeners are no longer willing to accept impersonal charts. They want personal service, and AI is allowing music vendors to provide just that.

AI engines track and analyze user data to generate personalized recommendations for each listener. Businesses may take different approaches to AI recommendations, but Spotify is particularly interesting, because it uses three different technologies to curate 100 million different playlists for each user every week (i.e., the Discover Weekly playlist). It is also highly successful.

The first of Spotify’s recommendation models is collaborative filtering, or other similar models. This essentially means that rather than analyzing songs, Spotify analyzes users to predict preferences. With collaborative filtering models, Spotify doesn’t have to make qualitative judgments about if Jackson Browne and the Eagles sound similar, if older listeners might like popular artists like Post Malone or Drake, or if country-rap collaborations like “Old Town Road” are a terrible idea.

In fact, to collaborative filtering models, bands and songs are just a bunch of 1s and 0s. Spotify’s system simply compares user profiles to recommend songs for each user. Most Eagles fans probably also listen to Jackson Browne, not Post Malone. Thus, if you listen to the Eagles, Spotify’s collaborative processing model is likely to put Jackson Browne on your Discover Weekly playlist and leave the Post Malone recommendations for someone else (This is a bit of a simplification of course: Spotify users listen to dozens of artists and hundreds of songs, so collaborative filtering recommendations are based on thousands of data points, not just two artists).

Another way that Spotify puts the right songs in front of the right people is with Natural Language Processing Models (NLPs). NLPs analyze words and textual data. For Spotify’s purposes, these models scrape tons of data from the web (news articles, social media posts, blog posts, etc.) to discover what words often appear in context with one another.

Returning to the previous example, Spotify would want to promote Jackson Browne if his name frequently turns up next to words like “good,” or “excellent,” or “nice.” When “Old Town Road” is linked to phrases like “trash” or “dumpster fire,” or “Oh God, my ears. Oh the horror, the horror,” Spotify will know to put it on fewer automated playlists.

Finally, Spotify also uses raw audio models to profile new songs. To do this, Spotify employs convolutional neural networks that analyze and compare the actual sonic quality of songs. When Spotify finds that songs, which you haven’t discovered yet, match the audio profiles of songs you frequently enjoy, then it will recommend those, too.

Spotify has more than 100 million users and a library of more than 30 million songs. Miraculously, Spotify makes playlists that feature the top 30 new songs (the top 0.0001% of songs) that each user is expected to like every single week. Even the biggest and most versatile of distributors don’t stock that many SKUs or service that many diverse customers.

Many distributors are still not using AI to match customers with new or niche products that they are likely to buy. They should be.

Distributors can use AI to satisfy customers and generate tons of revenue. Collaborative filtering models can effectively help distributors match similar customers and promote sales. Likewise, neural networks can find patterns in which products are often bought together, allowing distributors to promote those, too. NLPs can also be useful to distributors, as they can be used to source items for cross-sells and up-sells. Distributors can use these tools to easily translate Spotify’s “Recommended for You,” “Play It Again” and “Discover Weekly” into recommendations that would personalize the distribution shopping experience and increase sales (e.g., “Just You, “Re-order,” etc.).

What is more, distributors could apply these tools across all their channels. There is no reason that data-driven personalization must be restricted to digital platforms. Distributors can employ AI product recommendations across every channel to boost e-commerce sales, hone pitches from sales and customer service reps, and make mass marketing personal at last. Today, it’s not Jackson Browne or Post Malone that sells. With personalization, it’s both.