Artificial intelligence in retail analytics is a buzzy concept because it carries the promise that it will solve perennial problems like customer churn and onetime buyers, generally making the retail world a happier place. The truth is that AI is not a silver bullet that will get the job done all on its own, but if used properly it can empower retail organizations to tackle their challenges in ways that are smarter and more effective.

The amazing potential of marketing-style AI is its ability to discover customer segments that would be much more difficult and time-intensive for a human being to surface.

When not using AI, people tend to segment customers based on demographics: by gender, by age, by geographic location, etc. This is information that you more or less know about customers, and it can be used to target ads and marketing content. It can be meaningful in some ways, but it also has blind spots: Michael and Wes can have very similar demographic profiles (say, both men in their 30s who live in the Northeast) but have totally different taste in clothes (maybe Michael wears sporty clothes and Wes has a preppier look).

Segmenting by behavior is a much more effective way of grouping customers, but the challenge is that you have billions of data points: online shopping habits, online browsing history, visits to stores, calls to call centers, and on and on.

This is where AI analytics shines, by efficiently scouring all this data to identify pockets of users who are similar, based on the things that they do and the things that interest them.

AI can also identify segments based on the frequency of purchase and flag buyers for outreach at the right times when they show signs of slipping away. For example, Jenny hasn’t bought anything in 340 days, which might make retailers think that she’s at risk of churn. But AI analytics can easily tell those retailers that Jenny typically only purchases once a year, so actually, a win-back campaign wouldn’t make sense for her.

Meanwhile, Dasha usually buys something every month, but now it’s been two months since her last purchase, so she actually is at risk of churning. AI will take into account a customer’s habits and flag them for outreach accordingly — in Jenny’s case, after 14 months, while for Dasha it will be much sooner, after two or two-and-a-half months. If they veer off their typical schedule at the same time, the AI will group them together as “at-risk,” and allow retailers to reengage accordingly.

For another example of segments that AI can surface, which would be much harder for humans to detect on their own, let’s consider shopper types and personas.

Marketers and retailers have traditionally developed personas based on a combination of gut sense and character types we recognize from culture and society: “the soccer mom,” “the executive,” “the outdoorsman,” etc. But if you go into people’s closets and see what they’ve actually purchased, it often turns out that the soccer mom and the executive have a lot of similar purchases, or that two different outdoorsmen have fairly divergent shopping tastes. AI can get beyond the cardboard-cutout types that we’ve relied on for so long and let the data surface customer segments, so we end up with “shirts and skirts” groups, “heels and boots” groups, and so on.

AI can give a much clearer picture of who customers are, what they buy, and how often, which lets brands engage with them much more effectively. It can help you find more precise segments, discover pockets of users you didn’t know could be grouped together, and identify new opportunities. But to make the best use of these insights, it’s critical that your organization be positioned to act on them.

A common challenge for retailers is the gap between AI insights and operational efficacy. A company might invest in data scientists who have been using AI to find out what people’s top three product choices are, but the e-mail marketers haven’t heard about the findings due to silos in the organization. And it’s not just a problem of getting the e-mail marketers on board — if you have a critical segment that needs attention, you want every part of the business that customer could touch to be on board: the store, the call center, the web site, etc.

It’s also important to run experiments. AI will surface segments that would have been difficult to get without advanced tools, but to make the most of this benefit, you need to run tests to see what works best for your organization and your business circumstances. You can segment for customers who are fading versus engaged, or you can segment based on product clusters; try e-mail tests and promotion tests to see what gets the strongest response, etc. What’s going to drive the most value for your particular organization is variable, and it usually takes some experimentation to find out what works best.

Having your whole organization dialed into these experiments will maximize the benefit you can get from applying AI-generated insights. You’ll get more bang for your buck if the finance team knows about customers who are fading away so they can advise on win-back offers, or if the product team has a view on which products are no longer selling so they can adjust.

AI is a powerful tool that can dramatically improve your retail marketing efforts, but it’s only part of the equation. To be maximally effective, you have to put those insights to work across the organization. Think about it as a “super fuel” for a rocket ship — it’ll make you go faster and farther than ever before, but you still need the whole crew to run the ship.

Corey Pierson is chief executive officer of customer intelligence platform Custora.