Where I think Retail AI is headed in 2027

I spent three days at Shoptalk Fall in Nashville, listening to and participating in conversations about AI agents in retail workflows, demand sensing and planning, customer loyalty through social media, and how broader economic and cultural shifts are reshaping consumer behavior. These conversations gave me a good sense of what retail executives are actually thinking about when it comes to AI.

My takeaway after three days is that the first phase of retail AI was more focused on giving stakeholders visibility into processes, fragmented data, and the world around their brands. This next chapter will be more about using that information to make better decisions. Information is now coming from every possible direction, so the challenge has shifted to figuring out what actually matters and what should be done about it.

Here are my guesses for what else comes next. 

#1 Retail brands need help making sense of the signals they’re receiving.

Signals are getting louder and less predictable. Fifteen years ago, unhappy customers would leave a bad review on a product page. Today, sleeves that run a little short can start a cultural conversation on TikTok. You could spend months planning a product launch only for it to fall flat, while a product that has been sitting on shelves for decades can suddenly find itself in a renaissance. 

It seems like there’s no real method to the madness. For example, Clinique’s Black Honey lipstick (also the one I wear daily) launched back in 1971 and has found a loyal audience in nearly every generation since. 

For the past decade-ish, the answer to deciphering these signals was to mine more data. Signals would be pulled from various repositories and passed on to an analyst, who would organize them into insights for leadership. The problem now is that culture is moving extremely fast, and this process of data analysis is far too slow. Data is also showing up in more formats and from more sources than traditional spreadsheets and BI tools were really built to handle.

To add to the problem, too much data can create analysis paralysis, especially in retail and fashion, where creativity and psychology go hand in hand when trying to reach customers. Retail teams also tend to work in a highly coordinated fashion, so the more time people spend collecting and analyzing data, the less time they have for the discussion, collaboration, and strategic thinking that are crucial to the job.

What seems like a good middle ground is designing a decision intelligence system that continually senses different signals and proposes actions or decisions based on the information available. It would also need to leave room for collaboration, discussion, and cross-functional work rather than trying to automate the entire decision-making process.

#2 A good recommendation can also just be saying no or not doing anything.

It is easy to say that a brand should be cognizant of customer preferences and respond to them to the best of its ability. But the truth is that what customers want is not always what is going to create meaningful value for the business. The next generation of AI systems will need to process the multiple factors feeding into a decision and explain exactly why something is or is not recommended. And for that, AI will need to start developing a POV.

For example, if a retail brand is considering signing a major wholesale partnership, the AI system should not simply answer whether the company is ready based on the information available. It should explain why the company may not be ready, what would need to change before the deal makes sense, and how far the business is from reaching those conditions.

Another concern that came up repeatedly in the meetings I had was that agents tend to gravitate toward recommending more products, more SKUs, more promotions, and just more activity. That looks pretty good in a demo because the system always appears to be doing something. But a recommendation to leave something as is, flat-out decline, or postpone a decision can actually be much more valuable. Restraint is part of the job. A good AI system should be able to tell you when to do something, but it should also be comfortable telling you when to leave something alone.

We may also need to rethink what we mean by a recommendation altogether. Is the agent going to do everything and leave you with the final approval, or is it going to synthesize the information and give you something you can process and take forward as a human? It seems like different companies have different definitions of what that means. But regardless, I do think there needs to be some industry-wide encouragement for agents to also propose less activity.

Sometimes saying “no” is a pretty good recommendation.

#3 Demand signals only matter in context.

Most companies have years of data that they feed into complex machine learning models to predict future outcomes. What does not always show up in the formula is the context surrounding those outcomes. A product may have sold well during an unusually warm season, after a promotion, or because it went viral on social media. Historical data can tell you what happened, but really understanding why requires looking beyond just the numbers. That kind of nuanced interpretation should not always require a data scientist. AI should be able to help.

AI-driven demand sensing was a very hot topic of conversation. It gives teams more evidence to understand whether interest will actually survive the journey to checkout. But the inherent value of those signals depends heavily on what you sell, so prioritizing the wrong ones can actually confuse your models even more. An executive from Spin Master pointed out that, during one global product launch, the company was able to learn from earlier events in the launch calendar and quickly pivot its strategy for the ones that followed.

Demand sensing is great, but I expect it to become much more category-specific. It could be by product line, business unit, or even down to sub-verticals. It really just depends on what kind of company you are. The timing and surrounding context of those signals will also matter and, ultimately, lead to much more precise demand models.

#4 It’s not about the influencer, it’s about the world they inhabit. 

One of my favorite things to do as a kid was go to the toy store with my mom, who would let me buy exactly one Barbie every month. That experience and the time I spent in the store felt magical to me and are definitely some of my core childhood memories. Sometimes I wonder if kids today have that same experience. Turns out, a lot of brands are asking a similar question and trying to recreate some version of that feeling digitally. Social and live commerce are attempting to bring the human part of the buyer’s journey back through a digital interface.

Social media creators are helping to bridge that gap. But unlike a decade ago, influencer marketing has evolved beyond large followings and obvious sponsored posts. Brands are paying more attention to smaller creators with loyal and consistently engaged audiences. They may have less reach overall, but they can often reach very specific customer groups quickly. Maybe a smaller audience makes the creator feel more approachable, or maybe the products they feature simply have a lower barrier to purchase. Either way, the relationship between the creator and the audience seems to matter more than follower count alone.

And it is not just who is making the content. The content itself is changing too. I remember when I was in high school, I used to watch try-on hauls and in-depth makeup tutorials where you immediately knew someone was selling something. Now, more natural positioning seems to be becoming popular. The idea is to place a product into the life of a creator and let it show up in a more nonchalant way.

The new formula seems to be understanding what kind of world a target customer wants to inhabit, finding creators who authentically represent that world, and then figuring out how a product can exist inside it without feeling forced.

#5 Merchants will finally have more time to be creative and curious.

For all the attention given to AI, I was happy to see that the merchant remained an integral part of the conversation. The real job of a merchant is to combine intuition, an understanding of the customer, and a sense of the brand to create an experience people respond to. This job requires someone to hold the relationships between products, customer expectations, and brand identity all at once. It is almost like conducting an orchestra.

The near-term wins with AI in merchandising are unglamorous. They include assembling information, reviewing spreadsheets, detecting anomalies, comparing scenarios, and tracking exceptions. There are many merchandising teams out there that need help, and I think people who are too focused on autonomous retail are missing out on some very easy wins. Retail AGI sounds ambitious, but merchandising is not there yet, and that is okay. The help many teams need right now is still much more practical.

The more exciting thing to think about is what merchants are going to do with that time back.

I have my guesses. I think many merchants entered the field because they are obsessed with nurturing the relationship between a brand and its customers. And I think having more time will allow them to immerse themselves in observation. They can spend more time in stores watching how customers interact with products. They can compare how ideas actually execute across different locations. Or they can spend more time crafting in-person experiences that help them understand the customer more deeply.

My optimistic take is that AI may actually allow merchants to return to their creative roots. If systems can take more of the administrative burden off their plates, merchants can spend more time doing what humans are particularly good at: building connections.

#6 Good leaders aren’t necessarily AI-native, but they know how to navigate uncharted territory

The “T” word is not transformation anymore. It’s trust.

Most companies already have some blend of AI-native and AI-curious people, and most know they want to implement some sort of AI strategy. Many are already experimenting with several AI tools, but only a minority seem to be seeing the ROI they expected, even as training and familiarity with AI have improved.

So what’s missing? I think it’s good leadership.

I’m thinking about all of the amazing leaders who helped their teams navigate the uncertainty that came with COVID. No one was an expert in organizational change management during a global pandemic, but a lot of leaders with the right set of soft skills were able to help their teams navigate an incredibly uncertain situation. These are the people we need to re-engage for the AI age.

Today, companies appear to be dealing with some version of the same issue. Some employees want to embrace AI as quickly as possible, while others remain skeptical. Bridging that gap requires a leader who can translate the technology into a vision people understand, create reasonable guardrails around it, and actually help the organization get from where it is today to where it wants to go. Basically, someone sensible who has a plan.

For that, you do not necessarily need a technical leader who is deeply immersed in AI. You need someone who understands AI well enough, has strong soft skills, and can execute a plan.

A relatively small part of AI transformation is vendor shopping. In fact, that is probably one of the easier parts. The more painful exercise is the organization and execution around it. Good leaders will take the time to understand their organization, coordinate and evaluate pilot results, set up workshops and training so tools are actually being used effectively, and stay open to adjusting the approach as more information comes in. The problem is that sometimes the person buying the tools and the person responsible for organizational change management are different people. When there is little synchronization between the two, the results are probably not going to live up to expectations.

Good leaders are in high demand right now. And while most companies like to think they already have the right one, this next phase is going to reveal pretty quickly whether they actually do. 

What this means 

Retail is the next industry up for its AI journey. There’s a lot of excitement, an underlying agreement that the customer still matters, and a real opportunity for domain experts to participate in crafting AI strategy. Parts of retail operations have already embraced machine learning pretty heavily, so for those functions, AI may feel like a natural evolution. But for the parts of the business that have historically relied more on human judgment and intuition, AI presents a different opportunity.

I’m convinced that there is a way forward where we build systems that retain the integrity of the industry while making the people inside it more creative.

‍