AWS for Industries
Generative AI for Retail: Key trends to watch in 2025
It’s that time of year when the pundits opine on what next year will bring for retailers. Rather than listing the top technology trends in retail as I did last year, I’m going to focus on generative AI and highlight some key trends to watch. If 2023 was the year when generative AI exploded on the scene, and 2024 was the year of experimentation, then 2025 will likely be the year the technology matures even further. We’re getting closer to Gartner’s “slope of enlightenment” but we’re not quite there yet.
If you’re tired of hearing about generative AI, I can commiserate. Generative AI reminds me of the band Nirvana. There were many grunge bands that paved the way, yet Nirvana got all the attention at the expense of many other great bands. But no one can deny that Nirvana, and generative AI, have had a major impact on our world.
So, here are three retail-specific use cases and three technologies to watch in 2025:
- Virtual Shopping Assistants (use case)
- Hyper-personalization (use case)
- Virtual Try-on (use case)
- AI Agents (technology)
- Domain-specific Foundation Models (technology)
- Computer Use (technology)
Virtual Shopping Assistants
The concept is simple. When a shopper is unsure of what to purchase, they can ask in-store associates for expert advice—at least in theory. But what if you’re shopping online? Enter the AI-powered virtual shopping assistant, versed in such diverse topics as plumbing, digital networks, and cold-weather fashion. What tools do I need to repair an in-ground sprinkler system? Which Wi-Fi router works best outdoors? And what should I look for in stylish ski gloves? Getting answers to these questions was the idea behind Rufus, the virtual shopping assistant launched by Amazon.
What makes Rufus particularly useful is the fact that it’s conversational, allowing a back-and-forth exchange with shoppers until they’re satisfied with the answers. Just like a human expert, the virtual shopping assistant asks questions to help understand the shopper’s needs and preferences. Some might consider this conversational search.
This certainly doesn’t displace traditional search, so retailers should start by modernizing their Search & Product Discovery solution, then decide whether Chatbots & Virtual Assistants make sense for them. These solutions will likely increase buyer confidence, leading to increased sales and possibly fewer returns.
Hyper-personalization
Personalization using machine learning has been in use for 25 years, when Amazon first started using collaborative filtering that predicts a given customer’s preferences based on the likes and dislikes of similar customers. The next wave combines machine learning with generative AI to create individualized experiences for shoppers. This includes hyper-personalizing marketing communications, search results, product detail pages, and even chatbot conversations.
Eventually, each web store session could be customized for individual shoppers, showing them products using appealing themes, tailored assortments, and curated offers based on their personal preferences. This sort of white-glove service that was once reserved for the wealthy may trickle down to average shoppers. Retailers should consider how best to hyper-personalize each and every shopper interaction, leveraging data such as past sales, product data, and third-party customer data.
Virtual Try-on
Online sales have been hampered by the shoppers’ lack of confidence, especially in areas like fashion. Without being able to conceptualize the product, they may be reluctant to purchase, or they may order multiple versions and return the unselected ones. Generative AI opens the possibility of visually depicting products in context so shoppers can virtually try-on products. This is accomplished by combining two images, say a person plus a sweater, or a chair plus a living room, so the shopper can evaluate the look.
AI models like Stable Diffusion and Amazon Titan Image Generator are used to intelligently combine the images, showing shoppers what to expect and increasing their confidence to purchase. Retailers selling apparel, fashion, accessories, furniture, or other products that benefit from visualizations should consider this capability.
AI Agents
Chatting with a chatbot or virtual assistant can be informative, but rarely is it action oriented. Agents, on the other hand, play a role in achieving a goal. They are typically autonomous and have tooling that helps them accomplish specific tasks. You can even think of agents as being part of your team, making contributions and moving things forward. Products like Amazon Bedrock Agents can use chain-of-thought reasoning to break down and solve complex problems. For example, you might have a pricing agent that can scrape competitors’ websites for prices, examine product margins, and make price recommendations within its defined rules.
Imagine that when you purchase forecasting software, for example, it comes with a forecast agent that can operate the software on your behalf, updating and distributing your forecast as needed. Retailers should be looking for tasks that could be automated with agents, yielding greater overall team productivity.
Domain-specific Foundation Models
Most foundation models (FMs), like large language models (LLMs), are trained on a corpus of public data so they have general knowledge, but it’s also possible to build a model from scratch to focus on a particular domain. The Amazon Science team created a retail-specific LLM for use by Rufus that has been trained on its vast product catalog, customer reviews, and other similar data with the aim of enhancing the shopping experience. The hope is that its focus allows it to be smaller and therefore cheaper to operate yet still yield superior outputs.
Of course, building an LLM is a huge endeavor, and likely not within reach of most retailers. So most will choose to fine-tune existing models with their own data. Retailers should consider this cost-effective approach to improve the output of generative AI.
Computer Use
While still in its early days, it’s possible to let a FM, such as Claude 3.5, take control of your computer and use it much the same way you would. For example, you could request that it create a purchase order for you, then it will “look” at your screen, take over your mouse, and fill out the form. Generative AI could also be used for regression testing, making sure the changes to your web store work as expected. From the consumer side, a shopper could ask to find and purchase the lowest priced Apple AirPod Pros, for example, letting it do the research and ultimately select a product for purchase.
As FMs are trained to use software, especially browsers, they will be able to automate some routine tasks, thereby freeing humans to spend more time on creative pursuits. It’s a little too early to adopt this technology today, but keep an eye on it in 2025 for general availability.
Visit AWS and NRF
To help inform your decisions regarding generative AI adoption in 2025, we invite you to visit the AWS booth at NRF in January 2025 to speak with experts from AWS, Amazon, and our partners, explore exciting generative AI demos, and hear about the latest ideas in this area in addition to many other innovations across smart store technology, digital commerce, retail operations, and more.
Additionally, here are some great resources to monitor generative AI progress in the retail industry:
