This article was written by DMikey Vu and Samantha Hamilton. The original article was published by Bain & Company. You can find the article here.
In retail, many executive teams have by now had a chance to try generative artificial intelligence tools. That experimentation has sometimes taken on a surreal quality—like the picture above, created by asking OpenAI’s DALL-E tool for a photo of a panda bear on a skateboard wearing sneakers in Times Square. However, the emerging uses for generative AI in retail are concrete indeed and will have far-reaching consequences.
First, a reminder of where we were on AI and what’s changed. Traditional AI solves specific problems or makes specific predictions; it relies on algorithms to learn patterns and structures in data and can apply that learning to new data. Generative AI also analyzes patterns in vast data sets, but the big difference is that it uses that analysis to generate original, new to the world content, be that text, images, or music (video shouldn’t be too far off).
Agile companies across industries are starting to innovate with this new technology. For instance, Coca-Cola quickly launched an AI platform that allows digital creatives to generate original art using iconic Coke brand assets such as its distinctive contoured bottles. The platform, built in collaboration with OpenAI and Bain & Company, draws on DALL-E and its text-input sister, GPT-4. For Coca-Cola, it’s just the beginning of a broader generative AI push.
As things stand, generative AI’s emerging use cases in retail can be loosely grouped into four categories: personalized marketing, customer engagement and service transformation, operations and productivity, and customer and industry insights.
Generative AI enables personalized marketing at a scale that wasn’t previously practical. Take an email campaign targeting many thousands of consumers: The technology can adapt the core message to each recipient, using language and incentives that should resonate based on their past preferences. For a grocery shopper who cares most about deals, it could draft an email emphasizing value and include a coupon; for someone who sees themselves as a gourmet, it could extol the origin of the food and include recipes.
Marketing professionals can use generative AI tools to boost their productivity—for instance, by using them to generate a first draft of a blog post or creative imagery (in any style—photorealistic, cartoon, abstract, etc.) that they can then review and refine. Once the content is finalized, generative AI can also create derivative assets (such as resized pictures for social media) much faster, freeing up time for higher-order tasks.
Generative AI can be woven into the fabric of a retailer’s website or app through individually tailored landing pages, product descriptions, and illustrations. Apparel retailers have an opportunity to help customers better visualize how an item of clothing would look on them, using generative AI to alter their own photos.
There are also disruptive possibilities that don’t fit easily into marketing categories right now, such as a recommended recipe list based on a photo of what’s in a customer’s refrigerator, or a “closet concierge” that recommends outfits from photos of their existing wardrobe and the nature of the occasion.
Compared with more basic chatbots that use templated responses, virtual assistants powered by generative AI offer a much better experience for shoppers when they need help or inspiration, not least because the latest technology is better able to contextualize interactions with information it has gathered previously.
Consider a customer looking for new running shoes as she dashes between meetings. Talking naturally to her device rather than typing, she asks the retailer’s AI assistant which pairs are best reviewed for a runner who clocks up to 15 kilometers a week. The assistant highlights options, putting at the top of the list a shoe that the customer once bought in a previous iteration. After getting accurate answers to her spoken follow-up questions about whether her size is in stock and how long delivery will take, she buys the familiar pair.
Tools like this will offer an attractive fix for the poor search experience on some retail websites, given that plugging in generative AI expertise via an API will be easier than upgrading in-house search infrastructure. AI will also enable multimodal search, in which shoppers will no longer be limited to searching with text and keywords, but will have other possible starting points, such as photos, voice, and video.
For customer service agents, generative AI can recommend a script to follow in a call, suggest targeted offers that might tempt the customer, and produce summaries of each conversation. Generative AI is already turbocharging social media interaction by suggesting a range of possible replies to customer queries and comments. These prompts can help social media specialists engage with more people. However, customers will catch on to companies that appear to engage personally but don’t follow through on the points they raise.
Retail executive teams might already be tapping the potential of generative AI in a personal capacity—by using OpenAI’s speech recognition system, Whisper, to automatically take meeting notes, for example. This is just the beginning of the operational gains across the organization, as generative AI provides staff with tools that can help them speed through commoditized tasks.
Take frontline knowledge management. Retail associates are inundated with training documentation upon joining. It’s hard to retain and use that information when needed—for instance, when a grocery employee is posted to the meat counter and must observe strict hygiene rules—and there are only so many times a new hire can bother managers for clarification. Generative AI chatbots can give prompts and instructions just when associates need them, in a conversational tone that increases confidence as well as productivity. The technology can also codify best practice from unstructured data obtained from top-performing stores and integrate that information into the chatbot.
Vendor management is another operational area that’s ripe for AI assistance. Today, many retailers handle thousands of vendor relationships through large specialist teams. Generative AI can ease that load. It can automate—or semiautomate—some vendor interactions by performing tasks such as crafting request for proposals (RFPs), summarizing meetings, and drafting follow-up emails.
Other back-office uses include generating text for job descriptions. The technology will also unearth enhancements retailers don’t even know about yet—for instance, by identifying scalable best practice from masses of internally generated information that hasn’t yet been fully analyzed or collected. This “capture and codify” playbook should serve retailers well. And once they capture the best practices, they can use generative AI to automatically create or augment materials for training and onboarding.
New ways of analyzing customer sentiment and loyalty are emerging. Generative AI can monitor the content and tone of customer interactions—a phone call, an online chat—and then assess in real time whether that shopper is likely to be happy or frustrated with the information they received and the way in which it was delivered. Measuring customer advocacy through a metric such as Net Promoter Score℠ then becomes a predictive rather than reactive process.
Generative AI can make more sense of structured feedback such as surveys, generating sharper insights at the level of individual stores as well as offering a clearer view of the big picture. Its ability to process overwhelming quantities of unstructured feedback, such as freeform commentary on social media, will be transformational. Its capacity to synthesize this feedback with key industry-, market- and category-level data will enable real-time adjustments in products and assortment to match changes in sentiment, while also generating ideas for new products and services.
Online marketplace operators can also turn to generative AI to standardize the way third-party sellers list their items online, thanks to uniform keywords and product descriptions generated automatically from product photos.
As they analyze the potential uses of generative AI, executive teams face an acute strategic planning challenge. In many cases, the seemingly futuristic applications that emerge from their brainstorming sessions won’t be that futuristic at all: They will be possible within months rather than a decade or two. The recent strides taken by the technology really are that big.
That immediate applicability of these AI advances—so different from the slower burn associated with other futuristic technologies, such as blockchain and augmented reality—means that a wait-and-see approach is particularly high risk for retailers. There’s a strong case for bold application, even if the path ahead is uncertain. A test-and-learn approach today can develop a repeatable process that can be deployed widely tomorrow. AI will be core to the retail industry, and those that merely pause to take stock might never unwind the head start granted to their rivals.