Since OpenAI debuted ChatGPT a little over three months ago, “generative AI” has become a hot topic and no industry has been left untouched — least of all retail.
“The advancements in just three months feel like they should have taken 10 years,” said Darren Hill, Chief Digital Office at BrandX, which owns the Bon-Ton brand, among others, in an interview with Retail TouchPoints. “When you compare how quickly this thing is going to how quickly everybody got on the internet or how quickly everybody got an iPhone, it’s night and day. I actually think generative AI is going to be bigger than the internet or smartphones in ecommerce.”
All this excitement is generating a disconcerting combination of fear and urgency among retailers, but according to Hill that’s no reason to sit it out. “You’ve got to get in there,” he urged. “Otherwise you’re like a championship horse racer who’s ignoring the car for as long as possible. Being scared doesn’t help anybody, it just means your competitor’s outdoing you.”
But how can retail professionals separate the hype from reality? Given how quickly generative AI has gained widespread attention , its initial hype cycle may already be giving way to “a second smaller wave of the things that actually work,” according to Brian Hennessey, Co-founder and CEO of the AI-powered product information management (PIM) solution Talkoot in an interview with Retail TouchPoints. As the trial-and-error of generative AI plays out live, some very real, tangible use cases for retailers are beginning to emerge, particularly when it comes to the content that forms the meat of any digital commerce operation.
“My advocacy to my team always has been that a lot of these things — 3D, VR, AR — are really cool, but they might die on the vine, and they’re not typically applicable to retailers right away except in small ways,” said Rohan Deuskar, Founder and CEO of the visual merchandising solution Stylitics in an interview with Retail TouchPoints. “[Generative AI] is different. How do you get somebody to buy the $300 cashmere sweater versus the $50 sweater? The descriptions do matter, content matters, words matter, and this new AI is incredibly good at it.”
And a whole host of brands and retailers — including Stitch Fix, Adidas, Hershey’s, Coca-Cola and Henry Rose — are already putting this new tool to work across:
- Product information management, in particular developing product descriptions;
- Search engine optimization; and
- A range of other copy-based tasks including marketing and advertising, customer service emails and social media posts (not to mention what happens when you bring AI-generated imagery into the mix).
Product Descriptions: The Job Copywriters Don’t Want
In any job there is some element of drudgery, and for ecommerce professionals feeding the always-hungry beast of the internet with product copy certainly tops the list. As ecommerce has become a larger piece of all brands’ businesses, that job has ballooned.
Before launching Talkoot, Hennessey was the Global Writing Director at Adidas in the time before generative AI. “Adidas sells probably 800 different types of white socks — one with a little blue stripe, one that is ankle height, one that is slightly higher, one that has more cushion,” Hennessey said. “You would save that for like a Friday when you were just going to bust through that horrible, horrible job of writing 500 pieces of sock copy. It was absolutely mind-numbing. And that is the kind of stuff that AI is great at — the bullet points, SEO-optimized titles, everything like that. What that does then is it elevates the writers out of the trenches and they’re able to do more strategic work, writing for your high-profile launch products for example.”
Not only must brands have detailed product copy for their own sites (already a monumental task), but many brands also need to adapt that text to meet the requirements of other platforms where they sell such as Amazon.
BrandX’s Hill has trialed generative AI for just that purpose and noted that the improvements in efficiency and speed are incredible. In one instance he was able to generate the same number of product descriptions that it would have taken a writer a week to turn out in under 10 minutes. When generating copy at that scale, however, he cautioned that things can “get a bit repetitive,” especially if you’re working on descriptions in a category with very little difference between the items — like sunglasses or, for that matter, socks.
“[Generative AI] works best for products where you’re not trying to change behavior, you’re just confirming that yes, these are the kinds of shirts that you’ve always worn and love, or for something like trail-running shoes where the behavior and the need is already identified,” said Hennessey. “When you’re trying to change behavior and [introduce] something new that will revolutionize [a category or experience], that absolutely needs a writer to help change minds and get people excited.”
Most companies don’t have the staff to create product descriptions at the scale and speed now required for modern commerce. Recognizing this, Shopify recently introduced a tool to allow its merchants to create AI-generated product descriptions, and a number of solutions already in market, like Talkoot, are continuously working with the latest generative AI tech to evolve their offerings as well.
As one would expect Amazon sellers, and the solution providers that support them, are also quickly jumping in, particularly because Amazon listings are so formulaic while at the same time being absolutely essential to success on that platform. Jungle Scout just announced an integration with Open AI to create Amazon listings instantly, while a number of other solutions exist solely for that purpose.
Meanwhile, other more technologically advanced merchants are doing it in-house. Stitch Fix is leveraging GPT-3 (the large-language AI model that powers ChatGPT) for a range of text generation tasks, including product descriptions, and has seen “unbeatable time savings as well as excellent scalability without sacrificing the quality of descriptions,” said Tianlin Duan, a Data Scientist at Stitch Fix in a recent blog.
AI can now create 10,000 product descriptions in 30 minutes at Stitch Fix. However, it’s important to note that Stitch Fix hasn’t simply plugged in GPT-3 and let it run. The retailer is taking what it calls an “expert-in-the-loop” approach, wherein copywriters and merchandisers are still heavily involved in reviewing and editing the “algo-generated” content. The change for them is that they now work off this generated copy instead of writing new content from scratch. “This human expert-in-the-loop approach allows us to leverage the creativity and efficiency of generative AI while still maintaining human oversight,” said Duan.
Playing the SEO Game
One trickle-down effect of having more (and more detailed) product copy is the positive impact it can have on search rankings. “If you’re working with the same product as a lot of other people, it’s horrible for SEO because Google immediately knows you have the same product as everybody else, and you get no credit for it,” said Hill. “Your product descriptions on a lot of stuff don’t need to be the best, they just need to be different, and that’s where [generative AI] is super powerful.”
Using AI to generate variations in product copy and to update that copy regularly can have a huge impact in beating out the competition online. In fact, new research from BrightEdge found that while only 10% of enterprise marketers use AI today, 58% plan to use it for SEO and content generation in 2023.
This enthusiasm is understandable given that both Amazon and Google rank recent or “fresh” content higher, something that Hershey’s has capitalized on with the help of Talkoot. “You buy Hershey’s for different reasons during Easter than you do in the summer or around Christmas or Halloween,” explained Hennessey. “And for each different package size you have different reasons to use it — you don’t buy a 10-pound bag of Hershey’s Kisses for your own use, you’re probably going to use it for baking, and you bake different things in the summer than you do around Christmas.
“Just updating your content every quarter will help you rank higher to begin with, but then if you actually match search intent — so [that] if Google recognizes that people are searching for holiday cookie recipes right now and you launch fresh content that says how great your product is for holiday cookies, that will help you rank higher in both Amazon and Google,” Hennessey added. “It’s really about creating that up-to-the-minute, fresh content all the time, which was way too expensive before.”
Email, Social Media and Beyond
While product descriptions are a practical, scalable application for generative AI at retail, Lori Schafer, CEO of the product experience management (PXM) solution Digital Wave Technology, said she doesn’t think it’s the “most compelling” use case. On a recent webinar with Coresight Research she pointed to other, more advanced uses that she finds even more exciting, such as automating the tagging of product “attributes” (size, color, shape, style, etc.), which is vitally important to a website’s CX, but “very laborious” when done manually.
Another area ripe for AI automation is the creation of product copy across the entire customer journey. “Every product has a story, and you’ve got to match that to the consumers’ lifestyle so the consumer can imagine how that product’s going to be a part of their life,” said Schafer. “It’s all about giving the customers that great experience [on every channel]. On a website you need a whole lot of content — romance copy, descriptions, videos, images — but you go to TikTok and you need one video, you go to Instagram and you need a picture and a statement. This [technology] can automatically determine which parts of that product story go to which of the various digital commerce channels.”
In fact, before it started using GPT-3 for its product descriptions, Stitch Fix first honed the tool by using it to write ad headlines rather than having copywriters manually create new headlines for every advertising asset. Michelle Pfeiffer’s emerging ecommerce fragrance brand Henry Rose has done the same in partnership with global ad tech agency Constellation, leveraging the technology to deploy hundreds of personalized iterations of creative assets based on demographics and geographies across Meta, TikTok, Google and programmatic. The results have been impressive — the Constellation-generated ads brought in 83.2% more customers and increased website visits by 73.1%, with a 12.6% improvement in return on ad spend (ROAS).
And as McKinsey recently pointed out, because “striking gold” on social can be a numbers game, the more content you produce, the better your chances of hitting the mark. Not only can AI-generated content save time, but because AI can recognize patterns and trends on social faster than a human, it can more readily create new content that does just that.
Hill also pointed to more mundane content that, while less exciting, is vitally important to a business, such as privacy policies, triggered emails to update customers on their orders or even to suss out fraud. To be sure, the vast potential of generative AI can be overwhelming, so at the outset Jill Standish of Accenture recommends that retailers start by using AI “around the big chunks of where your working capital goes.”
“The last thing a retailer needs right now is to gamble on customer service-related issues with technology that is yet to be quite proven,” Standish said in an interview with Retail TouchPoints. “Use it for things that make [your business] more efficient and take cost out. That’s where it’s going to be an absolute necessity, and then play around with some of these conversational AI features. The last thing [any retailer] needs right now is to cut costs and [in the process] cut things where the consumer is going to notice and that will impact customer service.”