Marketplace Briefing: Advertisers are not won over by Amazon’s AI-powered ad tools
This is the latest installment of the Marketplace Briefing, a weekly Modern Retail+ column about the ever-changing e-commerce marketplace landscape. More from the series →
Amazon has been pitching brands on its latest AI tools for advertising. But advertisers are mixed on whether the e-commerce giant’s AI features are smart enough yet to power their campaigns.
Companies have been scrambling to develop their own versions of generative artificial intelligence ever since ChatGPT burst onto the scene last year. Amazon, for its part, is using its technology to attract more advertising dollars and boost its lucrative ads business. In the past year, for example, Amazon has introduced AI-powered self-service tools that generate image, video and audio ads. The retail giant is expected to earn more than $58 billion in ad revenue this year, up from $47 billion last year, per eMarketer. But it remains to be seen how much Amazon’s AI tools will drive that growth.
As is often the case with generative AI, the results aren’t perfect. Multiple brands and agencies that spoke with Modern Retail described a variety of issues, from bizarre effects to inaccurate product sizing and orientation. These products are meant to make content generation easier, but many advertisers said it’s easier to handle the creation themselves. Many also said they weren’t yet willing to use the materials that Amazon’s tools generated for customer-facing campaigns. While they agreed Amazon’s AI-powered ad tools will almost certainly improve over time, and they expressed optimism around the company’s newest features, brands and marketers are quietly testing and learning in the meantime.
“It’s kind of a mixed bag in terms of results,” said Ryan Walker, vice president of operations at Momentum Commerce, an e-commerce agency that works with consumer brands like Crocs, Lego and Clorox. “Sometimes it’s great. Sometimes it’s not usable.”
In an interview with Modern Retail, Jay Richman, vice president of Amazon Ads product and technology, acknowledged that mistakes are part of the development process. “Our approach is to launch these tools in beta with some advertisers and then have them evolve quickly as the science and the underlying technology matures,” Richman said.
Testing and learning
Sources mostly flagged Amazon’s image generator tool, which the company released last year, when citing issues related to its self-service AI features. Most products on Amazon’s sprawling web store are displayed as standalone images on a white background. But the image generator lets brands place those products in a lifestyle setting — like a toaster oven in a kitchen, for example. The goal is to remove creative barriers and enable brands to produce images that drive better return on ad spend. On average, brands that used the image generator between October 2023 and June of this year saw nearly 5% more sales per advertiser, according to the company. Amazon’s Richman also said that adopters of the image generator are submitting 39% more campaigns, on average.
Amazon’s tools are intended to be easy to use. All brands have to do is select a product image, and the AI is supposed to handle the rest. Brands also have the option to add instructions that describe what they want the AI to create.
However, some agencies like Momentum have run into issues with the tool. Occasionally, Amazon’s AI generates erroneous effects. In one example Momentum shared with Modern Retail, Amazon’s AI created an advertisement for a protein bar that portrayed the product as part of a water bottle’s label. The image also included a random sunburst effect in the upper right-hand corner. In another example, from a seller who requested anonymity to preserve their relationship with Amazon, the AI produced an image of a cheese grater in a kitchen setting, with the product partially obscured by a cloud of smoke. It was unclear to either source how AI generated the results.
Context is something that Amazon’s AI image generator struggles with, according to Danielle Waller, global chief media officer at marketplace consultancy Podean, which works with brands like Under Armour, E.l.f. and Danone. She cited an example of a toy car made for children, which was pushed through Amazon’s AI image generator and led to an image of the toy car on an open freeway rather than, say, a playground or a backyard.
“The context just made no sense whatsoever,” Waller said. “If you really want your brand to have certain standards around brand imagery, AI just hasn’t gotten to it yet.”
Scott McIntosh, who sells a phone-slash-cup holder device known as Cell Phone Seat through Amazon, reported a similar experience. In a demonstration viewed by Modern Retail, McIntosh tested out multiple scenarios to see how accurately Amazon’s AI could depict his niche product in a lifestyle setting — in this case, a car.
In some of the tests, McIntosh utilized the prompt feature to write instructions, such as, “A Cell Phone Seat in the cup holder of a car” or “A cell phone mount in the cup holder of a car.” In all instances, the images had at least one error. Sometimes the device was placed in an incorrect location, like on top of the dashboard or near the steering wheel instead of inside the car’s cup holder. In one example, the device’s sizing was too small. Another example had the correct size, but it showed the cup holder facing in the wrong direction.
For independent business owners like McIntosh, the appeal of Amazon’s self-service ad tools lies in their affordability and ease of use. But the highly specific nature of his product makes tools like the image generator challenging to use. He said that he’s run into similar problems with other generative AI tools, including ChatGPT and TikTok’s Symphony Assistant, a tool released in May that uses AI to generate TikTok-ready video content, including ads.
“It could probably work for some products, but it’s very hard to showcase a product like mine, which is an electronic accessory that has to be exact,” McIntosh said. “I can’t just stick the product in any environment.”
Amazon’s AI image generator works best with stationary objects, according to Momentum’s Walker. To illustrate, he cited an example of an ad created with Amazon’s image generator tool for a fire pit brand, which shows the product on an outdoor patio at sunset. “That’s a perfect use case,” he said. Looking ahead to the holiday season, Walker said Momentum may plan to use the image generator to publish ads for about one-third of the firm’s clients.
When the AI goes awry, it makes more sense to start from scratch than to edit over the generated image, multiple sources told Modern Retail. “At least for us, it’s still faster and cheaper to leverage our in-house creative team for one hundred percent of our ads,” an anonymous seller told Modern Retail. “Where we found value is having AI do the first pass, just to give the designers something to work from or to react to.”
To some extent, this is an intentional use case for Amazon’s AI tools. “The creatives generated need to inspire what comes next,” Richman said. “The back-and-forth experience is something that the Creative Studio is fine-tuned for,” he added, referring to one of Amazon’s newest ad products, which brings all of the company’s ad tools in one place.
‘Bumps in the road’
Despite the hiccups, there’s still plenty of hype around AI ad tools. Most sources told Modern Retail they were optimistic about where the technology is headed.
“It’s still pretty early, so there’s inevitably going to be some bumps in the road,” said Bob Avellino, senior innovation and growth director at Tinuiti.
An anonymous seller agreed. “If you’re a small brand where you have zero in-house capabilities, this is a godsend,” they said.
Amazon is racing to improve and add to its raft of self-service AI tools. At Amazon’s annual advertising UnBoxed conference earlier this month, for example, the e-commerce giant unveiled one new ad product that will let brands create audio ads. In September, Amazon launched an AI-powered video ad generator.
A brand client of Tinuiti’s tested Amazon’s audio ad tool ahead of the product’s official launch, with positive results. “We looked at a 94% higher ‘Add to Cart’ rate than previous campaign averages, so it was definitely something that ended up being very successful for us.”
Amazon’s Richman said the biggest piece of feedback the company received from clients when it launched the image generator was the desire for more control over how products were represented in the advertisements. Customers wanted features like the ability to adjust sizing, orientation and so forth. Ultimately, such feedback is what led to new product releases like the Creative Studio, which will make it easier to edit AI-generated content, he said.
Of all Amazon’s announcements at UnBoxed, Podean’s Waller said she was especially excited for the Creative Studio. “The idea of being able to take a product image and scale up to a video is so enticing,” she said. “If the promise is realized, this would be a game changer.”
Marketplace news to know
- The European Union is reportedly planning on launching an investigation into Temu over illegal products being sold on its platform.
- At its latest earnings, eBay posted revenue and EPS that were generally in line with analysts. Its guidance, however, was less rosy than Wall Street wanted.
- Amazon was supposed to kill off every brick-and-mortar retailer. Retail Dive looked into the smaller shops that have been able to fend off the e-commerce competition.