Mars’ Speed of Scale: How AI-Fueled Image Recognition Is Increasing Conversions
Ask 50 people how a product image should best display on a website, and get 50 different answers.
There may be no other segment within consumer goods that’s more prone to interpretation than product design. We all carry our own biases of what’s “good,” inherited from personal experiences and preferences; try as they might, even the most experienced and discerning marketing exec isn’t immune to their detrimental impact.
And while it was once upon a time acceptable to display just a handful of “good” packaging views on an e-commerce site, the bar has been raised to much more sophisticated levels, Roman Vorobiev, global director of design and artwork management of Mars Inc., tells CGT, in part due to growing influence of Instagram and other social media platforms.
On these platforms, packaging isn’t merely highlighted in an elevated way, it’s also displayed in a manner that showcases its relevance for the audience within that channel.
For a global company like Mars, this means serving as channel experts in order to bring a certain level of intelligence to the petcare market for such retail customers as Chewy and Amazon.
“You need to understand the field deeper beyond [your retail customers],” says Vorobiev. “In certain spaces you need to know your category even better than they do, and so be able to have the proper conversation [about] how to develop that space, and how to further improve their customers’ experience in that space.”
Upping the Game
As e-commerce experienced its Great Acceleration during the pandemic, Mars sought to reduce brand identity dilution at the crucial digital purchase touchpoint. Design executions were being collected globally, which in part led to an inconsistent experience for digital and e-commerce content because of an absence of a structured decision making approach to the design, says Vorobiev.
This often led to teams making arbitrary decisions based on what they liked vs. having the data to demonstrate what’s effective.
“Yes, there is A/B testing,” he says. “But it's always difficult to extract a specific design component of it, whether it was a price promotion that was attracting people [or] something else. Everything was a little bit one-off.”
Once the sprint toward a smarter approach to retail content began, the need for better measurement naturally accompanied it. Mars partnered with Vizit for use of its artificial-intelligence-powered image analytics software with the goal of scaling its efforts to develop a retail content assessment methodology.
It was a new approach for Vorobiev’s team. While they previously leveraged AI tools to assess video content and help analyze sentiment in ratings and reviews, the same couldn’t be said for images. What’s more, their previous tools lacked ease of use, he says.
The new technology is designed to remove the bias that can historically accompany the retail content process, beginning with providing clarity of measurement even before receiving a design brief, says Vorobiev. Now, Mars can identify a problem with an existing image, including how it could perform with a specific retailer, to help locate opportunities and potential obstacles before even getting to the design concept.
“When you look at an entire category, you start noticing other things that you probably missed,” he notes. “And so through the exploration process, it helps to move forward more decisions that helped to advance the effectiveness of the final work.”
Mars’ design teams have uncovered great value after two years of use, Vorobiev says, which in turn has led to solid confidence with the marketing teams on the insights it provides for such things as choosing the proper dog breed within a certain market.
This isn’t to say that traditional validation tools will disappear, he notes. “But we believe by using AI on the way towards that final validation will help us to get better [and] less biased output. It's more focused on the pet parent’s experience, rather than focused on our internal bias.”
While Vorobiev says it’s difficult to quantify the increased sales given the rapid petcare category growth, the company has recorded a relation between image effectiveness and conversion. For one set of SKUs on Amazon, they identified a correlation between the effectiveness of the content with a difference in conversions of up to 30 percent.
And while there were initial skeptics — Vorobiev acknowledges himself as one such cynic — Mars has built confidence through its ranks from not only competitive analysis but also by becoming more cognizant of identifying their own blind spots.
For example, premium cat food brand Sheba had been highlighting the food experience in almost a high-end restaurant manner (think silver spoons and soothing greenery). But upon scanning the images with the technology, they discovered that it would be more effective to include a cat in the frame.
It was one of those moments that seem so obvious in hindsight, says Vorobiev, but can be very common in working with digital retail content.
Looking Ahead
While the speed of scale that AI can provide within the process can’t be underestimated, Vorobiev notes that success still hinges upon having the right people to process these learnings. Building internal groups to serve as practitioners and advocates for the technology are critical for success.
As part of this, Vorobiev, who has been with Mars since 2008 in a variety of cross-regional roles, has helped establish what’s almost a social network within the company — a practitioners’ community that spans functions and segments to help connect these learnings and develop a common language for them.
For other consumer goods companies seeking to obtain a better understanding of the impact of images on their conversions, Vorobiev also advises taking an outside-in perspective. This includes looking beyond one’s own content and into the entire category across the globe for a deeper understanding of a particular region’s nuances. In petcare, even the number of dogs appearing on a bag can impact performance in one area vs. another.
“Be open to learning something new — and don't dismiss the learnings, because sometimes you might find out things [that] might be surprising and might even be controversial,” he says. “Try to understand and analyze instead of just rejecting them and saying, ‘This is not how we do things.’ Try to understand why AI is judging certain things this way, because it may be the opportunity to learn something new that you missed in your bias and noise.”
