Marketers are under pressure to find ways to incorporate AI across their marketing departments. It’s becoming something of a metaphor for CEOs to expect time savings from using AI and expect the benefits to be immediate.
Beauty business Cosnova, the world’s sixth-largest decorative cosmetics company with brands such as Essence and Catrice, was in a similar position, but Rebecca Schalber, senior manager of AI, said the key to successfully integrating AI into a business is to take your time and find what works, rather than rushing straight into it.
Cosnova’s AI team works in two streams. One is a general AI enablement stream that helps teams work with AI more “safely and responsibly,” and the other is an internal content stream that explores ways to run processes more efficiently.
Over the past year and a half, Cosnova has launched 15 separate pilots to see how generative AI can help teams work more efficiently. The company has overseen pilots in image generation, text generation, process automation, and even AI video, resulting in seven pilots that passed the evidence-based quality tests it applied.
“We’re always checking the technical feasibility. Is it good? Is it producing a product as good as it could be produced by hand? Is it viable for the organization? So, can we really do this in-house? And should we? And what about the process? Then, of course, we also consider brand fit. Is this what we really want to do for the brand?” Schalber told Marketing Week.
Sometimes the results were surprising. She recalls one pilot about image generation. There, fully synthetic photo shoots were found to be “just as expensive” as in-house produced photo shoots. Other pilots, such as one that used AI to reskin a model to make it feel more accurate to the region it was marketing to, were fired not because it didn’t work, but because the company decided it didn’t want to cross that ethical line.
Another hurdle to clear is around quality, with Schalber and his team finding that in some cases, generative AI tools are simply “destroying” the product. She admits that there are many success stories in fashion, where plain T-shirts and jumpers are relatively easy to create, but in decorative cosmetics, unique textures can make this more difficult.
“For example, I might create a nail polish with a rosy base and peach glitter,” she explains. “But the AI made this glitter silver or made it too big, so it ended up looking like specks instead of just a fine shimmer. This was where it always broke, and it was very frustrating for us.”
consistency and buy-in
Consistency is key when it comes to beauty. Customers want to know that the products they are purchasing will have the desired effect every time, so it’s important to Cosnova that everything they do with Gen AI reflects that reality. Schalber points out that, as a primarily digital brand, the content it puts out on sites like TikTok and Instagram needs to show its products “as real” and without surprises.
This is something that has been “ingrained” in the business since its early days, she noted, adding that the brand has always tried to avoid industry tricks like Photoshop and “before and after” model shoots to avoid “over-promising” product performance. The idea of a “digital twin” of each product, a digital asset that is a one-to-one replica of the product, was appealing to brands.
Working with digital agency Collective, the digital twin was one of seven successful initiatives for the brand. While most generated AI images tend to “break” with repeated iterations, the team was surprised to find that the digital twins were getting better and better.
“What surprised me was that the product managers overseeing our current workflow said it typically takes seven to 10 feedback loops before they can say an image resembles the actual product,” she says. “However, Collective was able to achieve 96% accuracy in the first loop, which is sometimes not achieved even after seven feedback loops.”
This is important because in a fast-moving field like beauty, where the company can change a large portion of its product assortment (typically around 50%) “on an annual basis,” Schalber wants the team to be able to ride trends and bring them to mass market early. Being able to review the amount of time it takes to take realistic product shots for social media and e-commerce can help you achieve that goal.
The team conducted tests to ensure the accuracy of digital twins and interviewed 2,000 people to see if they could tell the difference between traditional product shoots and those created using digital twins. Schalber said the team found that it was “more or less luck” whether users found the AI images, and that people preferred digital twins because they were often glossier.
There is no doubt that Cosnova has been thoroughly committed to its approach to embedding AI throughout its marketing department. However, this needed to be balanced with management’s expectations. Schalver credits the board’s decision to start experimenting and find efficiency gains quickly, which allowed the work to balance quick effects with long-term projects.
“We’ve basically balanced it with quick wins, where we can show that we can achieve the efficiencies that everyone wants today very quickly, while at other times we need to be a little more deliberate and spend more time,” Schalver says.
For example, digital twins were seen as long-term initiatives due to the challenges of getting the data right to be most effective, while other pilots were designed to run for six weeks and ultimately yield a yes or no answer. One was a trial of a new AI-powered video editing software that was found to reduce video editing time by 70%. Sure, it’s an amazing time saver, but the ultimate goal was what it freed up the team.
“What does this mean for a video editor in this role? Is she just shooting more video, or is she spending time developing more creative angles for video production, or is she spending time developing new skills for post-production? That’s just as important as the time saved.”
Data and decision making
One of the obstacles many companies face when incorporating AI throughout their business is that their internal data is not set up to allow AI to work effectively. This is an issue that Schalber has touched on in Marketing Week’s martech series, “Unpack the Stack,” and one that Schalber recognized as a challenge that businesses must meet in order to succeed.
Although she is careful not to publicly announce that the work is done. “This is a work in progress. To be honest, I think any company would say this is a work in progress, because no data is perfect,” she says. “Everyone I talk to, whether it’s a brand or an agency, says their data is never perfect, but what we had was a very high level of data readiness in the product data itself.”
This meant that companies had easy access to all the information AI tools needed to better understand their brand, including packaging, artwork, product claims, product benefits, product ingredients, and more.
And in areas where brands are “not so good yet,” they are building “knowledge graphs” to connect siled data and build new relationships around it.
“We don’t have to wait until the data is perfect. It never will be, but this technology allows us to learn from it,” she added.
And of course, companies are constantly reevaluating their tools to see if progress has been made. Schalber says her team has four members and they are always “listening” for new tools and opportunities, and “testing out” whether their models can perform a particular task better. She said most AI models “will never be as bad as they are today,” and that as technology improves, there will naturally be improvements. But what’s important to Cosnova is the “inflection point” when it comes to incorporating tools.
“We see little incremental improvements, but you never know when everything will be perfect. To really honor our brand values and brand promise, we want to only feature content that has this product precision so as not to mislead consumers,” she says.
She believes it’s important for brands that are still taking their first steps into AI, or are stuck on adoption, to ask tough questions about what they want to achieve with Gen AI. Think about things like efficiency, quality, and ideal use cases to come up with a framework to use.
“We were very clear that we didn’t want to use generational AI to replace humans,” Schalber says. “We want to use this to free people from tedious manual tasks and allow them to be more creative and strategic. That was the guideline we had from the beginning.”
Also consider what consumers want from your brand when it comes to AI. You might want to use it to increase efficiency and get more content out there, but if consumers don’t want to see AI content in any form, you’re simply “steering” in a direction that won’t pay off in the long run, she says.
For all the technology buzz around generational AI, the answers to how to get the most out of it tend to be the same.
“Lead with people, not technology,” she concludes.
