People in the marketing business know that the true power AI unlocks has more to do with improving strategies and targeting, and less to do with content creation. Pharmaceutical and consumer health giant Bayer has been partnering with AI-powered ad tech company Chalice for about a year, seeking ways to better target potential customers for some of its over-the-counter health brands. They exclusively revealed to me the smashing success they saw with a pilot program using display ads in Amazon’s demand-side platform to get new customers to purchase One A Day vitamins. While the campaign ran on Amazon, it wasn’t just to win over new e-commerce shoppers; many consumers still buy these products in physical stores. Bayer Vice President and Head of Media and Digital Platforms Maria Givens said the company is interested in powering the entirety of the consumer journey—something that Chalice can help with given their array of data from a variety of sources. Through the partnership, Chalice uses data to determine which people are the most likely to be receptive to One A Day online display ads, then its AI negotiates prices for the ads with Amazon’s DSP. Givens, Chalice cofounder and COO Ali Manning, and Bayer Senior Manager of Media Strategy Alexis Gossard talked me through the process. An excerpt from our conversation is later in this newsletter.Until next time.This is the published version of Forbes’ CMO newsletter, which offers the latest news for chief marketing officers and other messaging-focused leaders. Click here to get it delivered to your inbox every Wednesday. Artificial IntelligenceSmith Collection/Gado/Getty ImagesAs AI gets better, it’s appearing more and more online—and aside from a few blatantly obvious examples, many users can’t readily identify content that was computer generated. This week, Anthropic announced it will begin adding an invisible digital watermark to text and images created on its Claude platform, a move required by the EU’s AI law. For images, the move brings Anthropic in line with other AI and digital image editing companies, including OpenAI, Google, Meta, Microsoft and Adobe. But the proposal to digitally watermark AI-generated text and code created by Claude has many users up in arms. People commonly use AI to proofread human-written text, come up with ideas, make suggestions, or draft simple forms and basic text. The new watermark will stay on the text—regardless how much AI was involved in the final product. As it stands now, a short story generated entirely by AI will have the same marking as one written by a human and proofread by AI, calling its usefulness into question.Social MediaMeta was ordered to pay nearly $1 billion in fines by a New Mexico judge, who ruled the social media giant created a “public nuisance” over its handling of child safety on its platforms, and was a “significant” contributor to the state’s teen mental health crisis. The total—$942 million in all—comes from $567 million ordered by the judge for a fund to pay for awareness and prevention campaigns, as well as treat teens who were harmed by its platforms; plus $375 million previously ordered by a jury at the conclusion of a trial this spring.In his 68-page ruling, Judge Bryan Biedscheid compared Meta to a polluting factory. “[J]ust as noxious pollution produced by the factory can harm the common public right to reasonably clean air, the harmful effects of Meta’s platforms on children do not stay contained by its platforms and, instead, migrate to the internet as a whole and, perhaps most concerning, to the real world and create a common, societal burden on and harm to the affected children and their families and schools, as well as hospitals and law enforcement,” the ruling states.Meta communications chief Andy Stone wrote in a post on X that the company disagreed with the ruling and plans to appeal. Regardless of how the appeal goes, this may be just the beginning of Meta’s legal problems. On Monday, the U.S. 9th Circuit Court of Appeals ruled against Meta and several other social media companies in their bid to dismiss thousands of lawsuits that accused the platforms of being addictive to young people. The social media platforms—which also included Snap and the owners of TikTok, YouTube and Roblox—argued existing law protects them from such lawsuits. The appellate court disagreed, ruling that the law only provides a liability defense, and said the cases could go forward.Meta is back in a California court today for the beginning of another trial in which 29 state attorneys general accused the company of engineering Facebook and Instagram to be addicting to children. The trial is expected to last seven weeks, Reuters reports. On MessageExclusive: How AI Grew ROI For One A Day Ads By 60%From left, Bayer Vice President and Head of Media and Digital Platforms Maria Givens, Bayer Senior Manager of Media Strategy Alexis Gossard, and Chalice cofounder and COO Ali Manning.Bayer, ChaliceAI-powered online display ad targeting has proven to be a success story for Bayer and its partnership with ad tech company Chalice. I talked to Bayer Vice President and Head of Media and Digital Platforms Maria Givens, Chalice cofounder and COO Ali Manning, and Bayer Senior Manager of Media Strategy Alexis Gossard about a recent pilot for One A Day brand targeting on Amazon’s demand-side platform. This conversation has been edited for length, clarity and continuity. What does the AI do? Where does it come into play, and how is it different from something that is more manual?Givens: It is an amalgamation of traditional business and data intelligence within a proprietary AI algorithm. You’re not just chasing AI for the sake of AI. You are leveraging macro intelligence and your own business intelligence. The AI algorithm is the empowerment layer, and it’s helping to accelerate the work that you do in order to drive the outcome that you want.Manning: We have our AI technology take panel sales data that we get from different trusted partners that lets us see who has been buying One A Day for a long time, who is a new-to-brand customer, and who isn’t yet a customer. We can model out who is most likely to become a new-to-brand customer based on that data. We use LLMs to bring different audience datasets together. The LLMs can read the context between different audience datasets. In traditional modeling, you might just say, ‘This tranche of users look like they’re most likely to be new-to-brand, so we’re going to buy those 10 million consumers.’ We’re scoring every individual for the audience context around them compared to the audience context around Bayer’s new-to-brand customers, then we score them on how likely they are to be a new-to-brand customer. Then we go price that in market. We say everyone is a potential new-to-brand customer for Bayer, and they all have different values depending on how likely they are to reach for One A Day on the shelf. You’re paying for how likely they are to [buy the brand].How long have you used this strategy, and what kind of results did you get?Gossard: We started this strategy in Q4 of last year. That was the pilot program, but it had so much success, this is part of our regimen and it continues to perform. Until the results don’t prove out, we will continue to keep this as part of our strategy.In the first quarter we ran this, and then every quarter thereafter, we’ve seen four times more effectiveness. It’s had a 60% higher incremental ROI compared to the other targeting we were doing at [Amazon’s demand-side platform]. Predictive audiences have become a part of what we’re doing to drive the business.What does Bayer have planned for the next year through this partnership based on what you have learned here?Givens: One: Scale the success that we’ve had on One A Day to more of our brands. Two: Work on [predicting] attrition. We’ve proven incrementality and new-to-brand [targeting]. I think attrition would be the next step. Three: Linking to creative personalization, and seeing if we can do that from an automated perspective as well once we’ve identified the [customer] value. Maybe there’s an insight around [customer behavior] depending on the value, and that will influence the creative that we serve.Gossard: Right now, this is only running in the [Amazon] DSP, so primarily in display as a media channel. But we did pilot and test a similar type of strategy with Chalice for CTV. That’s another way to scale and grow: Finding the right way to utilize this technology in different mediums outside of just display.What advice would you give to CMOs that are considering using AI to better target their advertising?Givens’ advice:Don’t chase after the AI. Stay true to the business problems that you are trying to solve. A lot of times, it’s really around what are the gaps and what are the challenges in your strategy? You’ve got to jump in and test and learn. You can’t wait for it to be perfect. It’s not going to be perfect. The things that didn’t work are actually important for us to continue to iterate and learn from—just as much as the successes. Don’t underestimate or outsource the human intelligence. AI can solve a lot of things. It cannot solve everything. It’s not going to come up with the hypotheses like this. Sure, it can ingest data sources, but we’re human beings who have been living and breathing this business for so long. We’re in rooms, and the best ideas have come when we’re in rooms together, trying to understand what’s the next challenge that needs to be solved.Bring in partners that you believe understand your business outcomes [and] are willing to co-create for you. The questions we have, the business challenges that we have, are not off-the-shelf solutions. This is all custom work. We’re looking for partners who are ready and willing to be interoperable with our data and are really in a lab trying to create net new with us. Find the partners who want to co-create with you in a brand-safe, transparent way.Manning’s advice:If I was a CMO vetting an AI company, [I would] say, ‘Where’s your data coming from?’ If they can’t answer that very specifically and straightforwardly, and give you access to audit that data, they are selling something that's either not serving your needs directly or might not even truly be AI modeling. The most important part of any AI technology is the humans who give it instructions. Second is the data that fuels it. What’s important here is you’re able to find a data source that [you] trust. It’s not perfect, but it’s transparent. So many AI companies today are like, ‘Just hand us over your data, and we’ll give you a result.’ [If you do that,] your data’s empowering their technology. They’re the ones truly benefiting from it, and it could probably benefit your competitors. But if you work with a partner who’s transparent, auditable and works with your business goals, it becomes your proprietary competitive advantage.Strategies + AdviceA big part of BET President Louis Carr’s philosophy is serving customers—not making them wait. Here are ways he’s turned that customer-first approach into results.We all have first impressions of people, but they may be unfairly influenced by our past experiences. Here’s how to push through the fog and get an evidence-based first look at others.QuizWhich of the following people who was known as a child is now older and releasing his own music?A. Ryan Kaji of “Ryan’s World”B. Park Geonroung from Pinkfong’s “Baby Shark” videoC. Spencer Elden from the cover of Nirvana’s “Nevermind”Angus T. Jones of “Two and a Half Men”See if you got it right here.