Redefining Media Measurement with AI Insights

(00:00) The AI companion is on, and if you haven’t tested it out, I highly recommend it. What I like most is how it catches you up on things you might have missed—like, “What was that reference?” or “What did they say about that?” Anyway, I want to start this conversation with Tameka, who’s been one of my favorite journalists, thought leaders, and innovators in marketing, tech, and culture. When we started planning this Insider series and getting guest speakers, Tameka was one of the first people I reached out to because I knew she’d bring a fascinating perspective. So, I’m very excited to hear her insights on AI and what she's focusing on this year. Tameka, I’ll let you take it from here to introduce yourself and share your latest interests in the AI space.

(00:51) Thank you, David, and thank you all for joining! David, I think it's also interesting that we kicked off the year talking about AI at CES in a session together. So, I’m Tameka Kee, an advertising futurist, which means I focus on how advertising is changing and impacting various industries. I’ve been in this field for most of my career, and my current role is as Deputy Managing Director of SIM, the Coalition for Innovative Media Measurement, which is part of the ARF—another acronym! I took on this role at the beginning of the year, and it’s a dream job because I’ve always been passionate about advertising. I’ve loved it since I was a kid—I’d even record commercials, which had my family thinking I was strange since we were supposed to be recording TV shows. So now, working with the ARF, which focuses on research and advancing the practice of advertising, feels like a natural fit for me.

(02:02) At SIM, we have monthly virtual meetings where we produce research focused on media measurement. For example, we recently created a guide for the multicurrency transition to help both the buy side (media owners and publishers) and the sell side (agencies) understand different audience measurement methodologies. If Comscore says you have 10 million users and VideoAmp says 12 million, we provide guidance on how to calibrate that data. We’ve also looked into how privacy regulations impact measurement providers, which is a big issue since most privacy discussions focus on targeting and activation but overlook measurement. SIM’s work is very research-oriented and quite nerdy, but it’s also highly rewarding.

(03:32) I spend my time thinking about AI, VR, AR, and other emerging tech and formats. I’ve put together a short presentation for today on how to leverage AI in 2024.

This topic is vast, but I hope to distill it into practical takeaways. I recently moderated a panel on AI and the future of work for the Black Policy Lab, an organization using tech and policy to improve Black lives.

In today’s session, I’ll share some key takeaways for thinking about AI in 2024. One is to take a structured approach to “test and learn.” Testing AI is essential, but we need to be clear and rigorous about what we’re testing.

I’ll also talk about developing your “will-dos” and “won’t-dos”—what you’ll definitely use AI for, what you might try it for, and what you prefer not to use it for. This might evolve over time, but setting boundaries from the start is helpful.

(06:39) AI efforts should be cross-departmental. It’s critical to stay informed about how other departments plan to use AI, which tools they’re using, and what effects AI may have across teams.

For example, Sports Illustrated’s recent issues, where AI was poorly managed and communicated, could have been avoided with better cross-departmental coordination. Also, consider AI’s impact on interpersonal processes; it’s not just about the technology but about how teams interact and collaborate. Lastly, address bias in AI. While we won’t eliminate it, we can tackle bias by seeking new data sources to create a more balanced perspective.

(09:02) David, you asked about any big players that are addressing bias well. Amazon is one company doing some positive work. They acquired a company that specializes in adding data that better represents lower-income Black consumers, helping to balance their broader data set. The encouraging part is that more companies are openly acknowledging the problem of bias, which is a crucial first step. But because data sets are so vast and complex, fixing bias is challenging, especially when it requires a close look at the sources we use.

(12:51) Another approach companies can take is to curate data from within—content employees have searched for, are interested in, or have chosen. This curated approach lets you use AI like OpenAI for querying within a controlled data set, allowing for more manageable biases from internal sources rather than from vast public data. This controlled process could be valuable, especially when reducing bias is a priority.

(14:13) That’s a great point, Karan. Including humans at each phase of AI processing, whether refining prompts or reviewing output, helps manage bias. There’s no single solution, but having humans involved at every stage ensures we keep asking the right questions and making necessary adjustments.

(15:20) Bias is also subjective; it depends on context. For example, from an advertising standpoint, bias might even be seen as targeting—since advertisers often aim for specific audience segments, which naturally skews data. It’s more about knowing what the data represents and ensuring that you’re clear on those boundaries.

(17:05) At SIM, we recently conducted a study on targeting that revealed how layering multiple data sets can skew results. For example, combining video viewership data with in-store purchase data and demographic data significantly reduces the audience size, often leaving a skewed subset of the original group. The study wasn’t about proving data was biased but instead about showing how different data sources impact audience targeting.

(18:45) “Unintended algorithmic cruelty” is an interesting concept here, where layering data creates unintended biases in targeting. Home Depot once ran into issues when they based campaign targeting on zip codes, inadvertently excluding certain socioeconomic groups. It’s a reminder that data needs careful handling and review to avoid unintentional discrimination.

(19:52) Streamlytics, for example, compensates consumers for their data and focuses on Black consumer data specifically. If you use Streamlytics, you’re intentionally using a data set that’s skewed toward Black households. This isn’t necessarily bad; it’s just important to understand and transparently communicate the data’s skew.

(22:17) Testing AI rigorously is essential. This slide shows an approach Emory University used for AI-driven ad creation. They tested AI’s ability to produce display ads, comparing human-created ads with AI-generated ads. In this case, AI-produced ads outperformed human ones in terms of interest and purchase intent. This kind of structured testing is critical to see where AI might be more efficient than human-led work.

(26:44) Next year, we’ll likely see more growth in AI for audio and voice, as voice synthesis improves. Already, tools like HeyGen allow users to create AI avatars from just a 30-second voice clip. Combined with AI-generated text, this can create full video introductions or even spokesperson-like content. 

(29:57) We recently conducted a study on testing AI for different ad scenarios. For example, AI-generated ads for car brands focusing on “ruggedness” and “luxury” scored higher than human-made ads in terms of audience interest and purchase intent. This targeted testing approach allows companies to identify specific areas where AI outperforms, freeing up human teams for more strategic tasks.

(32:02) Some challenges that marketers face with AI include creating effective prompts and revising AI-generated content, which can sometimes be more time-consuming than starting from scratch. Editing poor AI-generated copy, for instance, can take more effort than simply writing it yourself. It highlights the importance of strong prompts and high-quality tools.

(36:44) Prompt engineering is an art in itself, and certain team members may be naturally better at it. Developing a structured, strategic approach to prompting AI can make the technology significantly more effective. Prompting tools correctly to produce refined, useful results is critical, and learning this skill is essential as AI use continues to grow.

(40:36) Another strategy is to establish clear guidelines on when and how to use AI. For example, I’m comfortable using AI for writing job descriptions but cautious about using it to select candidates due to concerns about bias. AI is now integrated into many parts of marketing, but it’s important to define what we do and don’t want to use it for.

(45:45) We must also consider how AI affects team dynamics. For example, if AI allows someone to work faster, what happens with the time they save? Does it create more workload for others, or does it allow for more creative tasks? Using AI efficiently means being mindful of its impact on the team’s morale and workload distribution.

(49:26) Finally, as AI tools become more advanced, like with AI-only creative teams outperforming human-led teams in some cases, it prompts companies to examine internal processes. In one case, an AI-driven team outperformed a human team in pitching ideas, which encouraged the company to reevaluate how they structure their work. AI can expose gaps in our workflows, pushing us to improve our processes and collaborate more effectively across departments.

(54:03) We’re at the hour mark, so thank you, everyone, for a great discussion! Tameka, it’s been a pleasure to have you share your insights, and we’d love to have you back anytime. For those interested in Tameka’s work, connect with her on LinkedIn, follow her podcast *Tech and Soul*, or email her directly.