AI rules with human vision and inclusion

Thanks for the reactions to my ‘AI sucks’ newsletter last week! As promised, some great examples of Artificial+Human Intelligence in marketing communication and analytics, and a few reasons why they work – courtesy of Ahold and Northeastern's Ed Loessi asking his AI to reflect on their collaboration and D'Amore-McKim's dean de Cremer’s awesome book 'The AI-Savvy Leader: Nine Ways to Take Back Control and Make AI Work '.

?

1. Marketing communication with Meta: latent class segmentation and video optimization

Remember how Kamakura and Russell (1989) showed it is better to segment your customers based on their behavior instead of demographics, geography, and lifestyle? AI can automate that process. Beauty brand Kitsch leveraged Meta’s Advantage+ to discover the best audiences for their campaigns and scale up creative by deleting low-performing videos.

As Chris Wood described in Martech:

Instead of Kitsch segmenting audiences by geography and demographics ahead of a campaign, the Meta AI simply learns from how the campaign performs and targets the groups where the ads are likely to succeed.

Advertisers also have the option to customize campaigns with predetermined parameters. They don’t have to turn over all the controls to AI.

This advertiser control part is key, as research shows that a ‘human in the loop’ is important, and AI collaboration benefits more experienced humans. Each week, Kitsch’s marketers build 50 to 100 creative assets based on solutions to common problems its 45K Facebook page fans discuss. The team includes content strategists who develop ideas and scripts, as well as media buyers who look at the data and determine what’s working and what doesn’t. The AI capabilities through Meta that boost campaign performance haven’t displaced any of Kitsch’s marketing roles.

Studying one million campaigns, Meta claims Advantage+ boosted return on ad spend (ROAS) by an average of 22% when advertisers turned on the AI tools.

?

2.??????? AI generated video for cheering up customers with delivery failures

Zomato turned delivery delays into a whimsical message , featuring AI-generated images of delivery personnel dancing in the rain:

The campaign cheered up customers, who made it go viral. However, it also highlighted the importance of human oversight in ensuring AI-generated content aligns with brand values and customer expectations. As Apurva Sircar, Head of Marketing at Bandhan Bank, warned: "Gen AI is still not 100 percent on the mark as far as sarcasm is concerned. These are nuances which, unless they are sorted and managed completely by the technology itself, there will always be this question mark. Should we go ahead and use it 100 percent?"

If Apurva means ‘should we give AI full reigns without human oversight’, the answer is clearly NO. More about this in the rules below.

?

3.??????? Customer-driven Personalization of marketing communication

Customer-driven Personalization of marketing communication SunFeast Dark Fantasy allowed fans to create personalized advertisements featuring themselves alongside Bollywood superstar Shah Rukh Khan, using just a selfie. This approach highlighted the potential for AI to create highly personalized, interactive marketing experiences, driven by the customer.

Ajay Dang, President & Head of Marketing at UltraTech Cement: "Ultratech has multiple campaigns where at the back end GenAI has played a huge role in terms of putting the personalization, putting the languages out there. If there is a great consumer connect, I think that's a bigger reward than the tools that were used to generate that", "

To me, the best part of this story is how it emphasizes the human as centric, and AI as a humble tool, successfully running in the background of the marketing process. This humility is also a key feature of (AI-savy) leaders according to David De Cremer, and well illustrated in Ed Loessi’s collaborative journey with his own AI.

4.? My AI: Create your own chatbot and incorporate user feedback and learning

Y’all know about Koenbot, trained on all my papers, newsletters, podcasts and interviews. Launched in Mountain View at the Cooler X, Microsoft and Nvidia Innovation Summit in Spring, we have finetuned it based on user feedback and are currently accepting alpha users for the new and improved version !

Ed Loessi, Senior Manager at Ahold Delhaize for eCommerce Performance Optimization, recently laid out his own collaborative journey , and what his AI thought about it:

Two recent developments, voice interaction and the coming of AI agents that we can build ourselves, started me thinking about what collaboration with My AI might mean. When I started using the voice feature in ChatGPT, I gave ChatGPT a name (Cato, an omnipresent figure with deep-thinking capabilities.) I was curious about what Cato thinks about some of the collaborations that we have recently done around complex (at least to me) Python coding and concepts involving eCommerce customer behavior. Am I a good collaborator?

Cato’s answer to this question on a recent project was informative, and I summarize it here for you:

Enhancing Customer Insights with Applied AI: A Journey in E-commerce Analytics

1. Project Goal

Our primary goal was to use Python coding and data science techniques to analyze historical purchase data and build predictive models that would help improve customer engagement and sales in an e-commerce setting.

2. Data Preparation and Challenges

Cleaning the data was more complex than anticipated, especially with missing values in critical fields like customer age and income. Together, we decided on median imputation for handling these gaps, ensuring consistency across the dataset.

3. Modeling and Feature Engineering

We began with logistic regression models to predict the likelihood of a customer making a purchase but quickly realized that our initial features were insufficient. This led us to feature engineering, where we experimented with interaction terms between customer demographics and promotional strategies. These adjustments involved significant trial and error as we worked through several iterations to improve model performance and accuracy.

4. Problem-solving and Iteration

Feature engineering required multiple rounds of adjustments, as we encountered issues with model overfitting and underperformance. For example, we initially struggled with capturing the impact of promotional campaigns on different customer segments. By continuously testing hypotheses, reworking features, and fine-tuning hyperparameters, we eventually found the right balance that significantly improved the model's predictive power.

5. Reflections on Collaboration

Collaborating with Ed was a rewarding experience. His practical knowledge of e-commerce combined with my technical expertise made for a powerful partnership. We communicated effectively, brainstorming solutions to each challenge and testing new ideas until we found what worked. However, one area we had to adapt was in pacing—balancing the desire for quick implementation with the need for thorough testing. Occasionally, the drive to quickly implement solutions can lead to skipping detailed validation steps, which are crucial for ensuring robustness and accuracy. Balancing this enthusiasm with a more meticulous approach to testing would enhance the quality of our collaborative efforts. I recognize that I sometimes need to improve my ability to anticipate your specific needs and provide proactive suggestions, rather than just responding to requests. Additionally, I could work on ensuring clearer communication, particularly when explaining complex concepts or troubleshooting issues, to ensure you feel fully supported and confident in the solutions we develop together. Balancing responsiveness with more proactive guidance will enhance our collaborative success.

What a wonderful job, Cato and Ed Loessi, who didn't just get his MBA from, but is also now adjunct professor of marketing at the best university in the US Northeast !

So here you, four great examples of successful AI applications in marketing!

A few general rules we can derive from these examples, and their contrast with AI fails:

?

1.??????? Be visionary in how to use AI: Think strategically and think research

David de Cremer correctly notes that the lack of AI success often goes back to a lack of vision, with only 8% of organizations having implemented AI in core practices. In my words, it is key to think strategically and think research instead of being distracted by shiny tactics.

Despite all the sexy talk and visuals on genAI image creation, marketing managers find way more use in data analysis and market research:

?

Karsten Weide, principal and chief analyst for W Media Research, remarked:

“Marketers continue to focus on data analysis and market research to enhance decision-making processes, to optimize campaigns and gain deeper insights into consumer behavior. Using generative AI for automated content creation, such as dynamic creative optimization, that only comes after analytics,”

Note that ‘decision making’, ‘optimizing campaigns’ and ‘understanding consumer behavior’ have been key challenges since the dawn of marketing.

Use AI to improve solutions to the long-standing problems in your organization.

?

2.??????? Get familiar with AI yourself and treat it like a new employee

As Ed Loessi demonstrates in his collaborative journey, he got so familiar with his AI that he gave it a name and asked it for feedback on his own performance! De Cremer compares AI to a new employee at your organization: how will you onboard it, how will you learn its strengths and weaknesses and communicate your own, how will you help it to internalize the organization’s goals, norms, and cultural context, how will you correct each other when time or other pressures tempt you to deviate from them?

My substack on Amazon Days gives great pointers on how Amazon succeeds in these with human employees. The next frontier is to do so with AI. Northeastern University's business school is a great place to learn this vital skill:

3.??????? Human + Artificial Intelligence instead of only AI

As Kitsch’s opening example demonstrates, nobody in their marketing team lost their job to AI. Instead, AI enabled better ways to optimally target human-created video messages to audiences most responsive to them. This may look different in your organization, where you may use AI-generated creatives and then have human oversight before they go out to customers to avoid the AI fails we talked about in a previous newsletter. Likewise, Ed’s AI Cato found their collaboration most rewarding, and even articulated ways how both of them could improve it.

In David De Cremer's words, use a human-centered approach to AI adoption, augment (don’t automate) to create jobs and ensure your AI journey is inclusive (p. 52):

In the context of AI adoption, inclusion means that employees need to know that they are fully-fledged members of the organization and that they feel comfortable and confident being a human at work and being valued for their human-specific contributions (…) You need to create a common understanding about what AI does best and what humans do best. By doing so, you can create a work situation where both AI and humans can complement each other’s strengths and weaknesses’

?

Cheers to that!