David Leibowitz, General Manager of Biz Dev & Strategic Partnerships @ Microsoft Offers Some Insights on the Impact of Generative AI tools and their Marketing Power
David Leibowitz has over 20 years of experience in data, analytics, and cloud solutions. David has held senior roles at corporations like Microsoft and Dell where he focused on transforming business through data and technology to drive innovation and growth in the retail and consumer goods space. In this article, David offers some insights on Genartive AI and its capabilities as a strategic marketing tool.
Can Generative AI outshine the marketing pros at Kroger when it comes to winning my loyalty? As a loyal weekly shopper, I’ve often found myself asking this question while flipping through the coupons and promotions that land in my mailbox. Don’t get me wrong; I appreciate a good discount, but I can’t help but wonder about the missed opportunities in personalized marketing.
My shopping list is fairly consistent, filled with staples like fruits, vegetables, and eggs. Yet, the grocer’s promotions sometimes leave me scratching my head. For instance, coupons for eggs or milk — items I’d buy regardless of the price — make me ponder the strategy.
While Kroger has a good grasp of what I regularly purchase, it seems to overlook what I don’t buy. Items like canned goods, household cleaning products, and over-the-counter medicines are conspicuously absent from my basket as well as their targeted marketing efforts. It’s safe to assume that I, like everyone else, use toilet paper. So why isn’t Kroger trying to capture that share of my wallet?
Though my loyalty firmly rests with Kroger, I also shop at online retailers and discount clubs to meet my family’s needs. I’m not unique; many shoppers, up to 67% according to Grocery Dive, visit two or three different stores for groceries each week. The question is, does Kroger know what they’re leaving on the table? These are the kinds of questions that keep marketers up at night — how to strike the perfect balance between retaining existing customers and making bold moves to expand their share of the consumer wallet.
“The absence of staples and cleaning supplies most likely means [the customer] is shopping for these elsewhere. It’s pretty safe to assume that everyone uses toilet paper.” — ChatGPT
A Focus Group of One. Me. and ChatGPT
To delve deeper into this question, I turned to ChatGPT, a Generative AI model, to serve as a hypothetical marketing strategist. The exercise involved common steps used by seasoned marketing and loyalty professionals:
Analyzing shopping history to glean insights about customer behavior. Could ChatGPT guess my demographics, preferences, and traits?
Crafting a customer persona based on the data. Would ChatGPT be able to create a vivid representation of me as a consumer?
Enriching the data with insights about preferences. Could ChatGPT recognize where the gaps and opportunities might lie in positioning new products and services to me?
Strategizing on specific hyper-personalized campaigns to enhance customer loyalty. Would ChatGPT weigh customer retention over basket expansion?
Suggesting brand partnerships as a recipe for success. What might be the best brands to use for co-marketing opportunities?
Measuring the effectiveness of the new marketing strategy. How would ChatGPT suggest we measure the tangible success of a new program?
Pioneering a novel ‘passive feedback loop’ to improve understanding of the customer. Surveys are biased and prone to providing weak signal. How might we capture more data to improve the model?
Setting the Experiment: ChatGPT’s Capabilities, and Limitations
To put these questions to the test, the experiment was devised using ChatGPT+ as a stand-in for a marketing analytics tool by uploading four weeks of paper receipt details, which included transaction date, product, UPC, unit, and price and whether the item was purchased at a discount.
Since this data isn’t anonymous and is easily verifiable (it’s my own shopping history, after all), we can more easily validate the assumptions made during the analysis.
Access to ChatGPT+ allowed its Advanced Data Analysis to convert the PDF receipts into structured data that could be interpreted, analyzed, and visualized. I recently walked through how ChatGPT could be used to automatically create the Python code to read receipts using OCR (Optical Character Recognition) based upon simple English instructions.
However, it’s important to note what ChatGPT doesn’t have access to. For example, it only had access to the description “Bolthouse Farms” and a flavor (Chocolate, Vanilla, etc.) along with the category of “Beverage.” From this purchasing pattern, ChatGPT later suggested a cross-promotion with Pepsi or Coca-Cola. However, given the drinks are soy protein meal replacement shakes, a promotion with a carbonated brand would fall flat.
Most grocers, including Kroger, would have access to this data, including nutritional content or subcategories for enriched customer profiles. This limitation sets the stage for our exploration: how closely can ChatGPT identify my preferences and market to me with a restricted dataset?
Crafting a Consumer Persona, Call Me ‘Adaptable Alex’
After analyzing the data, ChatGPT summarized traits, and assumed preferences. It even coined a name for the resulting persona, “Adaptable Alex.” For brands, the persona serves as a representation of a consumer cohort, based on data-driven insights to help marketers understand behaviors, needs, and opportunities.
Demographics and Personality: ChatGPT successfully pegged Alex as a family man likely in his 30s or 40s. The AI also picked up on some nuanced personality traits, describing Alex as flexible, practical, and balanced, with a dash of spontaneity.
Preferences and Lifestyle: Alex is the kind of consumer who enjoys the comforts of home cooking, as evidenced by his frequent purchases in the Grocery and Dairy categories. He’s health-conscious but not a purist; his cart has an abundance of fresh fruits and vegetables, but snacks can also be found. When it comes to certain categories like seafood, Alex is willing to pay a premium, indicating a preference for quality over quantity.
Goals: Alex aims for efficiency in his shopping trips and strives for a balanced lifestyle, avoiding extremes in any category.
ChatGPT also picked up on a sudden spike in spending, possibly indicating a restocking trip, and backed this observation with both qualitative and quantitative data that were surprisingly close.
In summary, most of the assumptions — including age — were surprisingly spot on. Even from a limited set of transaction history, ChatGPT was astute at determining basic purchasing traits and preferences like cooking at home and eating healthy, within reason.
Enriching the Persona with Questions
Asking ChatGPT what questions would enhance understanding of Alex’s preferences was the next logical step, to avoid making broad assumptions that might lead to costly marketing engagement mistakes.
The most salient suggested questions by AI were:
Why the inconsistency and week-to-week variability in Alex’s shopping? Is it promotions, lifestyle changes, or something else? Understanding this might help to tailor personalized campaigns.
Are the purchases for an individual or household? ChatGPT correctly surmised that “the variety in the basket — ranging from personal care items to diverse food categories — could suggest this is a household’s shopping and not just for an individual.”
ChatGPT was particularly astute at identifying categories entirely missing from the monthly basket, such as staple groceries and cleaning supplies. It also questioned the absence of cereal despite the high purchase rate of dairy — a common pairing.
This level of scrutiny reveals gaps in consumer spending that Kroger could potentially fill. For example, if staple groceries are missing, it’s likely Alex is buying them elsewhere. If so, why? Is it a price issue, or does another retailer offer a brand that Alex prefers? These are the kinds of questions that can lead to actionable insights for hyper-personalized marketing strategies.
By addressing these gaps, retailers like Kroger not only stand to increase its share of Alex’s wallet but also fine-tune its marketing strategies to be more aligned with actual consumer behavior.
From Insights to Action: Enhancing Consumer Loyalty
For the next stage, ChatGPT received additional data, confirming that Alex’s shopping behavior for the four-week sample could be assumed to be representative of six months, thus establishing him as a loyal customer. By validating that shopping behavior was habitual rather than situational, a more stable foundation for strategic marketing could be developed.
Enriching the customer persona with attributes like ‘loyal customer,’ established buying patterns, and a stable life stage which could be used to devise a marketing strategy. ChatGPT recognized the complexity yet necessity of expanding the share of wallet. ‘The consumer is already spending; the challenge is to capture more of that spending with you,’ it reasoned.”
Initial strategies varied widely, from broad approaches like ‘double down on retention’ to ambiguous ideas like aligning seasonal strategies with promotional calendars. Some were unreasonable, taking fantastic leaps like offering cooking classes featuring seafood or developing a tiered loyalty program to reward long-term behavior.
Expectations were recalibrated in the prompts, noting that these suggestions were overly broad, potentially too costly, and wholly unsupported by evidence from a single shopper. Diving right into pointed targeted hyper-personalized strategies that could be deployed digitally were the instructions.
Now armed with validated data about the consumer and specific guidance, ChatGPT was tasked with developing an optimal marketing campaign strategy:
Crafting a set of novel, yet economical hyper-personalized campaigns for the Adaptable Alex persona
Pinpointing consumer goods brand partnerships that would ideally align and possibly resonate with Alex. This is crucial as the average grocery store has about 33,055 SKUs, according to the Food Industry Association. It’s not feasible to promote every product on the shelf.
Metrics and methods to measure campaign effectiveness, and
Developing a passive customer feedback loop
Crafting Hyper-Personalized Campaigns
ChatGPT was then prompted to assume the role of marketing leader and to suggest optimal marketing campaigns. Would ChatGPT also recommend printed coupons for recently purchased items?
Of the next eight examples, most were generic sale announcements or variations on coupon promotions. But three strategies stood out as intriguing, and unexplored:
Personalized homepage: Prominent placement and highlighting frequently purchased items with current promotions could be a game-changer. While the Kroger app does showcase prior items to “buy again,” this feature is relegated to the lower third of the screen. The majority of the screen real estate is occupied by promotional products that may not directly align with consumer data.
A monthly meal planner: This novel approach to direct marketing showcases meals based on items frequently purchased, offering an easy way (a swipe on the app, a QR code in print) to add new ingredients to the cart. Unlike Kroger’s current seasonal mailers, which are not personalized recipes, this strategy leverages existing data from recent shopping trips to suggest meals requiring only a few additional ingredients. For example, “You already have chicken, salsa, chips, and lettuce; why not add shredded cheese and tortillas for a taco night?”
A “we missed you last week!” notification triggered by a missed trip, or a significant drop in spending, could serve as a gentle nudge to reengage the customer.
Personalized health tips: Given Adaptable Alex’s apparent aim for a balanced lifestyle, incorporating health tips or recipes could resonate well. For instance, if data suggests that potato chips and soda haven’t been purchased in six months, promotions could be optimized to feature healthier options.
These hyper-personalized approaches directly address consumer habits and needs while remaining cost-effective.
Exploring Brand Partnerships & Promotions
The subsequent task for ChatGPT mirrors a challenge many marketers face: identifying the optimal mix of CPG (Consumer Packaged Goods) brands that add unique value and encourage purchases without overwhelming the consumer with promotions. Similarly, brand partners aim for a high Return on Advertising Spend (RoAS), making a broad-brush approach to consumer marketing impractical.
First, ChatGPT was challenged to identify the right promotion categories for Adaptable Alex. It was then asked to get specific with brand recommendations, along with business justification. Of the potential partnership categories, ChatGPT found that the “aim would be to reinforce loyalty where it’s strong and incite new behavior where it’s lacking.”
Here, it noted unsurprisingly that dairy, fresh produce and beverages topped the reinforce loyalty category, further suggesting brands “offering organic or specialty produce.” For expanding Alex’s shopping list, Personal Care and Pantry Staples were suggested, with the aim of diverting those purchases to Kroger from other retailers.
ChatGPT recommended some specific brand partnerships that seemed to align with the data:
Dairy: Brands like “Chobani” or “Organic Valley” could be a good fit.
Fresh Produce: “Dole” could offer premium salads or fruit packages.
Personal Care: Brands like “Colgate” for dental care or “Dove” for soaps and shampoos could incentivize Alex to fill the void in his cart.
Pantry Staples: “Quaker” could offer oatmeal, cereals, and granola bars to fill the pantry gap.
Measuring the Marketing Strategy Impact
Traditional metrics such as engagement rate, conversion, and customer lifetime value are often the default for evaluating campaign effectiveness. ChatGPT suggested Net Promoter Score to gauge customer satisfaction as well. However, these KPIs don’t fully capture the nuances of consumer behavior or the potential for growth within specific categories. To address this, a more comprehensive metric was developed in collaboration with the AI.
Total Addressable Value of the Customer (TAVC), a fresh metric designed to offer a more nuanced understanding of customer value addresses total share of customer wallet, similar to a TAM (total addressable market). By comparing this against the spending patterns of similar consumer cohorts, TAVC identifies untapped growth opportunities. It also allows for a cost-benefit analysis, weighing the potential for category growth against the costs associated with acquiring new customers or encouraging new product adoption.
Total Addressable Value of the Customer (TAVC):
TAVC goes beyond the usual metrics to provide a deeper understanding of a customer’s value. It starts by establishing a baseline and identifying a cohort of similar consumers based on demographics, spending habits, and preferred categories. This expected basket or Total Addressable Value serves as a point of comparison.
Category-Specific TAVC: This segments the overall TAVC by category, revealing areas with growth potential. It helps in prioritizing categories for emphasis in personalized promotions.
Cost-Benefit Analysis: This involves weighing the potential growth in each category against the cost of customer acquisition or new product adoption. Categories with high returns and low investment costs become priority targets.
Key Performance Indicators:
Share of Wallet Shift: Compare “Adaptable Alex’s” post-campaign share of wallet for targeted categories against the cohort baseline. An increase suggests the campaign’s effectiveness.
Basket Composition: Track changes in the average basket size and composition. For example, are pantry staples now consistently part of the basket, and how does that compare to the cohort?
Incremental Sales Value: Measure the additional sales value generated by the campaign for targeted categories. This should be incremental to the baseline expected sales for similar consumers.
Category Penetration Rate: Check the percentage of “Adaptable Alex’s” shopping trips that now include the targeted categories. A higher rate indicates a more permanent shift in shopping behavior.
ROI on Promotions: Calculate the return on investment for each targeted promotion. High ROI categories might warrant further focus, even if their initial TAVC was lower.
Time-to-Conversion: How quickly does the shopper take advantage of promotions or adopt new categories? This can be a measure of how well the campaign resonated.
By employing TAVC and these refined KPIs, a more holistic and actionable understanding of campaign effectiveness is achieved, allowing for strategies that are both innovative and precisely targeted.
Automating a Passive Feedback Loop
Traditional methods like point-of-sale or email surveys often fall short in capturing the nuances of customer experience. They can be biased, have low response rates, and add friction to the customer journey. Instead, a more effective approach is to employ passive data collection methods that offer valuable insights into customer behavior without requiring direct input. Here’s how this can be achieved:
In-App Behavior: Tracking interactions within the mobile app can reveal key insights. For instance, which sections are visited most frequently? Are items added to the cart only to be removed later? This data can indicate areas of interest or potential pain points and can be compared against a broader cohort of similar consumers.
Purchase History Analysis: Analyzing buying patterns to alert when shifts or changes are noted. If Alex suddenly stops buying a previously regular item, it could signal dissatisfaction with that product or category.
Heat Maps in Store: Physical store maps cross-referenced with Alex’s purchases can show where Alex spends the most time. Areas where they linger could be of interest, while areas they avoid are also telling in the customer journey.
Compare against store inventory data: If we notice a drop in purchasing of typical products that Alex buys, is it possible there were supply chain issues that affected inventory or out of stock, rather than a preference issue? This is important to note, as stock-outs cause walkouts, according to a Harvard Business Review study.
By implementing these passive data collection strategies, retailers can gain a more nuanced understanding of customer behavior, allowing for more targeted and effective marketing efforts.
Is Generative AI Replacing the CMO?
As we delve deeper into the capabilities of generative AI, a pressing question emerges: Is the role of traditional marketing fundamentally changing?
In the exploration with ChatGPT, the analysis was astute and rapid. But it also required industry expertise for coaching. Early on, it wondered, “Is Kroger their only go-to, or are they shopping elsewhere for missing categories?”
To this, I replied, “there are no staples or cleaning supplies. We must assume that Adaptable Alex uses toilet paper, right?” Back on track, ChatGPT agreed and responded, “The absence of staples and cleaning supplies most likely means Alex is shopping for these elsewhere. It’s pretty safe to assume that everyone uses toilet paper.”
I was at once reminded that ChatGPT, is still at best, a copilot. An intelligent, tireless, but not infallible assistant. Understanding the unique facets of the business, and operational knowledge are still very much required to get the best value from generative AI. The ability to integrate with other internal and external data sources to understand the consumer and provide value are required.
As to whether ChatGPT knew me from the minimal data it possessed, let’s just say it was hauntingly close. It pegged my gender, age, and a number of other traits. Not wanting to lead the witness, I prompted ChatGPT to interrogate the data trends, revisit the stocking-up assessment, and surmise where their items might be purchased.
It was spot on that given “the consumer’s selective shopping behavior at this grocery store, I’d lean toward warehouse clubs or online retailers” for household basics like toilet paper, paper towels, and garbage bags. “They offer both the convenience and the cost-effectiveness that seems to align with this consumer’s pattern.”
So, is Generative AI poised to replace the CMO?
The answer is a nuanced ‘no’; while it can augment decision-making and provide valuable insights, the human element — understanding the unique facets of the business and operational knowledge — is right now, irreplaceable. But marketers that utilize AI and tools such as these effectively will be able to reach more consumers with strengthened and value-based relationships and expand their growth potential, at a fraction of the cost and time.
** This artcile was adapted from David Leibowitz's page on medium.com.