You Can’t Sell Shampoo to a Bald Guy: AI Best Practices for Marketing
Recently, my Facebook feed has become clogged with advertisements. Many of those advertisements are selling products that aren’t a good match for me, including ones for shampoo and hair care products!
I was offered a “session of our signature VE Protein Hair and Scalp Treatment plus our best-selling Pro-Hair Essence at only $28 (U.P $438)!”. All I had to do was click on the button that said, “Get Offer.” Instead of clicking on the offer, I sought more information about the advertisement and discovered that the advertiser “wants to reach people who are interested in TV reality shows, based on your activity on Facebook, such as liking pages or clicking on ads.” This information just deepened the mystery. I don’t have hair, and I don’t watch any reality TV. And why would watching reality TV require a person to seek hair treatments?
This advertisement was not an isolated anomaly. I started paying more attention to others. One was written in Chinese, which I can’t read. Another tried to sell women’s cosmetics to me. And another one was selling funeral services!
Facebook doesn’t have the limitations of television, radio, or printed newspapers. It allows advertisers to target their advertising to individuals. So, why were businesses targeting me for products that I’m not going to be interested in?
Out of curiosity, I began to click on the info button that promised to answer the question, “Why am I seeing this ad?” and was surprised by the answers. Many advertisements had selected me simply because I was older than 18 years old, and because I live in Singapore. Other advertisements said that the reason is that I speak English, including an advertisement written in Chinese! Some advertisements provided cryptic reasons, (e.g. “One reason that you are seeing this ad is that the advertiser wants to reach people who may be similar to their customers.”) One cosmetics advertisement told me, “One reason you are seeing this ad is that the advertiser wants to reach people who are interested in flowers, based on your activity on Facebook, such as liking pages or clicking on ads.”
I discovered that Facebook had incorrect information about me. It thought my interests included cosmetics, shoes, reality TV, and NFL, but in truth, I am interested in science fiction and rock music. My story is not unique. A colleague of mine with a strong computer science profile checked what Twitter thought about her and discovered that Twitter had incorrectly assumed that she was a he!
These examples illustrate the need for marketers to adopt two best practice approaches that weren’t available to ordinary businesses until recently.
Best Practice Principle One: Treat Customers as Individuals
Until recently, marketing teams didn’t have access to detailed data or powerful computing. So they used cluster analysis to segment customers into one of several groups, or worse still, they marketed the same product to everyone. The internet has made marketing campaigns easy, and the cost of an advertisement is low enough to make large campaigns affordable.
But broad-based marketing techniques ignore the true cost of advertising. Every time you try to sell an irrelevant product to a customer, you are teaching them to ignore your advertising. You are damaging your brand value and damaging the effectiveness of future marketing campaigns.
Customer segments are too broad. This approach treats all the people within a segment as being identical to each other, and that is not true. Compounding the problem is the use of unsupervised segmentation techniques, such as cluster analysis. These are algorithms that group customers using an arbitrary definition of similarity that may be completely unrelated to the characteristics that drive purchasing decisions. For example, it may group people by age and gender, whereas purchasing decisions may be related to geography and education.
Consumers are becoming more sophisticated. They expect to be treated as individuals. Best practice retailers, such as Amazon, know their preferences. I might go shopping on Amazon to purchase one product, but I buy five because they are so good at knowing what products to show me that I will get excited about.
Best Practice Principle Two: Give Human-Friendly Explanations and Allow Corrections
Over the past several years, there has been a convergence of data protection principles used around the world to treat consumers as individuals, as reflected in legislation and regulations. One of the key principles of data protection is that people should have the right to see what data you have stored about them and to have incorrect data corrected.
This principle can be extended to artificial intelligence systems. Europe’s GDPR gives consumers the right to review an algorithmic decision. Banking regulations such as the Fair Credit Reporting Act may require a lender to explain why a loan application was declined. Many of the AI Ethics principles published in the past twelve months include the idea that where practical, consumers should be given human-friendly explanations for algorithmic decisions about them, especially decisions that have a negative impact upon the consumer. Amazon has a section on its website where you can provide feedback to improve your recommendations. Facebook provides explanations for why it shows you each advertisement, although some of those explanations are too cryptic to qualify as human-friendly.
Prediction explanations and AI decision explanations have many benefits. They provide common sense checks of AI decisions at a detailed level before you deploy your model into production. By explaining why customers have been included in a marketing campaign, content and messaging can be aligned with AI decisions, powering better results. Customers can become more engaged if they understand why they were selected, highlighting which data fields affected the decision and focusing their attention on data values that may need correction.
Best practices are for your AI system to provide human-friendly explanations, allowing customers to correct any data that led to incorrect decisions.
Best practices have changed in the past few years, and with the power of cloud computing and automated machine learning, best practices are now available to all organizations, not just the tech giants.
DataRobot puts the power of modern computing and powerful predictive algorithms in your hands. With DataRobot, you can predict which customers will react positively to a marketing campaign. See this webinar for a live demo of a marketing use case that treats customers as individuals.
DataRobot provides human-friendly explanations for model predictions. Our XEMP prediction explanations explain how a customer’s data is different from a typical customer, and why that resulted in a different result or decision versus a typical customer.
About the Author:
Colin Priest is the Sr. Director of Product Marketing for DataRobot, where he advises businesses on how to build business cases and successfully manage data science projects. Colin has held a number of CEO and general management roles, where he has championed data science initiatives in financial services, healthcare, security, oil and gas, government and marketing. Colin is a firm believer in data-based decision making and applying automation to improve customer experience. He is passionate about the science of healthcare and does pro-bono work to support cancer research.