The first episode of “Mad Men” features a focus group in a brightly lit room. Chairs are placed neatly in rows along laboratory-grade clean counters. Along these countertops lie small shiny cylinders, each with individual tags and a box of tissues to accompany them. Amongst these chairs enter a line of secretaries, each assigned to a seat. Throughout the scene, a facilitator asks leading questions to the participants, prodding for feedback. The fictional company is Sterling Cooper, employed by Belle Jolie, a global beauty company, which solicits the opinion of future customers. In the scene, the lead protagonist Peggy Olson, played by Elisabeth Moss, comments (to the delight of the focus group facilitator) that the receptacle of disposed tissues used for removing lipstick looks like a “basket of kisses.” The tagline is quickly adopted by Sterling Cooper management — a hit slogan in the making.
Although archaic by today’s standards, this scene is reminiscent of traditional customer analytics, namely:
- ask a lot of questions,
- group the findings into a pithy tagline,
- and act upon the result.
Customer analytics grew from the bedrock of survey analytics, which is still practiced by many companies that champion a more traditional approach to public opinion. Be it for a product or a political candidate, survey analytics and focus group research play a role in the data journey for many firms — but this is only the beginning. There is a natural arc of complexity for commercial firms that range from surveys and focus groups to more advanced methods that harness machine learning (ML) under the umbrella of artificial intelligence. The following series of anecdotes presents a potential analytical journey to explain the techniques that drive modern commercial practices that help improve the customer journey.
How Segmentation Improves the Customer Journey
Following this roadmap, let’s go back to the scene from “Mad Men.” In this scene, Peggy Olson and her colleagues are participating in a focus group. This strategy of focus group research, represented by the secretaries of Sterling Cooper, is used by most Fortune 500 companies. The benefit of focus group research is companies can generate key questions that then feed into survey questions. Rather than guessing what the customer may care about, they create a representative sample of their target population in a small group. The survey, based on statistical methodologies of sampling and representation, is then communicated via phone, online, in person, or through mail-in pamphlets.
Once descriptive statistics are generated and pretty pictures of charts and graphs are communicated in PowerPoint slides from survey results, analysts can go one step further and generate clusters of customers, which is known as segmentation. For example, picture a company that sells outdoor and adventure gear. Based on their survey feedback, they may determine they have varying segments of customers that shop at the store. By using a clustering technique, known in ML as “unsupervised learning,” the analyst can create groupings of individuals based on the survey responses.
Maybe we have the Sierra Club adventure goers. These folks love hiking and protecting the environment, lean towards Democratic candidates, and tend to join groups focused on environmental conservation. A different group could also be concerned about the environment — albeit for different reasons. These shoppers lean conservative politically and support protecting land for resources like fishing and hunting.
To group both of these groups into the same category could be a detriment to the company’s marketing efforts because there is nuance to the segments. Different marketing materials would be needed, as well as different sales tactics and deals to draw them in. If the adventure store were to send mailings about hunting rifle discounts to the Sierra Club cohort, they would lose customers. These marketing strategies that follow, like direct mailings and advertisements that vary by group, are called micro-targeting. Used by political campaigns and companies alike, they are highly effective at identifying the proper group to use resources towards.
Attribution and Mixed Market Modeling to Improve Analytics
The next step on the analytics roadmap is leveraging online marketing channels. Most companies have online portals that process sales. With each mouse click, there is a new opportunity to suss out new information. There are myriad platforms to collect and synthesize this information, including two key players in the market: Google Ads and Facebook Ads. These platforms are used heavily to help companies determine not only who visited their websites, but how they got there. Laying a trail of breadcrumbs, websites can track the path you take to get to a certain product.
One such technique is called attribution modeling. For example, say you saw a video on YouTube about a new house-sharing program. You click on the video, which brings you to the house-sharing website. In this case, YouTube will receive 100% of the attribution for having sent you there. Attribution modeling is a core component of identifying how customers find what you are selling.
Another strong element of online marketing channels is called mixed market modeling (MMM). Think for a moment back to the house-sharing example. Say you see that same YouTube commercial but don’t click on it. Then you check Facebook and see that same ad in the corner of your screen. You click on that ad. Which ad receives the credit? Is it Facebook where you saw the ad and clicked or YouTube which introduced you to it in the first place? By using econometric models that underlay mixed market modeling techniques, you can infer a percentage to each resource. From that percentage, you can assign a dollar value to the ad click and create a targeted ad campaign from that information.
Machine Learning for Smarter Customer Service
Next, we get into the world of ML. One common example that is frequently used in ML to understand customer preferences is sentiment analysis. Using ML algorithms such as those based on Bayes’ theorem, or more complex implementations of neural networks (deep learning), the feelings of a customer can be uncovered. For example, a user can enter a chat with a company representative. This chat window will at first feature a bot that is built on top of a recommendation engine. This recommendation engine is another source of ML architecture that tries to compare your needs with those of millions of others.
Once the bot has given you a list of recommended options, the sentiment prediction kicks in. A score can be given to your responses to infer how happy — or upset — you are (0 representing incredibly unsatisfied and 1 as perfectly satisfied). Hypothetically, a company can then use that sentiment score to send you to an actual person (if you were really angry) or back to a list of options that you can self-select (if you are giving kudos to the product).
This same strategy can be employed when it comes to customer reviews. A review with a sentiment score under a certain threshold, say 50% (0.5), may be flagged for human response. This introduces the “human-in-the-loop” element of ML. Since company resources are not limitless and based on a predefined budget, having a human agent type or speak with each customer is rarely feasible. There is a cost associated with each human action. Since the ML algorithm is assigning a probability score between 0 and 1, the threshold can inform the company when a human is needed to enter the picture, saving time and money.
Emerging Possibilities in Computer Vision
Finally, we enter the world of computer vision. Remember how we determined sentiment? Well, stores can use computer vision to determine how happy a customer is. This algorithm, albeit not new in theory, is relatively new in conceptual application. The field of deep learning (neural networks) has become more in vogue in the past two decades and is a mainstay of newer operations in commercial use. Recently, Amazon has created its cashier-less stores, which use computer vision to identify you as a customer as well as the items that you purchase without you ever seeing a checkout kiosk.
However, computer vision is fraught with ethical dilemmas, such as, what applications will use the data generated from your facial images to personal property issues that arise around perceived privacy. This is a space that the AI community is actively watching and contributing to.
Moving through our roadmap of analytical capabilities from survey research — to micro-targeting, to recommendation engines, to computer vision — offers a mere sampling of the techniques that drive modern commercial practices to improve the customer journey. With the breadth and depth of big data advances and computing hardware leaps, newer and more robust AI techniques avail themselves each year, although there is a caveat to these technological advances. The path to a mature AI capability is not to jump to the latest advancement, but to have an understanding of the data at hand and to tailor your approach to the technique that fits your specific use case.
It is not always the most advanced algorithm that drives a company’s success, but the most understood and efficient use of resources. Peggy Olson in “Mad Men” understood that a simple tagline was most effective, and frequently companies will employ simple analytical techniques to reap the greatest reward.
Maven Wave helps drive the future of industry with innovative business outcomes, fueled by cloud, with risk top of mind. To help organizations maximize economic outcomes and advancements, Maven Wave brings a rich blend of industry-specific technological expertise, agile-integrated design, and best practices for transformation. Contact us to learn more.
Get the latest industry news and insights delivered straight to your inbox.