At a recent conference, I presented a talk on suitability of business problems to big data analytics algorithms and appliances, and one of the examples I shared referred to an article written by Charles Duhigg for the New York Times about habit formation and, correspondingly, the ability of companies to apply analytics algorithms to identify opportunities in relation to recognized patterns and corresponding habits associated with the customer community. My goal in sharing the story was to provide an example of how a large company is able to collect large amounts of data from a variety of sources and then analyze that data to find actionable knowledge.
Interestingly, after a discussion of the approach that the Target analyst took to identify a selection of products that indicates that a shopper (or “guest,” as they are called in Target) is pregnant, there is an anecdote about the presumed outcome of using the pregnancy analysis model. According to Duhigg’s article, “… a man walked into a Target outside Minneapolis and demanded to see the manager. He was clutching coupons that had been sent to his daughter, and he was angry, according to an employee who participated in the conversation.” The unnamed man was angry that his teen-aged daughter had received a Target mailer containing advertisements for maternity products such as clothing and nursery furniture. Confident that his daughter was not pregnant, he demanded and received an apology from the customer service representative. However, a few days later, this man apologized to the store because he had learned that his daughter was indeed pregnant.
The strategic proximity of the placement of this anecdote next to the discussion of the analytic model suggests the implication of cause and effect, namely that the Target pregnancy model was in place and was operational. The article does state that Target did agree that analytic models were being developed but did specifically reply to the author that “Almost all of your statements contain inaccurate information and publishing them would be misleading to the public.” At the same time, the author states that a Target executive shared thoughts about the best way to embed the targeted products in advertisements among other non-maternity items to help influence purchasing behavior in a less “spooky” way.
Interestingly, after my talk, a member of the audience introduced himself as a Target employee and told me that while the story is entertaining, according to Target this never happened. And while I am not about to get involved in a “he-said/she-said” discussion of the veracity of either party’s stance, the scenario does call for some exploration to consider characteristics of the integration of an analytics program with a marketing campaign to drive value and what needs to be in place to best execute against the strategy.
But first, a reality check. The existence of a comprehensive analytics program also implies a well-organized data management program and integration across the retailer’s various lines of business. A marketing campaign must also be supported across the end-to-end business process, and there must be clear metrics to understand whether the desired outcomes are achieved. Let’s see what that means for Target’s maternity marketing program.
The end-to-end process for a retailer might involve steps like this:
- Identify the target audience (pun not intended) to be considered for micro-targeting
- Collect the data, model, review and deploy the analytics application
- Perform the analysis and collect results
- Develop the marketing campaign around the resulting set of targeted customers
- Use the marketing campaign to influence customer behavior
- Ensure that the desired behavior can be supported by the retail sales channels
- Continuously measure the results
- Compare to the developed analytical model and assess the lift
This is a summary of the process, and leaves out a lot of the details of the process. For example, for step 2 (collect the data) this means that the retailer has comprehensive means for collecting information about customers from across the different channels and touch points and can determine who the entity is and consistently resolve that entity’s identity. Practically speaking, that means integration of customer data across all lines of business (brick and mortar, eCommerce, in-store customer service, call center, etc.).
Steps 5 (influencing desired behavior) and 6 (ensure support by the channels) also span multiple areas of the business. Most importantly, the supply chain must be able to support the purchase process. If a customer has clipped a coupon, driven to the store and found an empty shelf, obviously, the company’s goal will not be achieved. Other dependencies include the ability to correlate the purchase to the marketing campaign (which can be done using embedded codes), and that implies an additional data management process in place for tracking. That is what is implied by step 7 (continuously measure the results), which also requires dedicated staff for data collection, organization and analysis.
In other words, a lot of organization (program management), defined processes (planning and integration) and oversight (governance) need to be in place for all of this to work in a streamlined way. That being said, consider your own experiences with retail chains: How often are the shelves fully stocked with the products that customers are looking for? Can customers return items purchased online at the brick-and-mortar store? How well coordinated are the staff members to provide satisfactory support? Supply chain management, cross-channel integration, and customer experience and support are fundamental corporate capabilities. If your organization is not suited to perform these activities and processes well, it might be better to focus your company’s data analytics on understanding performance bottlenecks and opportunities for improvement.
SOURCE: Analytics, Hype and the Limits of Reasonability
Recent articles by David Loshin