When Whole Foods announced their plans for a line of grocery stores geared towards millennials, the announcement was deemed “offensive” and stock prices dropped. According to Marketing Mag’s Katie Martell, the Harvard Business Review, and a number of other outlets, there’s nothing inherently wrong with the plan. However, there is something wrong with Whole Food’s predictive marketing models.
Harvard Business Review’s Robyn Bolton writes that members of Generation X and Baby Boomers also want access to “lower-priced, organic, and natural foods.” While Whole Food’s product line announcement was frequently deemed “offensive,” Bolton believes the real problem is with Whole Food’s marketing models. Simply being in a demographic doesn’t predict certain preferences or behaviors. Organizations who make similar mistakes might not receive as much criticism as Whole Foods. However, they’re unlikely to achieve great results.
The advent of big data provides marketers with a new ability to build vivid customer segments. Behavior, beliefs, and intent can yield models that are much richer than demographic generalizations. However, many attempts to build predictive marketing models fall flat in terms of outcomes. Join us as we review the most common issues behind poor segmentation and predictive modeling results in marketing.
1. Poor Behavioral Data
Ultimately, the purpose of predictive marketing models is to predict how consumers will behave in the future. Without extensive or accurate knowledge of how your segments have behaved in the past, this is difficult to accurately model.
Infogroup research indicates that just 33% of marketers believe they collect enough behavioral data on customers. Only 21% are “very confident” about accuracy in profiling. If your predictive models rely on your transactional data or self-reported sources, you may not have accuracy.
2. Limited Context
Perhaps you understand who your customers are, but do you understand why they’re buying from you? Arjuna Solutions writes that predictive models need certain insight into how your segments “behave in the marketplace.”
Contextual insights are a critical part of marketing models in an age where consumer identity and segments are more and more fragmented. Fast Company’s Dan Hermanadvocates the idea of building models based on context and purchase motivation.
Instead of “40-something home buyers,” a mortgage company using more context could discover they are trying to reach “40-something primary home buyers,” “40-something vacation home buyers,” and “40-something residential real estate investors.” By identifying the reason for the purpose, marketers can better understand goals, pain points, and probable behavior patterns.
3. Small Sample Sizes
The vast majority of marketing professionals have some background in research methods and statistics. However, many organizations are trying to market with models that are based on a tiny segment of the population.
In the age of big data, companies can’t compete if their marketing models are based on first-party customer data. Their models may be inaccurate if they’re using focus groups, surveys, or other dated research models. Obtaining recent, accurate, 3rd-party data from a Data Exchange Platform is the best way to understand how populations behave with confidence.
4. Flawed Data Logic
Marketing models built internally and externally may fail because they’re based on terrible data logic. Affinity algorithms can reveal incredible patterns, such as the famous example of when Target’s algorithm predicted that a teen was pregnant before her parents knew. In other cases, affinity algorithms don’t hold up. Visiting a high-end furniture store within the space of six months does not necessarily make a woman a “trendy” food consumer.
As MIT research highlights, clustering and other methods of discovering patterns are “inherently unsupervised.” This can reveal surprising truths, but it can also output garbage if your inputs are inaccurate or your sample is too small.
For organizations struggling to develop accurate predictive marketing models, the problem could be related to their big data. Small sample sizes, old data, or a lack of 3rd-party insights for context can all result in predictive models that don’t reveal the future.