Retail Customer Analytics: The dirty little secret

If you’ve ever been asked to evaluate technologies for segmenting and analyzing customers, you’ll probably be shown beautiful visualizations or maybe enticed by highly useful metrics, such as customer buying predictions for different products, expected spending by customer, or risk of churn. At some point, you will ask the vendor how is it done, and you will see how easy it is to click and automate your way to great summaries and results. A little secret you are unlikely to be shown, however, is all the work that was required to prepare the underlying data before any of this magic can happen.

Most analytics solutions work best when tons of data are available to analyze, which compounds the data preparation problem. However, some solutions mitigate this problem by employing more advanced data science to drive useful, accurate information and customer predictions from comparatively little data. This is important because data is usually messy to manage. The more data required get the analytics you want, the more resources required to source, extract, clean, transform, enrich, pre-aggregate, structure, and provision the needed data. This does not count the storage management, back-ups, data governance, scalability, and security infrastructure surrounding it. The data costs could be many times the analytics itself.

Analytics solutions that deliver value from whatever easily assembled data you already have will enable you to improve customer retention, campaign ROI, segmentation, and personalization right away. It’s surprising what the right data science can do with relatively little data. Building up your big data assets and infrastructure will allow you to do even more, but you can do that incrementally. It’s a more practical, less disruptive, and lower risk path to science-based customer analysis and marketing.

Here are some things to consider about data for customer analytics:

  • If you buy customer data from a third party, what percentage of customers are covered? How accurate is it?
  • If your first party data is collected inconsistently, what will it cost to clean it up? What systems have the data and how will it be integrated into a consistent structure and meaning
  • If you use web data to score or segment customers, how will you do that for customers who are not active online?

At what point do the exceptions, compensating rules, and data quality compromises make things too complicated or less usable? Evaluate the value of analytics against the total cost of analysis plus the data preparation.


Peter Moloney is CEO of Loyalty Builders, a cloud-based predictive analytics service enabling marketers to get revenue lift from more relevant communications to their customers.

Tags :