Retail Customers: They Are What You Analyze
Retail marketers swept along with the big data wave hope that analyzing ever more variables will uncover hidden indicators: ones that provide more accurate predictions about which customers are ready to buy, what products are they likely to buy, what will they spend, and who will defect.
As a result, marketers sometimes try to collect everything – any and all customer attributes and behaviors they can get hold of — and soon find themselves tumbling down into a deep, costly black hole of data. But there is a simpler, far more cost-effective way to get even better results.
The key is to focus on the right customer data for a specific purpose then analyze to derive the most predictive power from that data, powering loyalty targeting and 1:1 product recommendations that drive more revenue. Here are four steps that retailers can follow for a higher-ROI from data-driven marketing.
The Big Data imbalance:
- 80% — Of total time and cost spent on data gathering, cleansing, integration, normalization
- 20% — Of total time and cost spent on data exploration, analysis, modeling, testing.
The imbalance increases as more source data bloats the analysis, leading marketers on an ROI-crushing – and unnecessary — chase for more useful predictor variables and a “360° view of the customer.”
To predict customers’ purchasing behavior, focus on their purchase history. It’s far more simple and cost-effective.
This doesn’t refer to the typical RFM approach (recency, frequency, monetary value). Instead, we can now make predictions that are either impossible with RFM alone or made much more accurate by deriving more predictive indicators. These include inter-order timing, products purchased, sequence, timing trends, range of categories purchased and other indicators, pulled from the basic transaction records. And by constraining the process to specific predictions derived simply from transaction records, the analytic process itself can be automated.
Transaction records are easy to access, always complete, always accurate, always unambiguous (if not, the company has far more fundamental problems to worry about), and, for businesses that record transactions by customer, always available for every customer. It’s also not considered “personal” data or controversial for internal marketing purposes. Additional data on customers can be helpful, but can lead down a slippery slope of increasing time and cost when, in fact, all the data you need is already collected and at your fingertips.
The alternative, analyzing tons of data on every customer, is daunting. All that disparate data needs to be sourced, extracted, cleansed, integrated, disambiguated, transformed, normalized, enriched, stored, governed, continuously updated, often synchronized with source systems, explored, analyzed, mined, modeled for making predictions, and tested. But not if you automate analytics
Using a few unambiguous and very predictive variables usually delivers the best accuracy. That’s because too many variables can lead to errors of over-fitting the data with false indicators or attributes that are already highly correlated. The real world is complex. It’s hard to model it correctly and adding poorly understood or ambiguous variables seldom improves accuracy. Instead, it can lead to unjustifiable complexity.
Fewer input variables reduce the cost of data acquisition and computation times while yielding models that are usually easier to interpret and optimize. The approach that delivers the most revenue from better recommendations and minimizes the cost of data preparation and analysis wins. It’s hard to beat simple but effective.
Where there is a limited, known, consistent set of data inputs and outputs, it’s possible to automate for scale, repeatability, and ROI. Automation drives down cost, reduces errors, saves time, and affords consistency and repeatability to the marketing process.
Loyalty predictions and 1:1 product recommendations can be calculated for each of thousands or millions of individual customers in one pass. With automated analytics, retail marketers can get ongoing, on-demand access to purchase predictions, risk and value scores, and lists for email, direct mail, catalog, web, mobile, and sales campaigns without the typical time, cost, and complexity of data preparation and data modeling.
Peter Moloney is CEO of Loyalty Builders, whose Marketing Lift Service offers a simple, cloud-based predictive analytics service enabling marketers to get revenue lift from more relevant communications to their customers.