Your Marketing Analytics Are Biased & It’s Costing You More Than You Think

By Swapnasarit Sahu

Unsplash

Data: It's the lifeblood of modern marketing. And yet so much of the data marketers rely on is incomplete, biased, and just downright flawed.

A unified, actionable customer dataset is now essential for any marketer who wants to deliver the kind of personalized customer experience that it takes to compete in the new digital landscape: from reaching the right audiences, to targeting campaigns on the right channels, and interacting at the right points in each customer's journey.

Equally important is the ability to enhance existing datasets to resolve identities and identify unauthenticated visitors.

But thanks to limitations in data collection strategies, too often customer data is incomplete or missing. And if this data is aggregated from multiple sources, the gaps in the data are magnified; eventually the model performance and analytics built on top of it simply fall apart. The data you're working with will deliver completely false or biased insights.

This problem of incomplete data sets and information is everywhere, across every vertical. The challenge is correcting this bias. (Since this bias is introduced in our data through procurement or sampling from external sources, the problem is called sampling bias.)

Sampling bias occurs when the sample collected from a population is not an accurate representation of that population. This happens when a sampling algorithm favors the selection of certain members of the population or because of collection constraints. The result is a non-random sample.

An example would be a country's population. Take India for example, which roughly consists of 60 percent men and 40 percent women. A sample of collected marketing data might show a distribution of 90 percent men and 10 percent women — which isn't real. Thus, the sample is biased and needs to be corrected.

Basic mathematical equations can address the problem of lack of information to produce unbiased analytics. Another example would be if a marketer only collects and analyzes data from shoppers with high spending capacities to launch a campaign targeting an entire database—which doesn't take the financial limitations of their broader audience into account.

Here are a few signs your marketing analytics might be biased:

Sign #1: Excluding multiple touchpoints from data-driven marketing campaigns

Customer journeys often zigzag through multiple channels and devices — including email marketing, social ads, desktop and mobile — and marketers need to take these different touchpoints into account to launch targeted promotional campaigns.

For example, a marketer who is preparing a broad marketing campaign solely based on desktop interactions will lose out on opportunities to appeal to customers who would usually interact with the brand on mobile, which would drive a higher ROI.

Sign #2: Lookalike audiences that exclude negative consumer interactions

Part of expanding reach to new customers is creating "lookalike" audiences based on behaviors among existing customers — such as lifetime value, purchase history and product affinities.

When launching a personalized campaign, a good approach by a marketer would be to leverage data from customers with a high lifetime value — but some marketers make the mistake of including customers in their lookalike audience who have a higher propensity to churn. This puts less value on customers with a high lifetime value, creating a campaign that is not representative of the most loyal customers.

Sign #3: Campaign optimization based on biased feedback

Say a marketer is looking to optimize an upcoming campaign based on data from recent customer feedback, but the feedback is limited to consumers who've had generally positive interactions with their brand. Because their analysis only incorporates positive feedback, there is a missed opportunity to improve the experiences of those customers who have had negative experiences, which can lead to a higher attrition rate in that group.

Recognizing data bias is the first step to fixing it. If your marketing analytics show any of these signs or is radically incomplete, your strategy could be wildly off base. And the sooner you solve for it, the less time you will lose optimizing for (and spending on) the wrong customers.

The views and opinions expressed in Marketing Maestros are solely those of the contributor and do not necessarily reflect the official position of the ANA or imply endorsement from the ANA.

Swapnasarit Sahu is the Chief Analytics Officer at Zeotap and is co-author of Correction of Sampling Bias to Produce Unbiased Analytics, part of I-COM Global's The Frontiers of Marketing Data Science Journal. He has over 18 years of experience in analytics and machine learning, working in top multinational corporations such as [24]7.ai, Airpush, IBM, and Komali Labs.