A great many market researchers are “numbers people”. This shouldn’t be a surprise, when we consider how central statistics, sampling, projection, and prediction are to our business. But the all-too-common gulf between market researchers and our data scientist colleagues is a surprise. This disconnect between market research and data science inhibits capabilities in the short term and hurts long-term competitiveness. Which is silly, because this disconnect is completely unnecessary.
More Data = More Silos
A colleague of mine runs research at a national US media company, with both advertising and subscription-funded properties distributed over many platforms in multiple markets. Their business is complex, and driven entirely by consumers’ rapidly evolving media consumption.
And like any sophisticated 21st-century organization, they’re collecting data from a dizzying – and growing! – array of sources: syndicated audience measurement, set-top box, consumer surveys, qual/quant combos, analytics of every flavor (web, app, social, video, advertising, etc.). If it’s data, they’re either already using it, or gathering it and planning to use it.
Like any large company, they’re not drinking from one fire hose: They’re guzzling from hundreds. This means multiple teams, with different skill sets, responsible for managing different sets of data and generating different types of insights. If that sounds like a recipe for organizational silos, that’s because it is.
And my colleague, who oversees a veritable forest of these silos, laments:
“I’ve got market research teams talking sample, and data scientists talking algorithms, and they just aren’t speaking the same language.”
For a Chief Research Officer or a Chief Information Officer, this is understandably frustrating. The glossy business magazines that the CEO reads on the airplane scream the importance of big data, shaping the Board’s expectations of how new data streams will revolutionize business. But as anyone who’s actually worked with data knows, the devil is in the details.
The Shared Roots of the Data Science Disconnect
Despite the starry-eyed visions of comprehensive AI, humans remain a vital component of generating insights. Machine-learning, relational databases, Hadoop clusters, NoSQL data stores, and adaptive heuristics might process the data faster or surface statistically significant data points – but they can’t (yet) tell us what they mean. Understanding implications and predicting consequences remains a uniquely human skill.
Happily, this ability to internalize evidence, extrapolate conclusions, and assemble a narrative is a fundamental skill shared by “traditional” market researchers and “modern” data scientists. Where the two “schools” diverge is in the tools and processes by which they apply this skill.
Different Tools, Different Worlds
My father used to tell me that when all you’ve got is a hammer, everything looks like a nail. That’s just the way our brains are wired, and so the perspectives of data scientists and market researchers are naturally shaped by their experiences and the tools they’re comfortable with.
Historically, quantitative market research grew out of a blending of sociology, economics, and mathematics. Quantitative market researchers rarely had the luxury of working with either census-level or observational data. Costs and available technology both contributed to a reliance on declarative responses gathered through surveys collected from (at best) representative samples. This trained generations of researchers to focus their attention on bias, weighting, attribution, etc. and elevated the cross-tab into a powerful tool to surface findings. Over time, these techniques evolved into more complex analytical tools like conjoint, CHAID, etc.
The modern crop of data scientists, by contrast, emerged from a background of computer science, physics, and mathematics. More often than not, the formative data sets on which they cut their teeth did not have to be sampled: Census-level data was readily available, and data could be collected observationally rather than declaratively. Though it’s a young profession, their formative experiences have often focused on entirely different concerns: Data cleansing, predictive modeling, significance testing, etc. Their analytical arsenal may include ANOVA, Bayesian modeling, ElasticNet regression, etc.
These are very different techniques, each requiring a different set of tools and a different philosophical approach. There’s a reason why data scientists are so enamored of the rapid-fire A/B test: With their tools and techniques, the incremental cost of acquiring a new iteration of data approaches zero. For traditional market research techniques, the incremental cost grows linearly until you’re able to achieve economies of scale. Though some might think that a lower incremental cost of data collection makes a data science approach “better”, that is not necessarily true.
Market researchers are often seeking to understand causation, and their tools delve into the depths of consumer psychology to illuminate motivation and context (“why did something happen?”). Data science – with its focus on observational data (“what happened”) isn’t designed to capture this type of context, or to easily explore the inner life of the consumer.
Too often, market researchers and data scientists are simply unaware of how their approaches differ or of the opportunities gained from each set of skills. Yet both schools start from a shared foundation and progress to a common output: They start with mathematics (and statistics, to be more precise), and they end with storytelling. And building on this shared foundation and common output is the key to bridging those silos within any research group.
Cross-functional Research Science
In my experience, the key to integrating market research and data science teams is to build a shared understanding of the capabilities and limits of each approach and each dataset.
No responsible market researcher will claim that a generic survey on a representative sample of sufficient size to deliver a 95% confidence interval is capable of answering any business question,just as no reasonable data scientist will claim that machine-learning heuristics on a single dataset can solve any business problem. And yet, by integrating context and evidentiary analysis, these two approaches can often plug one another’s gaps and provide a more complete understanding of consumer motivation, behavior, and how marketers can affect it.
Building this shared understanding means educating market researchers about the techniques and perspectives of the data scientists, and simultaneously educating data scientists about the techniques and perspectives of the market researchers. Each group is primed for this: They’re both approaching similar problems from different directions, and building on a shared statistical theory. Keeping both groups apprised of the data, studies, experiments, processes, and tools used by the other is key.
Creative research scientists will find opportunities to apply cross-functional techniques, generating real insight to shape the business going forward. The efficacy with which companies merge data science and market research groups will be a significant factor in their long-term competitive advantage, as those companies who build cross-functional teams that are able to use all of the available tools and techniques will have a stronger understanding of and impact on consumer behavior.
Research Science as an Opportunity for Vendors
This fusion of data science and market research represents a major opportunity for data science/analytics and market research vendors both. Vendors who fuse the underlying techniques and leverage tools from both worlds will be better able to provide value to their increasingly cross-functional clients.
For many companies, this will mean investing in skills and technologies with which they have limited experience. This may mean:
- integrating with new technologies (e.g. field-and-tab organizations integrating their output into modern visualization tools),
- developing new capabilities (e.g. analytics companies investing in psychographic profiling and attribution), or;
- acquiring skills or technology (e.g. acquiring and integrating neuroscientific data into their portfolios).
Build-versus-buy decisions will be difficult in the coming years, but the race for competitive relevance is already accelerating. As clients integrate data science and market research teams, vendors who provide support through this transition will win more business than vendors fluent in only one arena.
Because one thing that both market researchers and data scientists can always agree on is this: Data describes reality, and understanding reality is the first step to changing it.