So much of our marketing research is rooted in a descriptive insights culture. We segment consumers into supposedly meaningful groups that provoke thought. We use regression methods to determine attribute importance from brand research. We conduct A&Us to understand the “what and why” of usage.
We need to realize that our traditional focus is on explaining what has already happened. And this focus on descriptive analytics is falling short at driving action and providing much value in a digital, social, mobile age.
Lack of actionability is quite frustrating to both researchers and marketing. Evidence that traditional segmentation produces audiences that can be precision targeted via media is thin at best. Also, it is not unusual for tracked brand equity metrics to begin to diverge from share trends and we create stories as to why…insights often proven to be false by persistence of diverging trends. Derived importance modeling of attribute ratings against a dependent measure also can lead to an illusion of knowing that the brand will respond to a creative asset that stresses that attribute (there is even a fallacy I am happy to share…).
However, there is another path to follow…PREDICTIVE ANALYTICS…that is certain to be actionable. The world of precision ad targeting (a more general description of programmatic) is based on PREDICTION, which is proven to drive up ROI in a repeatable way and it is showing us a new analytics approach.
In real time bidding, the best known form of precision targeting, predictive algorithms use profiling variables at scale that are attached to users/cookies, or contextual factors. Think of a logistic regression model where the dependent variable is the desired outcome (e.g. make a purchase online, yes/no) and the predictors are everything we can use for targeting…but then filtered down to only using those variables that are statistically proven to add predictive power.
In this marketing world of competing on data rather than hunches, the researcher must begin placing priority on profiling brands and consumers using variables that are targetable. To do this, we need to conduct online surveys among those who can be matched to digital and social profiling variables on which ads can be bought.
So what should insights teams do differently?
- First and foremost, commit to linking survey results to digital profiling data on nearly everything you do. A number of online panel providers now offer the ability to link respondents to their profiles in Acxiom, eXelate, and/or Facebook (a different permissioning process). You can even connect survey results to frequent shopper data via Datalogix and other services. Sometimes, it is critical to link survey answers to actual clickstream behaviors on computers and smartphones and this is available as well. Finally, with an e-mail address, you can find 30-50% of respondents in social media. This gives the marketer a powerful ability to look at the social media conversation of a consumer segment of interest.
- Predictive insights come from new analytics applied to diverse data streams that address the prediction question. Use of somewhat different tools, such as logistic regression and ARIMA/ARIMAX will become more prominent.
- Build brand tracking from synthesizing all data streams that have predictive value. In December, WOMMA proved that social media and other word of mouth is predictive of sales. Now that you know that positive conversation going up is a driver of sales, how can you not track it and develop strategies to drive it up? I am certain that other behaviors such as trademark search, visits to website, etc. will also be powerful signals and drivers.
- Create forensic insights from ad serving results. When you profile out the responders, don’t just use survey variables, use digital profiling variables. You are likely to see a response profiling factor that is surprising and yet it is definitive…as you dig in more you are likely to get a powerful predictive “aha”.
- Track campaign effectiveness with prediction in mind. This means you will need to do some things differently. First, the turnaround to results needs to be lightning fast so you can move money around in flight based on what is working. Second you need to assess impression delivery against the intended audience. Did you really reach the segment of female outdoor enthusiasts you created or was the lookalike modeling not very good? Third, link campaign results to ad targeting profile variables so you can change targeting rules that improve business outcomes.
As the marketing research team desires to move from descriptive to predictive insights, how can they tell if insights are truly predictive? Here are two acid tests:
- Predictive insights are those quantified relationships from research that are shown to improve the prediction equations of ad response and sales trends. If an insight adds no predictive value, no R2, it might be interesting but ultimately not as valuable as it might otherwise have been had it addressed the prediction question.
- Predictive insights can arise forensically from ad serving response patterns. What are the persistent user characteristics of those who actually responded?
Predictive insights drive ROI improvement and profit growth that is otherwise hard to come by. Perhaps most importantly, prediction, rather than stopping at description, makes our insights actionable.