Adding Research Value with Predictive Insights

Most insights research I would call descriptive. Studies like segmentation, attitude and usage, brand trackers, brand equity studies… all descriptive. Many marketing research departments are called insights because that big “aha” is thought to be worth a lot – after all, don’t you want to develop some amazing insight that no one else mentioned before? Feels nice to be the brightest one in the room.

So here is where I might make you uncomfortable. Insights without prediction cannot be proven or disproven and are worthless.

The irony is that the predictions are embedded in your insights, you just have to extract them, expose them to the light, and see if they are correctly leading to desired future states. If you don’t do it, marketing will and they are not as good at it as you should be. Did you just cost your company money by allowing that to happen?

 

Segmentation

Let’s take segmentation, which is great to reveal a sizable segment of consumers no one thought of before. But so what? Should marketing direct more ad spending to that segment, taking away spending from other segments? Will innovating to the needs of that segment result in a successful new product launch? Do you change your pricing resulting in desired share and profit changes? Will the segments lend themselves to lookalike modeling in your CDP data profiling?

Be prepared to say what the effects on business outcomes are likely to be from those decisions. I have two negative war stories about bad segmentation. 

1. One was a laundry detergent segmentation I inherited while at Unilever. With much fanfare at Unilever, they presented their results. As a narrative, it seemed great… Until I noticed that brands with little purchase switching interaction indexed highly in the same segment! If it failed that behavioral test, it would fail everywhere.
2. The second was a product innovation firm’s segmentation that led to a new product innovation. When we tested the idea, the purchase intent was the same for that segment vs. other consumers. If you can’t turn on the segment you designed a new product for, your segmentation is trash.

 

Brand tracking

Take brand tracking. Just about every tracker I have been asked to review from marketer clients is a retrospective readout, saying little or nothing about the future, and marketers stop looking at it because they don’t know what to do with it. When I led model creation for The NPD Group’s approach to brand tracking, we built a predictive engine into it.

We created a calculation called “strategic share”, which was the share your brand was entitled to, given its attribute ratings. The idea was that if actual share was below strategic share, your brand was poised for growth. We married that with ARIMA modeling and proved, with a fairly high degree of accuracy, we could predict the direction of next quarter’s tracking results.

 

Models of consumer behavior

Consider models of consumer behavior using either the NBD Dirichlet model or the Beta distribution.  These are inherently descriptive models of patterns of market share, repeat rate, and penetration levels. How do we make them predictive? Combine them with a logit model of conversion rates and apply some calculus and you will realize that those with 20-80% probabilities of buying your brand must be, mathematically, more responsive to advertising than, say, non-buyers with a low probability of purchase (the fallacy of reach-based media planning, but don’t get me started!).

So, in support of the MMA and in partnership with Neustar (and fueled with Numerator receipt scanning data), we went from a purely descriptive model to one that directs programmatic buying and targeting that can result in a 50% improvement in ROAS (return on ad spend). The white paper is here.

 

Conclusion

While most aha-s don’t have the value you, as a research technician probably think they have, the predictions embedded in them do!

Just go the distance… Be quantitative, rigorous, and accountable for the business impact of the actions your predictions lead to. If your insights are true and you go all the way to predictions that are proven to be correct, you will have a big impact.