I have had a number of consulting assignments through the years with a goal of improving a client’s brand tracker. You know the story…nothing moves and when something does move, the client doesn’t know what to do with it. And every one of those disappointing trackers was missing a critical question…constant sum.
The constant sum question asks respondents to allocate 10 points across brands they would consider on their next purchase. They can give all 10 points to one brand if that is the only brand they would buy, or 0 points to a brand they definitely would not buy…or any pattern in between…as long as the points add to 10 across all brands.
Validity
In my experience from dozens of trackers and hundreds of brands, this question returns highly predictive user level data. Recently, I used it on a brand equity study for a financial services brand where we had actual account opening data merged in. The patterns were highly confirmatory of the value of the question (e.g. close to 0 conversions from those giving 0 or 1 point, and closer to 10% account opening rates for those giving a high number of points.
Bias removal across countries
Many of you know that purchase intent and net promoter scores are highly affected by culture. Top box scores are much higher in French and Spanish cultures for example without implying more trial. NPS is useless in Japan where scores are always really low, again without implying your business is about to implode. Not so with constant sum.
The fact that the respondent is making choices and sacrifices (e.g. they will have no points left for a brand they like if they give all the points to some other brands) makes the patterns unaffected by culture. On the other hand, a respondent could give top (or bottom) box PI answers to every brand they are asked about, if they choose to.
Useful
One of the most important aspects is that constant sum is really useful. Those who give between 2-8 points to a given brand are the Movable Middle and studies have shown that they are 5 times (or even higher) more responsive to advertising than non-Movable Middles. So in an addressable media world, take the IDs that are in the Movable Middle that you have accumulated over waves of tracking and on-board them as a seed sample to your ID/device backbone at scale (media agency or DSP might be the keepers of this.) Using lookalike modeling, you can create a targetable audience at scale of Movable Middles and this can lead to a 50% improvement in advertising ROI.
Another useful aspect is that it reveals who you directly compete with. In the financial services example, it was really clear that the online banks were in more direct competition, the credit card-based business were another segment, etc. So, who are your direct competitors? The covariance patterns (e.g. if one brand gets high points from certain respondents, another brand tends to also get high points…) tell you.
Final trick of the trade…the constant sum data and attribute ratings are usually coherent for given respondents…but not always. For example, those who give 5 or more points to a certain brand tend to rate it very highly across attributes and perhaps most interestingly, they do NOT rate other brands highly…they find the brand unique.
Now what is really interesting is when the attribute ratings defy that pattern for certain respondents. Those who gave you a lot of points but do not rate you highly on attributes are VULNERABLES. Those who rate you highly but where you got few points are PROSPECTS. My most cited paper (cited over 1200 times according to Google scholar) showed that such discrepant patterns were highly predictive of individuals’ brand choices one year later. And now you have a new, powerful, predictive brand equity metric.
My advice…use constant sum in your brand research. It just might be the mic drop you are looking for.
