I’ve had plenty of consulting assignments via the years with a objective of enhancing a consumer’s model tracker.  You already know the story…nothing strikes and when one thing does transfer, the consumer doesn’t know what to do with it. And each a kind of disappointing trackers was lacking a important query…fixed sum.

The fixed sum query asks respondents to allocate 10 factors throughout manufacturers they’d think about on their subsequent buy.  They can provide all 10 factors to 1 model if that’s the solely model they’d purchase, or 0 factors to a model they positively wouldn’t purchase…or any sample in between…so long as the factors add to 10 throughout all manufacturers.

Validity

In my expertise from dozens of trackers and a whole lot of manufacturers, this query returns extremely predictive consumer stage knowledge. Lately, I used it on a model fairness examine for a monetary companies model the place we had precise account opening knowledge merged in.  The patterns have been extremely confirmatory of the worth of the query (e.g. near 0 conversions from these giving 0 or 1 level, and nearer to 10% account opening charges for these giving a excessive variety of factors.

Bias removing throughout nations

A lot of you understand that buy intent and internet promoter scores are extremely affected by tradition.  Prime field scores are a lot greater in French and Spanish cultures for instance with out implying extra trial.  NPS is ineffective in Japan the place scores are all the time actually low, once more with out implying your enterprise is about to implode. Not so with fixed sum.

The truth that the respondent is making decisions and sacrifices (e.g. they may haven’t any factors left for a model they like if they provide all of the factors to another manufacturers) makes the patterns unaffected by tradition.  However, a respondent might give prime (or backside) field PI solutions to each model they’re requested about, in the event that they select to.

Helpful

Some of the vital elements is that fixed sum is absolutely helpful.  Those that give between 2-8 factors to a given model are the Movable Center and research have proven that they’re 5 instances (and even greater) extra attentive to promoting than non-Movable Middles. So in an addressable media world, take the IDs which can be within the Movable Center that you’ve got gathered over waves of monitoring and on-board them as a seed pattern to your ID/gadget spine at scale (media company or DSP could be the keepers of this.) Utilizing lookalike modeling, you may create a targetable viewers at scale of Movable Middles and this may result in a 50% enchancment in promoting ROI.

One other helpful facet is that it reveals who you instantly compete with.  Within the monetary companies instance, it was actually clear that the net banks have been in additional direct competitors, the credit score card-based enterprise have been one other phase, and many others.  So, who’re your direct opponents?  The covariance patterns (e.g. if one model will get excessive factors from sure respondents, one other model tends to additionally get excessive factors…) let you know.

Last trick of the commerce…the fixed sum knowledge and attribute scores are normally coherent for given respondents…however not all the time.  For instance, those that give 5 or extra factors to a sure model are inclined to price it very extremely throughout attributes and maybe most curiously, they do NOT price different manufacturers extremely…they discover the model distinctive.

Now what is absolutely attention-grabbing is when the attribute scores defy that sample for sure respondents.  Those that gave you loads of factors however don’t price you extremely on attributes are VULNERABLES.  Those that price you extremely however the place you bought few factors are PROSPECTS.  My most cited paper (cited over 1200 instances in line with Google scholar) confirmed that such discrepant patterns have been extremely predictive of people’ model decisions one yr later. And now you will have a brand new, highly effective, predictive model fairness metric.

My recommendation…use fixed sum in your model analysis. It simply could be the mic drop you might be in search of.