Building a Social Spine in Tracking Research
Larry Friedman, TNS
- turn tracking research from a rear-view mirror into something that is windshield oriented, looks to the future
- historically tracking is survey research
- lots of information is observable, passive, that we didn’t have available before – social data
- we need to use obtrusive (surveys) and unobtrusive research together, MR rarely uses unobtrusive methods
- The Trouble With Tracking – insights not actionable, flat-line metrics, far from real time, long vague surveys
- is social media data a real alternative? real time, sensitive to events real consumers talking to each other in their own words at their own pace
- has social worked? hasn’t made the case yet, mostly because there is a lot of discussion and it’s not always what we’re interested in, hard to find the signal over all the noise. hard to understand what the metrics really mean. people want a one-to-one metric with external measures as in how we compare survey metrics
- what does the number of negative mentions in social really mean? how does it compare and to what?
- some worry that social isn’t representative because not everyone is online, not everyone uses these tools [and everyone participate in surveys??]
- the real issue however is predictability. even if it’s not representative, if it predicts it’s got value [heck yeah! first time I’ve ever heard someone other than myself shout this out]
- should social data look the same as survey data? why should it? shouldn’t it really be different?
- great success in predicting brand health by putting social and search data in the model
- data cleaning is extremely essential, discussion must come from the correct geography – avoid english conversations around the world, avoid the coupon and sales bots, ensure you are getting apple computers not apple pie
- social lets your track your brand equity today instead of 3 weeks from now when something major could have happened and been dealt with
- make the social equity score a leading indicator, plan for several weeks ahead, prevent problems sooner
- most social analyses are mass market analyses, we can’t throw away surveys just yet. surveys are needed for deep dives
- models need to be updated more frequently – pinterest and vine are brand new and suddenly huge, social sites will come and go very quickly
- google modeled trends in flu reporting, people search for flu symptoms online – model could anticipate trends in flu that CDC was reporting, model worked well for a couple years. model is no longer active perhaps because google changed how it does it’s search terms, changing their own data changed their own model
- old tracking model was a project on it’s own, you discussed not changing the tracker so as not to change the trend.
- new tracking model is an integrated program from a variety of sources to give a forward look into the market place
- need to consider what consumers say (survey), think (social), and do (POS)
- able to predict car registrations by using all forms of data together
- difficult part – we see a trend that is decreasing and so take action to improve it… and then the trend increases. but was your prediction accurate? did you fix the problem or did the trend just increase as it was going to anyways?
- why do you use tracking surveys if you make no changes to affect those ratings? why bother tracking numbers that never change [fabulous point]
- we need to learn more about which and when spikes matter