Well, that makes things interesting. Here is a very cool curve ball pitched to the market research and publishing industries; basically the Adsense model applied to market research questions:
They are pitching their micro-survey paywall method as costing about 1/10th of standard online research, but I’m not sure the two are that easily comparable. Although there is the potential for everyone to win here (except readers who hate interruptions ;), I have some pretty big reservations about the potential quality issues. At first glance I can only see this being useful to those who want to quickly ask one or two questions to a large, untargeted group.
They say they can do targeting, but this will be based on the estimated audience profile of the content publishers they are working with along with cookie info and IP data, rather than information held at an individual level. Also, their post-stratification is heavily reliant on a large-scale third-party survey of internet use. That will likely be problematic if you are somewhere (like NZ) that doesn’t have a good info base for stratification or IP ranges that don’t lend well to geo-targeting.
And they say they can find correlations between questions even when they only ask people one question at a time. To do that they will either have to do data fusion (potentially dodgy, especially if it is on top of already estimated demographic info) or ask subsets of people multiple questions which they then use for correlation analysis (possibly as a base for data fusion). Either way they’ll be increasing the error around any correlations identified.
Google is teeming with smart people, and they have done some pretty good benchmarking to prove that it is possible to get results that are at least as accurate as online panel-based estimates, so they are no doubt taking at least some of those issues into account in their calculations. And a big tick has to go in their box for being open about the limitations of their method. However, this will essentially be a self-service offering, just like Adsense. There won’t be a quant-head available to provide advice on the potential pitfalls of relying on the information and I suspect most end-users won’t understand the types and sources of error in the estimates that pop out.
That said, many market research providers probably don’t give great advice to their end users anyway!
My take is that this service will be more of a competitor to crowd sourcing offerings (e.g., Amazon Mechanical Turk, Crowdflower) than online market research panel providers, particularly for those wanting to do some ‘quick and simple’ surveys. That segment is still a large growth market, so either way Google and content publishers win.
I’ll definitely be taking it for a spin :)
(here is an earlier post of mine relating to this)