Big data now represents the next step in improving television program selection and targeting precision. Merge, append, fuse and model are all terms that can bring increased precision for reaching
high value prospects. These methods are discussed below:
Data MergeThis is is a great way to understand the television viewing preferences of current best
customers. A customer database, such as a file containing frequent shopper data (first-party data) is merged on an exact name and address basis with set-top box household viewing data. The
new enhanced data set provides a basis for understanding differences in between-segment viewing preferences -- for example, active customers vs. high-value customers vs. purchasers of particular types
of merchandise vs. lapsed customers who we need to re-activate.
To avoid privacy concerns, the customer data and the set-top-box data are merged by an independent third-party supplier.
Following the data match, personally identifiable information is removed. The enhanced data set is then returned to the owner of the set-top box data, who works with the advertiser to access and
analyze the resulting merged data set.
A shortcoming of merging data sets can be low match rates. If there is a 30% match -- for example, the people in the advertiser’s loyalty card
data file match 30% of the households in the set-top-box panel -- we are then limited to using 30% of the set-top box viewing data. Nevertheless, a set-top-box panel of one million and a 30%
match rate would provide viewing behavior for 300,000 households. We just have to trust that there is no systematic bias in the matched vs. non-matched records.
Data
AppendSimilar to a data merge, here a data set containing records from a third-party data set is matched on a household or persons basis with set-top-box household viewing data.
The major third-party data sets include Acxiom or Experian for demographic characteristics data, Polk for car ownership data, American Express and MasterCard for retail purchase data, Kantar Shopcom
or dunnhumby for frequent shopper loyalty card data. In addition to these solutions, Nielsen offers a broad range of syndicated solutions utilizing credit card data, car purchase data and similar
solutions.
Match rates between set-top-box data files and third-party data files tend to exceed levels achievable through a first-party merge. As a result, appended data sets tend to be
larger than merged data sets. The benefits of having larger data sets include the ability to conduct more granular analyses of television viewing behaviors: for example, your best customers vs.
customer segments that you do not currently serve.
Data FusionData fusion is used in cases where neither a data merge nor a data append is possible, when the files
that we’d like to merge or append have very few members in common. For example, one file contains persons and their internet behaviors and a second file contains persons and their television
viewing behaviors. There are very few instances of a person/household appearing in both files. We need to understand the Internet and viewing behaviors of a particular group of households or
persons.
The rationale
behind fusion is the “birds of a feather” idea. Through the fusion process, records are matched on the basis of household/persons characteristics and behaviors, also known as
“hooks” that link back to television viewing behavior.
The set-top-box file would be considered the “recipient” file, and the file to be fused into the set-top-box
file would be considered the “donor” file. A “donor” household might be a husband-wife household where both heads of household have similar levels of education, household
income, children between the ages of 12 and 17, and live in a suburban community. The “recipient” set-top-box household would have the identical set of characteristics so that records in
the file can be linked. At that point, the fused data would be used to “fill-in” the missing characteristics not included in the recipient file.
Over the last five years or so,
data fusion has become increasingly common in the U.S. Examples of syndicated fusions include the Nielsen’s National TV / Online Fusion, Nielsen’s National TV/ MRI Fusion, the
comScore / MRI fusion.
Data ModelingData modeling is similar in many ways to credit scoring, in which credit card companies “grade” individuals to
determine credit-worthiness. The marketing application is about identifying attractive target prospects based on past purchase history, demographic characteristics and other predictors.
The analysis provides a means of determining the viewing preferences of different groups: for example, the viewing preferences of the correctly predicted highly attractive prospects vs. the viewing
preferences of those predicted as being attractive who have not yet purchased. Actual viewing preferences may differ significantly.
So, we have discussed the process of merging, appending,
fusing and modeling consumer characteristics and behaviors into the data sets that are used to profile television viewing preferences. Each case will vary depending on data availability and
affordability. As advanced as all of this sounds, we are just now beginning to tap into the power of big data
advertisement
advertisement
.
Joe, this is an excellent overview of a timely and complex issue. I'd be happier, though, if you delete the word "precision" from the title.
All the various techniques of utilizing big data do enhance our understanding of people's behavior and thus improve the quality of buys to deliver more qualified audiences... None however add precision.
I agree with Gabe. While a big fan of such techniques (we've been using them for around 20 years here in Australia) marketers shouldn't be seduced by the concept of more precise targeting. Targeting precision is all about (cost) efficiency. The flip side of the coin is marketing effectiveness. Grab a book called How Brands Grow by Dr. Byron Sharp of the Ehrenberg-Bass Institute here in Adelaide. It will challenge lots of modern marketing beliefs and with empirical evidence show that what some marketers call 'waste' (through the lens of efficiency) some others see that as 'growth opportunity' (through the lens of 'effectiveness'. In essence - to grow you need to get you brand in front of people who had no idea they would like it.
Thanks for your comments. We're always interested in delivering the right message to the right prospect - this includes the idea of segmented messaging. We think the idea of targeting precision applies to both individual television properties as well as to the portfolio that we purchase to reach a particular target prospect. If, in the process of developing targets, we're able to identify a group that represents significant business potential, it’s important to first understand how to unlock that opportunity.
While I buy into the "big data" idea as a way to refine targeting in ways that are nearly impossible using standard TV rating panels with their small samples, the danger lies in the nature of the data we are talking about. So far, big data, as it applies to TV, relies mainly on set usage indicators----in other words "household ratings". There are no measurements showing who in the home is viewing. While this may seem like a detail to some----it isn't. If you study the Nielsen findings you will see that homes with younger HH heads and upper incomes are TV's biggest set users, while older homes, mainly in the lower income brackets tend to be lighter set users. Why? Younger/upscale homes have more TV sets and more residents to use them----including youngsters. Older homes are usually without children and have only one or two residents. As a result, when an older home does tune in a program, the chance that one or both adults are watching is very high. But when a younger/affluent home tunes in the desired "target adult" may not be watching at all---or even be at home.
Solving the problem of identifying the viewer, not merely the home, as being "reached" is going to be a tricky one. Perhaps statistical attribution offers a partial solution but, so far, I've seen no indication that this issue has been adequately resolved.
Thanks Ed. We have found the data and methods discussed above to be useful supplements to Nielsen Ratings in cases where the groups that we are attempting to reach have a low incidence and limited representation with the Nielsen panel. This is especially important in the evaluation of buying opportunities within long-tail (low-rated) cable networks. Finally, this is the only option for the 100 or so cable networks that currently have no persons-level ratings measurement.