Ahead Of The Cross-Channel Curve

Marketers are struggling to get a handle on cross-channel attribution, judging by several efforts now in progress.

For example, the Data & Marketing Association (DMA) has created a sample RFI and glossary, defining every term from “match rate” to “truth set.” And the Coalition for Innovative Media Measurement (CIMM) will release a white paper, “Best Practices in Multi-touch Attribution and ROI Analyses” during its Cross-Platform Media Measurement & Data Summit in New York on Feb. 16.

Now being developed with Sequent Partners, that paper will focus on attribution modeling, marketing-mix models and what CIMM calls “single-source analyses based on exposure and purchase data matched at the household level.”

Why this attention now? The purpose is to “bring some consistency and clarity,” to the wild-west environment that now prevails in the field, said Jane Clark, the CEO and managing director of CIMM, during a recent briefing.



One problem is that vendors and marketers use different language to describe the discipline. “We came to the conclusion that it’s still very early days,” Clarke said.

For this reason, marketers may not be getting the precise solutions they need.

Comparisons are difficult. In general, though, older data companies “are taking a bit more care in making sure they are doing it correctly,” Clarke said. Some newer ones may be developing probalistic algorithms that “can’t possibly be validated.”

She added that “even companies that you think are 100% deterministic are not really 100%.” 

At the same time, many companies in this space are “buying each other.”

Then there are the “macro-level” data issues. In a recent paper titled “Best Practices in Cross-Device and Cross-Channel Identity Measurement,” CIMM defined these as:

  • Incomplete population representation
  • Small sample size
  • Data integration issues
  • Variance in data quality

MediaMath’s Ari Buchalter concluded in the paper that “Some of the data are … just perpetually inaccurate. People move. They create new log-ins. There’s a percentage of those data that are always inaccurate and out-of-date.”

Where does that leave the end-user marketer? Asking the following questions: “Where are you sourcing the data, how much is black box, can you figure out all sources, how often they’re refreshed and updated, and how solid and reliable they are? You can’t take it at face value,” Clarke said.

CIMM is also trying to “bring more granular measurement to television,” Clarke added.  The question is: “Can you use these census-based measurements to get away from panels, and be able to link people across devices and offline and online media?”

Finally, cross-channel marketers have to cope with the privacy issue. Clarke cautioned marketers that “you’re as responsible as the vendor if you overstep the bounds into something that’s not a good use of the data.”

In its recent paper, CIMM said that “accuracy in the context of cross-device identification, is typically calculated as the number of matches and non-matches correctly identified, or the number of times a deterministic or probabilistic prediction was correct.”

However, CIMM added that “accuracy scores vary based on approach, and how this measure is calculated varies from solution to solution. As a rule, most accuracy scores do factor in non-matching predictions to calculate accuracy.”

One vendor commented that “the fact that numbers are being put out there that, in practice, count useful and non-useful matches as being equally valued is not helpful.”

 CIMM advises marketers to:

  • Focus on measuring precision
  • Understand the truth set that is being deployed
  • Align methods with marketing goals
  • Work with solution providers

The February summit will also feature updates on global cross-platform measurement, kids and teens, multi-platform TV and video.

Correction: Coca-Cola will be represented on a panel at the CIMM summit but will not be presenting research, as an earlier version of this column stated.

4 comments about "Ahead Of The Cross-Channel Curve".
Check to receive email when comments are posted.
  1. dorothy higgins from Mediabrands WW, January 23, 2017 at 12:15 p.m.

    An enormous glitch in the works is that in buying inventory through ad networks and audience aggregators the client does not know how many times any individual (or population) has been exposed.  Our ability to parse a frequency distribution for digital is hampered by inconsistent "tranlarency".  While some may decry these as traditional metrics the statisticsl reality is we cannot even think about attribution without a tighter read on exactly how many exposures each individual receives over defined time periods. 

  2. Jared Frank from KHQ, January 23, 2017 at 4:42 p.m.

    dorothy higgins, perfect attribution is most definitely flawed. However, to what can digital attribution be compared? The gold standard has been TV, but even Nielsen's measures are seriously lacking. Why are we comfortable taking broadcasters' measurements as gospel, but squimish when we look at a report from an ad network?

    In the end, the author is correct: be very careful about the data you use to target your audience. Cross-channel marketing still has a long way to go, mostly because people's media consumption is fractured if not platonic.

    Or am I missing the point?

  3. Ed Papazian from Media Dynamics Inc, January 23, 2017 at 4:59 p.m.

    Jared, we don't take broadcasters' ratings as TV buying currency, we take ratings from an independent company---Nielsen----which has always made it a point to avoid any conscious effort to decieve advertisers or ad agencies and certainly not to favor one seller over another. Moreover, TV has none---repeat, NONE---of the ad viewability/fraud issues that plague digital media. Add to that the fact that Nielsen's national rating panel covers all commercial in-home viewing channels;consequently, with TV it is possible to analyze reach attainment across channels---something that no individual digital ad seller can equal. Dorothy is quite right. There is a huge difference between what is used for TV relative to seller-supplied "audience" tallies for digital. This has got to change before one can even talk seriously about attribution, "audience targeting", etc. on  a serious and non-theoretical basis.

  4. John Grono from GAP Research replied, January 24, 2017 at 3:40 a.m.

    Ad to further the point, TV does not count by device. plus the majority of TV viewing the magnitude of temporal duplication is much smaller.

    But being a panel it automatically 'de-duplicates' when Little Johnny leaves the bedroom (but leaves the TV on) to come out and watch Game of Thrones in the family room. Then when he watches it again on the PVR it again de-duplicates - Little Johnny is only counted once.

    When you rely on cookies (even PID cookies) de-duplication is amazingly complex to account for, so getting the online audience right is a huge challenge.

    And by the way, did I mention that then de-duplicating the TV and online audiences is even harder? We're working on it and v. hopeful.

Next story loading loading..