Loading ...
Sorry, an error occurred while loading the content.

24636RE: [webanalytics] Conversion Attribution

Expand Messages
  • Andy Fisher
    Dec 29, 2009
      >Of course, no tool can do any kind of data correlation, including
      >deduplication, involving data that it doesn't have access to.

      True - but in the Atlas, DCLK and the Tagman case, the tool can decide which other tool records the conversion event. So if you have two affiliate networks, you only fire the affiliate network tag when you think they should get credit. The idea is to make the affiliate counts the same (theoretically) in the affiliate tool as well as the ad server for that particular affiliate. This is important because, in most CPA deals, you pay for the conversions that the affiliate claims.

      >If we assume that every referral and conversion event is reported to
      >Webtrends (as an example), which isn't a difficult assumption to turn
      >into reality, then yes, Webtrends Analytics could report on how many
      >conversions are attributed to each referral source in such a way that
      >each conversion is attributed to only one referral source.

      Definitely true. However, the trick is making these numbers appear in the affiliate tools themselves (i.e. the numbers you pay on). Your solution is theoretically correct and easy to implement. It's just hard to use it for billing (ah - politics). Maybe some of the WA tools will go through their audits and actually be a system of record.

      It's also worth noting that that neither the ads servers nor the affiliate measurement tools typically use referring URL in their measurement systems. Rather they tag the ads, server the redirects tag the pages and, essentially using pathing analysis. This is what allows for view based conversions.

      >Provided you're okay with "last touch" attribution, which is all that's really
      >supported by the web analytics marketplace right now. This level is
      >pretty easy.

      Both Atlas and DCLK do an OK job of doing multiple touch attribution. Atlas has engagement mapping and DCLK lets you pull all of the preceding data for converters at the event level via the DCLK API. You can then play with it yourself or use a tool like Clearsaleing. Both Atlas and DCLK could use some improvements, but they are on their way there. To the best of my knowledge, Coremetrics is working on this too. Now there is a special problem here with affiliates... you can get the affiliate impression data into multi-touchpoint attribution systems on the ad server side but it is very hard to 1) get that data into the affiliate systems and 2) write contracts on that data. Best practice I've seen is to model out the data from a multi-touchpoint attribution approach and then scale the average change into your desired CPA/CPC/ hybrid. I'd love to hear other people's approach to this.

      Using the Visitor Datamart product to drill into individual
      visitor-level data, and a bit of JavaScript and/or cooperation with
      your affiliates to sync up your visitor IDs to theirs, you can drill
      into individual conversions, match that up with external data sources

      Absolutely (one sneaky thing you can do is treat every piece of content on your site as a conversion). Trick here is that it turns out to be maddeningly hard to join that the impression data at the event level. You get tons of mis-matches, errors, etc. You can do it manually, you can try things like Omniture genesis integration with DFA, but all of these things have issues. It's actually one of my big pet peeves with our industry. On the WA side of the game we are good with clicks (particularly natural search and inbound links). We are lousy at impressions. On the Ad server side of the business, we are good with impressions but lousy at natural search and inbound links. Maybe someone will make the two systems talk together... someday. Sigh. At least attempts are being made. Coremetrics, Genesis, etc...

      Great conversation. Happy holidays
      Andy

      PS: I don't work for any vendor, so that lets me be super cranky ;)


      From: webanalytics@yahoogroups.com [mailto:webanalytics@yahoogroups.com] On Behalf Of Wes Contreras
      Sent: Monday, December 28, 2009 2:52 PM
      To: webanalytics@yahoogroups.com
      Subject: Re: [webanalytics] Conversion Attribution



      On Fri, Dec 25, 2009 at 1:51 PM, Andy Fisher <andy.fisher@...<mailto:andy.fisher%40razorfish.com>> wrote:
      >
      > But I am under the impression that were are not talking about deduplication of data
      > within a tool such as webtrends. Rather we are talking about deduplication across
      > multiple platforms that will fire multiple tags on a conversion event. This is a big
      > deal because many different networks will take credit for a single conversion and
      > If you buy on CPA this is a big deal.
      >
      > So for example you have two affiliates. I don't think that webtrends (or any of the
      > other WA tools) can dedup conversions that are claimed by both affiliates. Altlas,
      > DCLK, Tagman and some other can do this (of course there are all sorts or special
      > case an exceptions). Please correct me here if I continue to be wrong about this.

      Of course, no tool can do any kind of data correlation, including
      deduplication, involving data that it doesn't have access to. If we
      assume that every referral and conversion event is reported to
      Webtrends (as an example), which isn't a difficult assumption to turn
      into reality, then yes, Webtrends Analytics could report on how many
      conversions are attributed to each referral source in such a way that
      each conversion is attributed to only one referral source. Provided
      you're okay with "last touch" attribution, which is all that's really
      supported by the web analytics marketplace right now. This level is
      pretty easy.

      Using the Visitor Datamart product to drill into individual
      visitor-level data, and a bit of JavaScript and/or cooperation with
      your affiliates to sync up your visitor IDs to theirs, you can drill
      into individual conversions, match that up with external data sources
      like Atlas and whatnot, and produce some really interesting reports.
      With this level of data, you could implement an attribution model
      other than "last touch," as well as combine data across multiple
      platforms and deduplicate or further analyze from there. Really, this
      is access to event-level data, so the only limits on what you can do
      are your skill with data manipulation and analysis, but this level
      takes more work to implement.

      Disclaimer: I am a Webtrends employee, but not speaking for Webtrends.
      This is personal opinion only.

      --
      Wes Contreras



      [Non-text portions of this message have been removed]
    • Show all 14 messages in this topic