« previous next »
Pages: [1]

Author Topic: attribution modelling  (Read 5949 times)

Gurtie

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 1628
    • View Profile
attribution modelling
« on: May 11, 2012, 07:15:22 AM »
so - we've developed a fairly whizzy dashboard which pretty much manages to connect the dots together so we get full interaction paths to conversion, including connecting in offline conversions with online views for some clients. Still needs work, and will never be 100%, but its good enough to be useful.

next we enter a world of pain - attribution modelling.

add a complication - for the majority of clients, we don't do *all* online marketing - there is normally at least one outlier which means we have someone arguing that we're underweighting, overweighting or generally trying to cheat someone out of credit. We're not, so actually this won't change what we do, but it does mean we need to justify things with more than a gut feel argument.

and of course, internally, although we pretty much agree on the big points, we all have our own preferences. I guess its a bit like seo - more than one right way to do it.

Anyone been through this and reached a decent model? Or can point me to good research papers?
Logged

keano

  • Inner Core
  • Full Member
  • *
  • Posts: 211
    • View Profile
Re: attribution modelling
« Reply #1 on: May 11, 2012, 08:04:37 AM »
Econsultancy are normally pretty useful on this subject (particularly if you're a member!). There is a freebie report that you can get at http://econsultancy.com/uk/reports/marketing-attribution-valuing-the-customer-journey which may be useful in some way.
Logged

Gurtie

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 1628
    • View Profile
Re: attribution modelling
« Reply #2 on: May 11, 2012, 08:21:54 AM »
cool, thanks.  Now have to remember what the password is!
Logged

jetboy

  • Inner Core
  • Sr. Member
  • *
  • Posts: 433
  • Hens of warfare!
    • View Profile
    • Email
Re: attribution modelling
« Reply #3 on: May 12, 2012, 01:34:49 AM »
Avinash Kaushik has written about attribution models (http://www.kaushik.net/avinash/).

Personally, I've disowned last click and am going down the route of distributed attribution, heavily loaded towards first click. In part this has been led by tracking offline conversions, and needing to decide which tracking phone number to display (the first click that owns 50%+ of the sale dictates the phone number); in part a desire to put non-brand PPC and SEO in a good light (i.e. the channels which bring in *new* business); in part wanting a relatively simple model that can be explained in a couple of sentences. That rules old things like having a decay curve for touch-points.

There's no right way of doing this, and there's value in looking at the data in various different ways. However, for deciding marketing spend and reporting on effectiveness you've got to pick a model and stick with it. It dictates how you spend your marketing budget. Weighting things so heavily towards first click means that I'll probably never spend money on banner ad retargeting again. Measured on last click only, retargeting is a cost-effective channel.
Logged
Pages: [1]
« previous next »