2
$\begingroup$

I am a beginner at modelling. I have to build a multi channel attribution model that attributes revenue to various channels touched over a period of time.I want to know how suitable will Structural Equation Modelling be for my problem? I have been reading up on Structural Equation Modelling for the past few hours but I don't understand how do I fit my problem statement into the model?Any help of any kind will be appreciated.

$\endgroup$
4
  • $\begingroup$ This was my first Google result for "multi channel attribution modeling" kaushik.net/avinash/… $\endgroup$ Commented Jan 14, 2015 at 14:56
  • 1
    $\begingroup$ I know quite a lot about SEM, but I don't know what "multi channel attribution model that attributes revenue to various channels touched over a period of time" means, so I can't answer. $\endgroup$ Commented Jan 14, 2015 at 18:01
  • $\begingroup$ You can read about multi channel attribution here kaushik.net/avinash/… $\endgroup$
    – Nisha
    Commented Jan 14, 2015 at 19:55
  • 1
    $\begingroup$ Basically,visitor to a website, before making a purchase could have interacted with many channels.He could have clicked a paid link or received and email about it or anything.SO, the revenue of the purchase made by the visitor should not be credited to the last channel alone.All the channels(here email and paid link) which have brought the visitor to the website in previous visits also have a credit to share.So, what I am asking is, which modelling technique should I use to attribute revenue to various channels and can SEM be one of such modelling techniques? If yes, how? If no, why? $\endgroup$
    – Nisha
    Commented Jan 14, 2015 at 20:01

2 Answers 2

1
$\begingroup$

How about Markov chain process?

You could have for example two channels: add in the print and web banner where you have links to the product/service page in you companys web pages. In print it could be an QR code which is read through mobile phone.

Now you know transition probabilities and perhaps even shares of potential customers who will accept you offer. You can then easily attribute some part of the revenue into channels.

You can easily extend this setting into further channels if you have some knowledge about transitions between channels.

$\endgroup$
2
  • $\begingroup$ Agreed.Will read up on Markov chain.Any idea how SEM can be applied to this attribution exercise? $\endgroup$
    – Nisha
    Commented Jan 14, 2015 at 15:38
  • $\begingroup$ @Nisha transition probabilities can be parameters or variables in your model which has 2 equations if you have two channels etc.. $\endgroup$
    – Analyst
    Commented Jan 15, 2015 at 12:13
0
$\begingroup$

About Markov Chains:

First you will need an analytics to track clickstream data and store the users touch points across their journeys. Second, you will calculate the transition matrix, which is pretty simple to do with python

The transition matrix will describe something like this:markov_chain_ilustration

Where the states are the channels or campaigns.

With that you will have a probability for a user walk from start to the end, or in a marketing explanation: from the first touch point to the conversion.

Then you can calculate the contribution of each channel through the removal effect. To do this you need to remove the channel from the chain and measure the the change of the overall probability. This change will be exactly the weight of the contribution of the channel

There is a paper that you can explore, although some things are not explained in a explicit way. Paper

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.