Selecting the Right Attribution Model for Inbound Marketing

  Analytics, Attribution Modelling   19 Comments

 

Ok guys I am finally back with a post on web analytics. Those of you who have been requesting an analytics post from me may find this post to be the most informative of all the analytics posts I have written so far. I am going to talk about attribution models today and try to break the myth surrounding the widely used last touch attribution and more recently the ‘first touch’ attribution model. I am also going to introduce you to ‘Proportional Multi Touch Attribution Model’.

 

What is attribution modeling in Web Analytics?

Attribution modeling is the process of understanding and assigning credit to the acquisition channels which eventually leads to conversions (goal completion).  Acquisition channels can be paid search, organic search, email marketing, direct traffic, referral traffic, display advertising etc.  The purpose of attribution modeling is to understand the buying behavior of your website visitors. Why people buy from your website? What happens before they buy? What prompted them to make a purchase or complete a predefined goal? Which are the most effective acquisition channels for investment. An attribution model give answers to such questions.

 

What is Last Touch Attribution Modeling and why it is flawed?

In the last touch attribution modeling, conversions are credited to the most recent acquisition channel. For e.g let us consider the following conversion path:

 attribution modelling

 

1.  A person say ‘Himanshu’ reads a blog post on your website

2.  After 3 days he saw your display ad on a website

3.  After 2 days he reads a review of your product on some website.

4. After 4 days he decided to make a purchase. So he searched using a non branded keyword and clicked on your PPC ad on Google.

5.  Just to make sure that he is going to get the best deal, he went to a product comparison site. Being satisfied with your product pricing he decided to make a purchase during weekend.

6. During weekend he searched on Google using a branded keyword, clicked on your organic search listing.

7. He made a purchase from your website.

 

Here Himanshu was exposed to multiple acquisition channels (blog post, display ad, product review, paid search etc). Each of these exposures is considered as touch.  So Himanshu was exposed to 6 different acquisition channels before he made a purchase.

Now according to last touch attribution model, the conversion (making a purchase) is attributed to organic search.  Most analytics software by default use ‘last touch’ attribution so they will also report to you that a person searched for your website on Google through a branded keyword and then made a purchase.  So acquisition channel responsible for your sale is ‘organic search’.  As you can see from the chart above, this is not true. 6 acquisition channels have played an important role in the conversion on your website.

 

What is First Touch Attribution Modeling and why it is flawed?

According to first touch attribution model, the conversion is attributed to the first touch (in our case blog post). The visitor to your website read your blog post and then he made a purchase decision on that basis. This is also not true. As the visitor also saw your display ad, read a review of your product, clicked on your PPC ad, visited a product comparison site and click on your organic listing before making a purchase. All these acquisition channels influenced the purchase behaviour.  Just as a last touch attribution can lead to mis-allocation of resources, over crediting first touch can mislead as well.

Neither first touch nor last touch provides a good understanding of conversions.

 

What is Multi Touch Attribution Modeling?

In multi touch attribution modeling, the conversion is attributed to multiple acquisition channels instead of just the first touch or last touch attribution.  Here the middle touches also come into picture. This model aligns well the real life situations as people rarely make a purchase through one or two acquisition channels. For e.g. it is highly unlikely for someone to read your blog post and then make a purchase later via organic search. Similarly it is highly unlikely for someone to see your display ad on a website and then later make a purchase via paid search.  He may read reviews of your product or go to product comparison site before making a purchase.  So we need to take all of the touches into account.

 

Introducing Proportional Multi touch Attribution Modeling

Now the problem with multi channel attribution modeling is assigning value to multiple touches.  Not all acquisition channels are equally valuable. For example, in the example above, Himanshu reads product review and went to a product comparison site before making a purchase. These two touches seem more valuable than the exposure to the blog post, display ad and the PPC ad as they play a very important role in the purchase decision. Had Himanshu not satisfied with the product review or pricing, he wouldn’t have made a purchase.  Consequently these touches should be given more credit.

 

Here the proportional multi touch attribution model comes into the picture. In this model values are assigned to touches in proportion to their contribution in a conversion.  The acquisition channel which assists the most gets the maximum value and maximum resources are allocated to it regardless of it being a first touch, last touch or middle touch.

 

For example:

In the chart above, if clicks on the paid search ad have helped the most in conversions over a period of time say 30 days, then ‘paid search’ should be allotted maximum resources even when it is neither the first touch or last touch.  Similarly if viewing of a display ad has helped the most in conversions, then ‘display ad’ should be allotted maximum resources regardless of being the first, middle or last touch. All other touches should be allotted resources in proportion to their attribution.

 

I am calling all the analytics n analytics-ninjangas out there to share their views on this new attribution model. If you like this post then you should subscribe to my blog and follow me on twitter.

 

 

Other Posts you may find interesting:

 

 

Himanshu Sharma About the Author: is the founder of seotakeaways.com which provides SEO Consulting, PPC Management and Analytics Consulting services to medium and large size businesses. He holds a bachelors degree in ‘Internet Science’, is a member of 'Digital Analytics Association', a Google Analytics Certified Individual and a Certified Web Analyst. He is also the founder of EventEducation.com and EventPlanningForum.net.

My business thrives on referrals, so I really appreciate recommendations to people who would benefit from my help. Please feel free to endorse/forward my LinkedIn Profile to your clients, colleagues, friends and others you feel would benefit from SEO, PPC or Web Analytics.

 

 

  • http://www.irisholidays.com Iris Holidays

    Good thought. How do you execute Multi Touch Attribution Modeling with GA?

    • http://www.seotakeaways.com/ Himanshu

      Through ‘Multi Channel Funnel’ feature in Google Analytics.

  • AnalyticsNerd

    Somehow i completely missed this post. What can i say Himanshu. It is an awesome write up. I really like your new attribution model. Holistic and discrete analysis of mutiple touch points is essential in gaining insights about our visitors and your model seems more promising than the multi touch model itself. I always have doubt about the validity of the first touch and last touch attribution modeling. Thanks for confirming their flaws. The days of untargeted marketing efforts are over.

  • JasonGASA

    Great analysis of multi touch attribution modelling. Many marketers are still using last click attribution let alone first touch, multi touch or proportional multi touch attribution. Today it is imperative to look beyond first, middle and last touches to find hidden sales opportuinites and to fully optimize marketing initiatives. Himanshu your post seems ahead of time. I think it will take a long time to adopt multi touch model before even considering proportional multi touch modelling. One request. I would really love to see a post on actually developing an attribution model for multiple channels using Google Analytics. Many thanks for all the information you are putting on your blog.

  • Joanie

    I like your article on attribution modelling. But the challenge is how to understand it and use it in a way which produce tangible results for our clients. The attribution model used by Google Analytics is based on the most recent session. Same is the case with Google Adwords. Both analytics and adwords give an incomplete and very biased picture of visitors’ buying behaviour. So clearly last touch attribution is flawed. I agree with you on multi touch attribution. Proportional multi touch is something new for me and i will ponder on it. Thank you for the wonderful post.

  • Jitender

    Superb post Himanshu. Thanks for constantly reminding me that i am not using Google Analytics as well as i should be ;)

  • lukewarm

    Multi touch attribution is one of the most serious problem in internet marketing today mainly because of the lack of statistical framework. Where other attribution models have failed yours may work. Thanks for sharing your thoughts. I was reffered to your blog by my colleague. I also like your post on ‘web analytics 2.0′. Keep up the Good work.

  • http://www.theinsideman.co.uk/ MarwickSEO

    Thank you for your post. We have used some of the ideas from this post and put together a multi touch attribution model for our client’s site.

  • newGeek

    Brilliant post as usual. I agree with Jitender here. I am clearly not utilizing GA properly. Thanks for sharing your new model. I have been waiting for posts like this for ages. I am currently analyzing the impact of PPC and Display advertising on our SEO efforts and your model may come handy in allocating resources to the right channels. I would love a follow up for this post on how to implement this model in Google Analytics.

  • Brian

    So in your first example, how did Himanshu find the blog post? :)

    I’ve been considering a multiple variable attribution model for a big e-comm site, but it’s difficult to determine the weight each channel or touch point should have. It’s hard to know what’s best for the business because I don’t have all the data (like you mentioned), so my scope is limited and to be honest, if I implemented something today it would essentially be a glorified guess.

    • http://www.seotakeaways.com/ Himanshu

      For the sake of simplicity Google Analytics keeps the conversion path length to 11 and then put all the longer paths in the 12+ bucket. Conversion path length is the number of marketing channels a visitor is exposed to before he made a purchase. So what Himanshu did to find the blog post is not as relevant as what Himanshu did after reading the blog post and up to the point of making a purchase. There can be ‘N’ conversion paths, that is why Google Analytics has the top conversion paths report. You need to learn to segment the data. Most of the data is already their in your analytics just waiting to be segmented. Focus on Assisted conversions. Often campaigns which don’t directly result in conversions like ‘display advertising’ or social media are assisting the conversions in some way. Determine the assisted conversion value of these campaigns. If they are not even assisting then stop investing money in them.

  • Brian

    Thanks for the detailed response. You made some good points.

    I fundamentally disagree with “So what Himanshu did to find the blog post is not as relevant as what Himanshu did after reading the blog post…” however. Had he not found the blog post originally, he might have never learned about the brand/site at all, never clicked on the display ad, and thus never completing any sort of conversion.

    I understand what you’re saying about some actions being more significant to the conversion than others, and I agree with that completely, but I strongly believe that the way a person originally finds the site is definitely an important step in the process.

    That’s what makes the proportional multi-touch model difficult to implement. The importance of individual touch points is subjective and hard to quantify.

    • http://www.seotakeaways.com/ Himanshu

      Hi Brian! I think i didn’t make myself clear. I know what you are looking for, the very first touch when Himanshu was first exposed to one of the marketing channel. But this very first touch is important only when it pushes conversion/e-commerce transaction within the normal ‘time to purchase’. There is a possibility that the very first touch happened several months/years before the actual conversion. For example i became a SEOmoz pro member after reading their blog for almost 2 years. In this case my very first exposure to seomoz is irrelevant for optimzing my conversion rate. I consider exposure to that marketing channel as the first touch where majority of my clients start entering into conversion funnel. So first touch is where majority of my clients enter into a conversion funnel and last touch is where majority of my clients are exposed to a channel before they make a purchase. I agree with you that multi touch attribution is difficult to implement. But we have to start somewhere to solve this attribution problems. I don’t claim that my model has resolved all the attribution problems in web analytics. This model is an attempt to get one step closer to solving this mighty issue. Moreover it is incrementally better as outlined by ‘Avinash’ himself :)

  • Rob

    There is a major flaw with any attribution model (with perhaps the exception of Last Click which has its own flaws that you’ve outlined above) Any cross-session model relies on the integrity of ‘unique visitor’ which is currently outside the technical capability of most analytics packages as a wider spectrum of devices are increasingly used by individual users. For example, if Himanshu has an iPad, iPhone, laptop at work, laptop at home and a smart TV, with each step to purchase occurring on a separate device, the attribution path will be completely fragmented and practically useless. This is an extreme example, however even across 2 or 3 devices, the attributed values start to look very different. I think we need to begin looking at device attribution as an additional dimension to channel attribution…

    • seohimanshu

      yes i agree with you. Great insight. Thanks.

  • Liviu T.

    Hi, what is the difference between this model and the LINIEAR model? In the linear model if we have 2 touchpoints from paid you will attribute twice the amount to that channel. For me it’s the same as linear model.

    • seohimanshu

      Linear model assign equal credit to each interaction in a conversion path.Whereas Proportional Multi Touch attribution model assigns credit to interactions in proportion to their contribution in a conversion. So the interaction/exposure which contributes more get more credit.

      • Liviu T.

        Let assume the following funnel: Organic -> Direct -> Paid -> Referral -> Direct > 1000 usd.

        The conversion value is 1000 USD.

        For linear model we have:
        Organic = 200
        Direct = 400
        Paid = 200
        Referral = 200

        Please attribute a value to each channel for the model you propose.
        Thx.

        • seohimanshu

          Hi Liviu! I think you are mistaken about how linear model works. In case of linear model attribution each interaction gets equal credit. So it can’t be organic=200 and Direct = 400. The model which i have proposed assigns credit to interactions in proportion to their contribution in a conversion. Now in order to actually assign a credit, you must first know the role played by the interaction in a conversion. So for example, if Display ad has played biggest role in getting conversion then Display ad will get the maximum credit. If email has played the second biggest role in getting conversion then email will get the most credit after display. We just can’t put a value to any channel without understanding the role it has played in a conversion. I hope it helps.