It has been a while now since Google Analytics made their multichannel module available (MFC, for Multi-Channel Funnel). Naturally, it was immediately declared the best thing since sliced bread, as with anything Google releases. Multi-channel analysis, and especially everything surrounding attribution models, has been all the rage in Online Marketing for some time now. With such diversity in channels in which marketers can invest now, those questions are absolutely relevant. Heck, if I can know with a good level of certainty where to spend my budget, and especially where not to spend it, I want to know!
Still, I am not totally comfortable with Google Analytics new module for various reasons. I will review a few of those points.
First, experienced digital analysts will know that any multi-session analysis is always very much at risk for a simple reason we’ve always lived with: cookie deletion. As long as we were focusing on the single session, you know, “come to the site, look at stuff, and covert or not”, we were OK. But now with the need to go full purchase cycle, over multiple sessions, through several possible sources, we are more than ever susceptible to cookie deletion impacts. I don’t want to start a debate about deletion rates; you probably know enough about your target audience to estimate how high it’s happening for you. Also, you need to factor in the typical purchase cycles; the longer it takes your customers to make up their mind, the more chances they’ve had to delete their cookie, oh and yes, come again via other computers, mobile devices, etc.
Right here we could call it quit.
Then there’s what Google Analytics declares as being the source of a visit. It has been rather puzzling, to say the least. By that I mean GA does not always assign a referrer according to what logic would require. I would assume that the source of a particular visit should be just that; what brought the visitor to the site this time. Still, GA will attribute a visit, say Organic Search Engine, if the visitor has that reference associated with his/her cookie, even though this time s/he came directly to the site. However, Justin Cutroni from Google tells me that it is not the same situation with MCF, that Google does something to attribute the visit to Direct. Good then, we just wish they would apply it to the regular analytics reports.
Then there is Organic Search. It’s been a known fact for five years now (and I extensively wrote about it then) that many, many people make use of Google search engine as a way to get back to sites they already been to. Look at yourself: do you type a site URL directly in the browser field theses days, or just type the company’s name in Google (which is probably in the launch page of your browser), knowing it will bring you there, even if you make a typo? It works so well, we don’t mind the extra click we need to get to the site. So, in the type of multi-session purchase cycle I discussed above, Organic Search found in the last steps are most probably people coming back, not searching. You should then filter for branded search terms to have a more accurate view of Organic Search contribution to the MFC. Generic, product category-related terms early in the cycle denotes true search, acquisition potential, whereas branded-term search visits toward the end of the purchase cycle are from people coming back with a higher probability to convert. However, such visits should not be credited with the conversion/revenue in an attribution model, I believe.
Finally, think of all the possible combinations in steps. I mean, look at this:
These are the top 10 patterns (or combinations) discovered by the tool in one instance. What kind of decisions can you really make? And I am not even adding even more complexity by bringing up the client segmentation question!
Sure, I guess we have to commend Google Analytics for at least trying. But, as of today, I am wondering if we are not just chasing ghosts.
And you? What do you think? Am I too negative?
My sense is that when analysts are aware of the limitations of MCFs, they can function appropriately within the tool to find true and meaningful insights. Are there issues with cookie deletion or multiple devices? Yes. But those issues exist for all cookie based analytics software and aren’t limited to MCFs. Notwithstanding some of the frustrations you mentioned (I do hear you on the direct traffic thing), I think that MCFs are a tremendous gift to Google Analytics users. IMHO, the strengths and benefits in the reporting outweigh the challenges you raise in your post. Hopefully, MCFs are just a springboard to more robust visitor level segmentation capabilities moving forward.
Hi Jacques,
I agree that the MCF has been slightly hyped up and although it produces some wonderful new perspectives I can’t help but think sometimes you’re looking for something that is not there.
I think you have picked on the right report “Channel Grouping Path”. Although that would appear on initial viewing to be the most promising report within MCF it is actually the least useful. I am finding the Assisted Conversions, Time Lag and Path Length more useful. In particular Assisted Conversions is enabling me to advise clients that their media investment needs to go beyond simple last click rules. We knew that anyway but this report gives real substance to it and makes clients thinks more about what channels are delivering valuable traffic up the funnel. However I do take on board your deleted cookies and different PCs/mobiles point which I suspect will potentially grow.
Hello Jacques,
Thank you for this post. I often asked myself this question.
I think this analysis can be helpful for marketers.
This channel grouping path shows us how prospects go through phases in the buying cycle.
If the stages of the buying cycle are taken into account when planning and executing marketing campaigns, it seems to be possible to interpret this kind of analysis.
So, if the goal of the display advertising campaigns is to create attention, they achieve it. If the goal of paid search is to create interest and trigger off an action (conversions), yes they get it.
But if emailing campaigns are here to generate conversions, well, there is a room to improvement. You have to tell your email marketing team to do differently.
It would be helpful to know which kind of combinations generate the least amount of conversions.
One of the marketers concerns is to shorten the buying cycle.
They have to engage with prospects at the earlier stages of the buying cycle and then build a relationship with them. This strategy is about timing and providing valuable messagings and contents.They need to know which kind of campaigns are performing well and at which stage of the buying cycle.
I think they need this kind of analysis coming from Google Analytics or from Channel Optimizer powered by AT Internet.
Benoit Arson
AT Insight
Attribution modeling is actually not a hard problem IF your tool is able to work with individual visits and IF the tool is built to handler higher-order analytical methods (i.e. NOT DUMB TABLES).
I can remember from long ago a Customer Advisory Board meeting for a certain vendor where the customers urged the vendor to apply even rudimentary confidence intervals. If they had started with that, they could now be in position for doing the conjoint analysis and other techniques (logistic regression, simple regression, multivariate regression) that would be moving our little profession into the 21st century.
I’m at the point now of using the analytics tool for producing big data sets that I then analyze with real statistical methods. It’s a shame it has to be so roundabout. I’m also using the raw logs to produce even bigger datasets of individual visits or transactions. I don’t know if the web analytics tools will ever move beyond simple business-level analysis.
I don’t see the DAA saying anything on this topic either.
But our users are getting more sophisticated and asking about it, that’s for sure.