Why Accurate Online Sales Attribution Is Impossible
Let's imagine I go online to buy some headphones.
I might first start my "online journey" with a generic search.
This might be simply typing in "best headphones" into Google and browsing the results.
After a little research I know that I want noise reduction headphones and that Sennheiser are a decent brand with some good reviews.
Over the next 3 days I am exposed to several display ads (only some of which I click, the rest I barely notice) and watch some video reviews on YouTube.
I finish my journey by typing "play.
com" into Google and then I purchase the headphones from Play.
com.
A few years ago a lot of folk in advertising were only seeing the end of this journey from their data.
They knew that people were converting on a branded search (such as "Play.
com") and were not seeing that people's journeys were actually starting with a generic search.
This caused some to move budgets away from generic keywords, thus some sales - that were kick-started with a generic search - suddenly vanished.
Oh dear.
Display advertising also felt the brunt of this attribution method, known as 'last click wins', as branded search (often used by consumers to navigate to a site for a purchase) effectively hijacked the credit for the sale from display ads that featured in the journey.
Affiliate advertising is now hijacking many sales due to their "cash back" incentives.
But attribution solves this problem...
doesn't it? So along came an idea called "attribution".
Here the number crunchers looked at the entire online journey.
Every display ad, impression or click, search ad, affiliate ad etc was considered and given a weight.
Then the credit was shared out according to a clever algorithm - often known as a 'black box' whose mechanics were rarely explained to an advertiser.
So...
problem solved? Not quite.
The big praise for these models were that they were now considering the entire user journey and so every touchpoint was being considered.
This is simply not true.
The new problem for advertisers For low purchase decisions you can start your journey and convert online in a few clicks (buying that Journey single that you've just heard on the radio whilst sat at your computer for example).
In these circumstances you may feel confident that you've captured the entire user journey.
But a lot of online purchases are considered for days, if not weeks and span multiple devices.
Just thinking about my own experiences, I might "Google" some reviews whilst on the train to work if I'm considering buying a new phone.
Then at work I'll continue with a few natural search researches.
I might see a few display ads across the network of sites I visit and then I go home and buy online from my home laptop.
That one journey, in one day, has spanned three devices.
Whoever then analyses this data will see two non converting journeys (on my phone, then on my work PC) and one short journey that does convert (my home laptop).
Then there's the topic of cookie deletion where entire journeys are being deleted as users close their browser.
Often natural search is not tracked either so is not included in attribution models (though technically this is possible and is performed by some agencies).
And what if I buy offline? So what can marketers do to get a true understanding of online performance? I feel the answer lies in not trying to seek the easy solution where we can tally up an entire journey and allocate a weight in a report.
Current attribution models are an improvement on a 'last click wins' model but they are not the perfect model.
The truth is there will never be a perfect model.
Technology will give advertisers more and more ways to communicate to consumers so we will never be in a position to capture every dataset and link them all together.
An extreme solution - but possible? The only answer I can see (and it is as extreme as it is unlikely) is an internet access key.
Imagine that internet piracy causes governments to impose a law that everyone must identify themselves when going online, no matter what the device (the same way you can allow people to vote online without people being able to rig the system).
This is the only way I can see marketers being able to link datasets together (so the first part of a conversion journey on an iPhone can be matched with a conversion journey on an internet cafe PC when you're on holiday and the entire journey is then analysed).
I might first start my "online journey" with a generic search.
This might be simply typing in "best headphones" into Google and browsing the results.
After a little research I know that I want noise reduction headphones and that Sennheiser are a decent brand with some good reviews.
Over the next 3 days I am exposed to several display ads (only some of which I click, the rest I barely notice) and watch some video reviews on YouTube.
I finish my journey by typing "play.
com" into Google and then I purchase the headphones from Play.
com.
A few years ago a lot of folk in advertising were only seeing the end of this journey from their data.
They knew that people were converting on a branded search (such as "Play.
com") and were not seeing that people's journeys were actually starting with a generic search.
This caused some to move budgets away from generic keywords, thus some sales - that were kick-started with a generic search - suddenly vanished.
Oh dear.
Display advertising also felt the brunt of this attribution method, known as 'last click wins', as branded search (often used by consumers to navigate to a site for a purchase) effectively hijacked the credit for the sale from display ads that featured in the journey.
Affiliate advertising is now hijacking many sales due to their "cash back" incentives.
But attribution solves this problem...
doesn't it? So along came an idea called "attribution".
Here the number crunchers looked at the entire online journey.
Every display ad, impression or click, search ad, affiliate ad etc was considered and given a weight.
Then the credit was shared out according to a clever algorithm - often known as a 'black box' whose mechanics were rarely explained to an advertiser.
So...
problem solved? Not quite.
The big praise for these models were that they were now considering the entire user journey and so every touchpoint was being considered.
This is simply not true.
The new problem for advertisers For low purchase decisions you can start your journey and convert online in a few clicks (buying that Journey single that you've just heard on the radio whilst sat at your computer for example).
In these circumstances you may feel confident that you've captured the entire user journey.
But a lot of online purchases are considered for days, if not weeks and span multiple devices.
Just thinking about my own experiences, I might "Google" some reviews whilst on the train to work if I'm considering buying a new phone.
Then at work I'll continue with a few natural search researches.
I might see a few display ads across the network of sites I visit and then I go home and buy online from my home laptop.
That one journey, in one day, has spanned three devices.
Whoever then analyses this data will see two non converting journeys (on my phone, then on my work PC) and one short journey that does convert (my home laptop).
Then there's the topic of cookie deletion where entire journeys are being deleted as users close their browser.
Often natural search is not tracked either so is not included in attribution models (though technically this is possible and is performed by some agencies).
And what if I buy offline? So what can marketers do to get a true understanding of online performance? I feel the answer lies in not trying to seek the easy solution where we can tally up an entire journey and allocate a weight in a report.
Current attribution models are an improvement on a 'last click wins' model but they are not the perfect model.
The truth is there will never be a perfect model.
Technology will give advertisers more and more ways to communicate to consumers so we will never be in a position to capture every dataset and link them all together.
An extreme solution - but possible? The only answer I can see (and it is as extreme as it is unlikely) is an internet access key.
Imagine that internet piracy causes governments to impose a law that everyone must identify themselves when going online, no matter what the device (the same way you can allow people to vote online without people being able to rig the system).
This is the only way I can see marketers being able to link datasets together (so the first part of a conversion journey on an iPhone can be matched with a conversion journey on an internet cafe PC when you're on holiday and the entire journey is then analysed).
