Recently, I was asked to comment on my favorite attribution model for paid advertising. Off the top of my head, I didn’t really have one — and the more I thought about it, the more I realized that all attribution models are fundamentally flawed in a way no one is thinking about. Or at least, not that I hear discussed regularly.
The issue isn’t with the actual models, per se, it’s that no model really nails what the “total value” being measured actually is.
Let me explain.
The point of attribution
Marketing managers have a lot of options when it comes to how they spend their PPC budgets. Getting the right balance between online channels can be tricky because not every channel drives a direct sale.
Some channels, like social media ads, are known as top-of-funnel channels — they target consumers early on in their purchase journey when they are simply browsing. The goal of these channels is to arouse interest in a particular brand or product even if a sale doesn’t happen in that instant. Conversely, other channels, like Google Shopping, are a lot more likely to drive a direct sale because consumers happen upon them much later in the purchase funnel.
If you focused your budget exactly proportionately on the channels that drove direct sales — known as “last-click” attribution — you would miss out on a lot of top-of-funnel marketing activities that drive awareness and desire for your products.
Attribution models were created so that we could assign due credit to the right channel for delivering a certain percentage of some “total value” that advertising brings. That way you would know how many resources to put into each channel for the greatest effective marketing strategy.
My issue with all attribution models is that they assign credits to touch points, which assumes an accurate measurement of the value of a transaction generated by advertising. Attribution models can help to estimate what percentage of the total value should be assigned to each touch point, but the total value itself is inaccurate.
What is the total value?
The most simplistic answer to this question is order revenue, which on its face makes sense — you want your ads to generate the maximum possible sales revenue. But in e-commerce, things are never that simple.
Using something as basic as revenue means that you are devaluing the impact of a lot of very important factors for e-commerce success.
For example, if you consider that every customer has a lifetime value, wouldn’t it be worth more to acquire a new customer (who has a full lifetime value ahead of her) as opposed to an existing customer (who only has a partial lifetime value ahead of her)?
A purely revenue-focused attribution model wouldn’t be able to make this distinction of value between customer types. It also ignores secondary effects like customer referrals, newsletter signups, social media shares and product reviews — all of which can have a big impact on sales.
There’s also the question of which stock item is sold via which touch point. We’ve spoken before about how not all product sales are created equal. Selling something that is taking up expensive warehouse space and of which you have an abundance is far more valuable than selling something that’s running out of stock quickly. Yet, when revenue is tracked, a full attribution value is assigned to both sales equally.
Revenue models also fail to assign different values to sales at the various product life stages. This is particularly problematic when you consider the sale of perishable goods (I include products with extreme seasonality and fast fashion in this category) because the value of their sale changes dramatically throughout their life cycle.
Finally, while two products may have the same price, they could have wildly different margins, making one of them a much more profitable sale than the other. But, since revenue and profit are not interchangeable, current attribution models can’t account for this difference.
Solve the value first, then optimize
My point is that there is a lot more to advertising value than pure revenue. Treating every transaction the same will eventually lead you to allocate the completely wrong value to the touch points in your customer journey.
Most discussions of attribution models talk about how we can tweak this metric or that to divide the allocation differently. But, if what you’re dividing in the first place is ambiguous, incomplete or wrong, the value of that division is completely lost.
I don’t claim to have the answer to what the total value should really be (yet, anyway). Even Google’s latest stab at data-driven attribution doesn’t take into account the intricacies of product life cycle management. But as an industry, I think this is the conversation we ought to be having.
We should put our brains and our models to work to figure out how to accurately calculate the true value of each transaction before we go about attributing value to the touch point that led us there.
I may not have been able to come up with my favorite attribution model, but I can say that any model is only as good as the action you are able to take based on its data. It’s essential that we create bidding systems that are able to “see” the attributed values and base bid adjustments on them automatically.
Some opinions expressed in this article may be those of a guest author and not necessarily Search Engine Land. Staff authors are listed here.