An email I got from Avinash Kaushik this week got me thinking about multi-channel attribution.
Assigning value based on position in the customer journey is like saying it’s more important for outcomes if the surgeon closes the wound (the last part of the operation) than removing the cancerous legion. Obviously, the surgeon needs to perform every step in the operation or the patient won’t have a good outcome. And, post-op care by nurses, medicines, and rest all contribute to a positive outcome, even though those things happen AFTER the surgery is complete.
Yet, hospitals and physicians need a basis for charging patients. Part of that charge comprises the inherent value of services performed and the costs associated with those services. They don’t use a multi-channel attribution model to do it. Why do we?
What is multi-channel attribution?
Image courtesy of Coast Digital
Think about the customer journey.
Along the journey, prospective customers visit various pages within your website, your social networks like Facebook and Twitter, and maybe visit a physical store to see your product in action (in fact, Best Buy filed a lawsuit that they’re performing a role in the sale, but not getting the reward because customers buy online).
Multi-channel attribution assigns a value to each step in the journey. Commonly, businesses use “last click” attribution which means assigning all the conversion value to the last step in the journey or “first click” attribution, in part because many analytics software products are set up that way. Alternatively, firms assign value to various steps using some formula, although this is much less common.
Questions is assigning value to steps
Google Analytics provides a variety of models for attribution like the one to the left — I’ve cropped the image to protect the privacy of client data, but yours will show the $ value of each step in the conversion.
Determining which model involves asking some difficult questions about how customers move through their journey.
Is one of these steps more important than the others?
If you eliminated one or more steps, would you enjoy the same sales volume at a reduced cost?
Do all customers experience the exact same journey or are there a nearly infinite number of combination of steps?
Does the customer progress through this journey at a regular pace or might there be multiple gaps before making a decision, thus extending the journey over months or years?
Does the customer progress through the journey in a linear fashion or might he/she use a more circular path by jumping up and down the funnel — or even sideways?
What happens when a prospect leaves your site? Do actions you can’t measure impact your results?
How does traditional advertising, which is much harder to measure, impact the results you see in combination with social actions?
And, these are just a few of the questions plaguing multi-channel attribution modeling. Hence, Kaushik suggests only the most sophisticated firms should even attempt to measure multi-channel attribution.
Measuring multi-channel attribution
The first step in developing multi-channel attribution analytics is understanding the customer journey — from awareness through purchase to after-purchase support, which is the most overlooked aspect of the entire customer journey. The ultimate goal is developing loyal customers who spend a significant percentage of their purchases on your brand (we call this share of wallet), recommend your brand to others, and potentially becomes a brand advocate — defending your brand from others.
Once you extend the customer journey to its final steps, you see multi-channel attribution models fail miserably when it comes to assigning value because they totally ignore aspects of the journey occurring after the sale.
So, should we totally ignore attribution, as suggested in the email from Kaushik?
So, how do you implement attribution without running afoul of arguments that attribution models suck?
Recognize that each step in the customer journey provides value and accept that you can’t calculate the value of each step in the journey. Also recognize that various paths through the customer journey exist and some are more effective than others.
Next, you’ll need a custom model representing the customer journey options and assign value to each customer journey path. In the image of Google Analytics, you’ll find a link to create a customer model is at the bottom of this drop down menu. BTW, this used to be available in premium analytics, but Google made it available for everyone. Cool.
A great way to build your custom model is by looking at the funnel options in Google to see all the pathways folks take into and through your website. If you get really stuck, here are some from the Google Gallery of attribution models from leading analytics gurus.
Within this data visualization, you can see how visitors move through your website, where they came from and where they drop off. You can highlight any entrance option to see just how visitors from that source moved through your website. Notice, this is my website since I never share clients’ analytics. I don’t have an ecommerce site, so yours will likely look substantially different, but the idea is the same.
Crafting your custom model is fairly straightforward and Kaushik gives a great step-by-step on Occam’s Razor.
However, as good as all these models is, they fail to capture the value inherent in after purchase behaviors since many of them involve your website to a lesser degree. My suggestion is to build a more nuanced model by including other interactions likely to take place after the fact, such as accessing contact pages, FAQs, or review pages that might indicate a post-purchase behavior.
OK, you’ve got some value assigned to various channels and journeys based on real data. There are several ways in which this data provides valuable insights for decision-making — which is why you want this data in the first place.
You really shouldn’t use the data to determine a particular channel doesn’t work, so you should just drop it. That decision really requires more data. Plus, the true value of a channel might be obscured by the number of different ways it impacts customer purchase that are difficult to measure. Instead, use multi-channel attribution data for:
1. Comparison (benchmarking)
As a single brand, determining what the numbers mean in terms of decision-making is a little challenging. If you have multiple brands in the same market or work with an agency that might provide insights, you should really look at how a brand compares with others in the same customer market.
Difference between your performance and that of others indicates opportunities lost to ineffective strategy.
Rethink your strategy in an effort to improve your performance to match or exceed the competition.
2. Improve poor performers
Let’s say certain social channels don’t perform as well as you might expect.
Determine why these channels aren’t working and focus attention to improve their performance rather than simply limiting efforts to concentrate on channels that do perform better.
3. Set performance goals
Use current performance across a particular channel to set performance goals for the next period.
4. A/B testing
Using A/B testing you might determine how much investing an extra $ on a particular channel translates into improved market performance. Don’t assume that spending an extra dollar will improve performance in a linear fashion as that’s probably not a valid assumption.
Once you know the ROI from a particular channel and how market performance improves by investing more money, you have the tools to forecast sales into the future. This makes budgeting more effective and allows better planning.