What to Expect When You’re Expecting [to Write Your Own Multi-Touch Attribution System]
At some point, you’ll decide to build your own marketing attribution system. This isn’t a decision to take lightly. But it’s also not one to fear. This is a sign of success! You’ve gotten big enough to have hit the limits of out-of-the-box systems. To keep growing, it’s time to molt and move into a new shell.
Having your own flexible attribution system will allow you to deepen your understanding of what’s happening in the customer journey, to think more expansively about the kinds of incentives you want to create for growth, and to be more proactive about collecting the data that will help you make better decisions.
I can’t tell you exactly what those decisions will be. I also can’t tell you exactly how to design your system. But I can tell you what signs to look for to know it’s time. And I can tell you about the common patterns.
The step up from single-touch third party attribution to multi-touch first party attribution is a big one. It’s also part of a natural progression from deterministic thinking to probabilistic thinking:
How you got here
You should be able to get to tens of millions in revenue with the help of only one or two paid marketing channels. That might be paid search. It might be outbound sales. It might be SEO. It might be Facebook or TikTok or podcasts or any number of other channels. But the point is: Focus.
Focus will allow you to live in the self-contained world of the platform you’ve committed to. You’ll use their third-party reporting — whether that’s the Google Ads UI or the Facebook UI or whatever — to test new tactics and analyze results. You won’t worry about cross-validating what they report with other systems because you won’t have other systems. Everything will be internally consistent, and you won’t think much about external consistency at all.
And that will be fine! In fact, it will be better than fine. Your whole team will be aligned to the incentives of the platform you’re focused on. Measures of success will be obvious. Decision-making will be simple and speedy.
Trouble will arise when you start to branch out into channels beyond the reach of your original platform. Usually you add these other channels when you start to hit diminishing returns in the channel you started with. Maybe SEO traffic has stopped growing, so you decide it’s time to do more demand generation. Or maybe your sales team is seeing lower and lower open rates on outbound emails, so you decide to do account-based marketing. Or maybe paid search bids have gotten too competitive and you want to explorer greener pastures.
Regardless of the reason, once you start to diversify your channel mix you’ll be thrown into a state of confusion. The perfectly-aligned measurement of your first channel is now threatened by a different worldview. Branded searches are up — is that because of the new Facebook spend or just a function of the growing customer base? Facebook click-through rates are up — is that because of the new brand awareness campaigns or simply better copy? People will start to hoard credit and point fingers.
How to know you’re ready
If you’re scaling into multiple channels, you should have first-class problems. You’ve probably grown your business to tens of millions of revenue. You probably have got people capable of asking sophisticated questions. You probably have lots of internal data and an analytics team and the sense that you could do a better job of answering these sophisticated questions than the one-size-fits-all ad platforms could. And you know what, you’re probably right.
(If instead you feel like you need multi-touch attribution because you don’t know for sure if any of your channels are working, take a step back. More sophisticated attribution is not the answer. Instead, your problem lies in getting the basics right. Find a single channel that works and keep pushing it. Only when you’re sure you’ve hit the limits of one channel is it time to add more, and only then will sophisticated attribution be the limiting factor.)
So again, these conditions should probably all be true before you even consider doing attribution in-house:
You’re spending at least a few million dollars on paid ads
You’re spending in two, and probably three or more, channels
You have an in-house analytics team with a solid data warehouse
You own your purchase funnel and the data it produces
How the process will unfold
“Someone’s first session started on a paid search click; a week later they came back direct; a day after that they clicked on an unpaid search results; then they used a coupon we shared on a podcast. How do we divvy up the credit?”
You’ll start by building a dataset of all your marketing touchpoints — sessions with UTM parameters, maybe coupons, maybe data from your referral program, maybe App Store attribution, etc. You’ll then build a framework for allocating credit across these touchpoints. You’ll explore different credit-allocation options. You’ll consider various weightings. You’ll experiment with time decay.
And then you’ll find that most of your customers have only one or two touchpoints, and that these decisions, while important around the margin, are often usually moot. So you’ll make a decision — say, given 40% of credit to the first touch, 40% to the last, and 20% to all those in between — and move on.
You’ll roll out this first multi-touch system and spend months answering questions and battling misunderstandings. At this stage, it’s impossible to over-communicate what is happening and why. Whatever road show you’re planning, double your investment. This is your one chance to really get marketing, sales, and finance on board that the system is transparent and fair. Do whatever you can not to get stuck in the buy-in process.
Once everyone is bought in, it will become clear that the limiting factor is more data. You’ll implement a “how did you hear about us” survey to the checkout flow. You’ll start using coupons tied to specific channels and subchannels. You’ll try to estimate the effects of cookie blocking.
Some of these new datapoints might throw your old assumptions into question. For instance, at Squarespace, one of our first two big channels was podcast advertising. Back in 2012 or 2013, when the company was spending $10m on marketing, almost half of it was on podcasts, and for a time, we were the undisputed biggest podcast advertiser in the world. We knew these ads worked because we used coupons that were specific to each show.
But after we added a checkout survey, we had to wonder: “Were podcasts even more effective than we’d thought?” Of all the customers using a podcast coupon, 90%+ said in the survey later that they first heard about us on a podcast. But of the folks saying in the survey that they’d first heard about us on a podcast, only a third were using a podcast coupon. It turns out there had been tons of false negatives. Podcasts were potentially 3x more effective than we’d thought. We spent more, and faster, with this knowledge.
Eventually there won’t be any new data to gather and the remaining unmeasured effects of your marketing will fall into two categories. First, view-through effects. Everything from Facebook to billboards to TV to Spotify could fit in this category. The desire to really pin this down will cause you to start systematic lift testing in the digital channels that allow it. You’ll move to the third level of the pyramid from earlier in the post.
Second, interaction effects. Once you’re spending in a bunch of channels, the whole may be greater (or less) than the sum of the parts. Facebook campaigns might be making paid search more effective, or TV ads might be making Facebook more effective. These complex problems might lead you to implement a mixed media model, such as Recast. But don’t cross this bridge unless you’re sure you have to, and only once you’re spending at least millions of dollars per month on media.
Not just facts, but beliefs
Teams often start down this road thinking they need the answer to a narrow question: How are we allocating credit? They’ll quickly add a second question to the list: How can we gather more inputs?
But having more control over attribution will allow you to ask more substantial questions about your business. What’s our underlying theory of how we grow, and how do we represent that in our attribution system? What kinds of marketing interactions are inherently more valuable and why? How aggressive do we want to be in giving credit to speed spend and growth? When do we want to hit payback positive, and how can we represent that desire in the time weighting of attribution credit?
Simply put: What is our strategy?And what information are we passing back to ad platforms to encourage behaviors that further our strategy?
You’re building an incentive system, and the question of what you want to incentivize is a subtle one. Your answers will be grounded not just in facts but in beliefs. The beauty of an flexible attribution system is it will allow you to toy with the facts, but at the end of the day, you’ll be doing so to test your beliefs. Know that going in, and own it. Control of the inputs and control of the outputs means you get to plot the course.