Welcome to the first chapter of the Fairing Attribution Academy! I’m Matt, CEO and Co-founder of Fairing. I’ve been in the e-commerce and digital marketing space for a little over 14 years, and have been fortunate to work with multiple 8 and 9-figure brands. In this time, I’ve seen dramatic changes in measurement and attribution methodology and marketers' access to data.
In this chapter, I hope to distill those years in the market into a timely, impactful update that serves to educate you, the reader, on modern day measurement best practices. With that, let’s set the stage by looking at The State of Attribution today.
The genesis of digital attribution can be traced back to the simpler times of last-click attribution, a model that for years held sway over digital advertising performance. It’s not something I’d recommend in 2024, but alas, simplicity often wins. (Something many measurement practitioners could learn from.)
As new media types were introduced and marketers could diversify their budget and advertise on multiple channels, last click’s limitations became glaringly apparent. This led to the advent of multi-touch attribution (MTA) models whose goal is to distribute credit across multiple touch points.
To make matters worse, the introduction of smartphones added another wrinkle—the need to track the customer journey across multiple devices. Arguably, this is a problem that will never be solved with MTA’s bottoms-up approach.
Today, most marketers actually don’t necessarily care about each touchpoint (and certainly not the last), but rather the total return on each investment in ad spend. If I spend more money on this channel, will I get a greater return than if I didn’t?
This heuristic is also often combined with the ease in which a channel can be measured—blackbox multi-channel measurement will most likely not prevail. We’ve come a long way from if it can’t be directly measured, don’t allocate spend, but we’re certainly no where close to where we were in the pre-digital marketing era.
2014 to 2021: Pre-Apple Privacy Shift
From 2014 to April 2021, a combination of both platform-reported measurement + Google Analytics last-click was king. The majority of ad spend was on Facebook & Instagram, AKA Meta (and still very much is). Brands built a competitive advantage through their ability to acquire customers at scale. If you worked through this time, the phrase “LTV:CAC ratio” might trigger some interesting memories, although we’ll save that topic for a different day. We’ve migrated from What’s your CAC to What’s your MER?—a question that’s more representative of a brand’s P&L than their unit economics.
One of the biggest learnings of this time was that building a business on projected cashflow from returning customers for a non-subscription product is in no way sustainable.
The Methodologies of Today
Fast forward to today. While there are a slew of marketers relying solely on platform-reported attribution, it’s clear to most that diversifying measurement methodologies is a key to long-term success at scale. This is becoming even more clear with Meta’s current performance.
The core methodologies today are:
- Single touch attribution (i.e. last click) – this methodology is thankfully fading, but is often used in conjunction with other measurement methods to analyze the bottom of the funnel. Legacy finance teams also can’t let their grip on LC go!
- Multi-touch attribution (MTA) – MTA can be simple (first + last touch), or quite complex (programmatic weighting). MTA simply looks at all touchpoints that led to the path-to-purchase. Bottoms up approach.
- Media Mix Modeling (MMM) - historically reserved for large marketing budgets & revenue and performed at a multi-month, bi-annual cadence. New MMM models have emerged with the ability to measure regressions in days or weeks, not months. The main value of an MMM is its ability to measure environmental factors, like seasonality. Top-down approach.
- Platform-Reported - platform-reported simply means optimizing budgets based on the reporting output of the platform (e.g. Meta) itself. It’s often regarded as “grading your own performance”, but arguably the incentives are aligned.
- Incrementality Testing - This methodology relies more on experimentation than deterministic click-based measurement. It’s gaining in popularity, yet often faces budgetary constraints and is most often used in conjunction with MMM or MTA. In a most simplified depiction, incrementality testing will use hold-out groups plus a control to determine if advertising is having a positive effect on conversion or not.
- Attribution Surveys - attribution surveys utilize direct-from-consumer responses to questions like “where did you hear about us?” to understand top-of-funnel channel performance. Very often combined with other methodology, attribution surveys help marketers understand the harder to measure channels like podcasts, influencers, TV, etc.
Outside of the core methodologies above, brands also often rely on discount codes and vanity URLs to help understand performance of offline channels like Podcast and OOH. For brands advertising on TV, lift testing is the common measurement method.
On the Horizon: AI, Incrementality, and Legislation
But was it incremental?
Incrementality and credit allocation are perennial problems in the world of measurement. Did the branded search term actually assist in the conversion? What percentage of retargeting ads are being shown to consumers who were going to buy anyway? Was that SMS message necessary? After all, brand awareness exists and advertising clearly isn’t the sole driver of revenue.
If you’re a marketer relying solely on platform-reported metrics you’ll quickly realize that Meta is going to take credit for way more sales than they drove. All platforms do it. If you’ve ever added up the attributed sales from all your marketing platforms, you’ll know exactly what I’m talking about.
Incrementality is what every marketer should be striving for—it helps you understand the causal lift of your campaigns. As a company matures, it becomes all the more important.
Measuring incrementality is typically done with controlled experiments. Choose 2 groups, and hold one as the control. Target the other with different ads or media mix, and then measure the differences in outcomes. Assuming you’re going into this with a set hypothesis, this experiment will either prove or disprove your initial assumptions.
Unlike MTA, I’ve yet to see anyone run incrementality experiments in-house, and at scale. Running a valid experiment typically requires deep statistical knowhow and that experience is typically not available in an in-house team. Nor is it typically worth it (yet) to attempt to do this in-house, although I’d be happily proven wrong.
Grappling with Future Privacy Restrictions
Privacy initiatives have always affected marketers ability to measure, but it arguably wasn’t until Apple launched its App Transparency Tracking (ATT) in iOS 14.5 when everyone started taking it seriously. Chrome’s impending (and continually delayed) deprecation of third-party cookies will similarly restrict cookie-based tracking. Today, Chrome is blocking third-party cookies by default for 1% of its users.
Privacy regulations and tracking restrictions are kryptonite to bottoms-up, data-centric methodologies—for example, if your models lose signal from anyone who purchases on an iOS device, that feedback loop disappears.
The future of attribution, therefore, is one that weaves together innovation, ethics, and consumer empowerment. In this new, cookie-less world, marketers must pivot towards or at least be prepared for privacy-first strategies, leveraging first-party data and contextual targeting to sustain the relevance and efficacy of their campaigns. This is why MMM solutions are gaining traction today–given their top-down approach they’re somewhat immune to these changing variables.
The AI and ML Advantage
An advertising platform’s ability to understand when conversions happen is paramount to its success. Modern day advertising auctions and media buying platforms rely heavily on machine learning trained on various data sets. When a conversion happens, that data point is added back into the model, thus improving its ability to target lookalikes. Simple enough. This is why server-side tracking is seeing wild adoption—they improve Meta’s ability to create a flywheel as they are resilient to pop-up blockers and other mechanisms that may block a client side pixel.
Synthetic Conversion Data
It’s unknown how Apple will continue to disadvantage those who rely on cookies and other tracking mythologies - they clearly see it in their best interest. Chrome is currently restricting third-party cookies by default for 1% of users with the plan to ramp this up to 100% – is also creating a path of unknowns.
is restricting third-party cookies by default for 1% of Chrome users to facilitate testing, and then ramping up to 100% of users from Q3 2024.
What platforms have started to do is to utilize probabilistic models to inject synthetic conversion data to fill the gaps where tracking has dropped off or the consumer journey has gone dark. They don’t need to know 100% that a conversion happened, but with a great deal of confidence that can still optimize a campaign.
These companies will always be looking for additional data sources to train these models—and they will succeed.
Someone once told me that Meta is as much a DMP (Data Management Platform) as it is a publisher—and why they’ve built an incredible moat by creating the best-performing advertising platform. Current and future competitors will continue to face the cold-start problem with getting initial user data to optimize their ML models.
Build vs Buy
The complexity of some of these methodologies makes buying software to solve the problem the likely path for most. I’ll continue to recommend that the best methodology chosen is the one that the team can best understand. Building an MTA model in-house with Segment click data dumped into a Bigquery database is no longer an uphill climb, although those are skill sets that might not be readily available on your team.
Your team’s ability to analyze the output of a model is paramount. If a healthy dialogue around the output of an MMM is not something you see anyone on your team having, then it may not be the best solution for you…at this current moment. Marketers are also usually guilty of cherry picking models and data that produces the results they want to see, rather than the ones that are most reflective in reality..
I say this having seen (and having been on both sides of) countless budgets wasted and immense finger pointing (at the vendor!) when in reality there wasn’t good initial alignment.
What’s next for attribution?
The road ahead for attribution is complex, but ripe with opportunities for the intrepid marketer to find that alpha! By staying informed, adaptable, and committed to customer-centricity, DTC brands can continue to grow efficiently in this new era of measurement.
I hope this post has been an illuminating look at the state of attribution. Stay tuned for more deep dives and practitioner spotlights in future issues.
