Attribution is a hot topic in the digital marketing industry. Any advertiser will admit they have ‘attribution problems’ and bemoan the absence of a perfect solution. Most marketers don’t fully trust attribution. Below we explore different types of attribution and what the future of attribution looks like.
What is Attribution?
Attribution is the process of determining how much impact a particular marketing tactic has on sales. It helps you answer questions like:
– Which channels are driving the most conversions?
– How much value do each of my marketing tactics add to sales?
When most people talk about attribution, they’re often talking about ad platform attributed results (Facebook, etc) or results attributed to certain channels inside of Google Analytics. These are your last-click and last-touch models.
When more technical marketers mention attribution, they’ll probably be talking about statistical modeling that factors in more than one channel’s impact on sales.
Next we’ll review the different types of attribution:
Last-Click and Last-Touch Attribution
Last-Click and Last-Touch Attribution are the simplest forms of attribution and the most common. It works like this: The last piece of marketing a customer engages with is credited with generating the conversion.
The problem with these models is that they doesn’t take into account all of the other factors contributing to a customer’s decision to convert. For example, if someone sees an ad on Facebook and then clicks on it but doesn’t buy anything right away, then later purchases from a Search ad, Facebook won’t get any credit for that sale, even though Facebook played a role in converting that customer.
This model works fine if you’re only advertising in a few channels, and even more so if those are all acquisition channels. Last-Click and Last-Touch become more problematic when you start deploying brand tactics that won’t get much credit on these attribution models. Most marketers default to these simplistic attribution models because they’re the default in free tools, which 62% of advertisers use exclusively.
Media Mix Models
Media Mix Modeling (MMM) began in the 1950s and uses statistics to predict how changes in ad spend will affect sales. Media Mix Models are still used by many companies to analyze how much money they should spend on various marketing strategies so they can maximize their return on investment (ROI). Machine Learning and A.I. are being added to increase the accuracy of the modeling.
Media Mix Models are best used when conversion events can’t clearly be attributed to advertising efforts, like an in-store purchase for CPG brands. Another strong application of Media Mix Models is to estimate the impact of unmeasurable channels like radio, print, or out-of-home (OOH).
The biggest weakness of MMM is how little data is used as an input. This means that the modeling relies heavily on assumptions and estimates. Media Mix Models can easily overlook the individual nuances of advertising different products in different industries to favor generalized assumptions about what an ad channel will do.
Multi-Touch Attribution (MTA) is the attribution model that tries the hardest to generate perfect answers, taking into account dozens of variables. This makes MTA a more expensive methodology because of its complexity. Multi-touch attribution uses a combination of sources, such as display, search and social, plus estimation around variables like time of day, the weather, etc., to assign a fractional credit to each touch point on the path to purchase.
The biggest weakness of MTA is the assumptions and biases that inform models. Advertisers who use MTA are essentially trusting a ‘black box’ to give correct answers. Every MTA platform has a different methodology and biases built into the model, so you’ll never get the same results from multiple providers.
Incrementality testing is a method of A/B testing that helps marketers isolate one variable at a time and see how it affects their overall campaign. Instead of traditional A/B testing where a website element or campaign creative is tested against another, the experiment is conducted at the channel or campaign level. Incrementality often involves selecting a ‘hold’ group where the advertising tactic you’re looking to prove will be turned off. Then, you compare the decay that occurs in the ‘hold’ group with the control group.
Incrementality testing is largely focused on growth – helping advertisers understand what channels are reaching and converting net new audiences and where they can invest budget to increase sales.
Incrementality testing is great because it’s accessible to any advertiser and is a useful way to prove any channel’s contribution to results.
The biggest weakness of incrementality testing is how long it takes. You can only run one experiment at a time in any geography, so proving out the value of each channel in a complex media mix would take months. Another related problem of incrementality testing is the limited data you can capture. Seeing the difference in conversion data is just about all the information you’re able to gather, so your analysis is limited to the data you can capture.
The Big Problem with These Attribution Models: Seeing the Whole Picture Accurately
Marketers don’t trust attribution data for two reasons: the old methods of getting attribution data don’t show you the whole picture, and the part that they do show isn’t accurate. Imagine going to the Louvre to see the Mona Lisa only to see that half the picture is covered up, plus the uncovered half is blurry. You’d hardly know what you were looking at and you wouldn’t really appreciate the painting.
This is what marketers face when they use outdated methodologies to get attribution data.
Uncovering the Other Half
The first step to seeing the attribution picture clearly is to make sure you see both ways that an ad channel adds value.
First, let’s explain the different ways advertisements can add value because this is probably new information for you.
2 Categories of Users
Any advertisement can reach two categories of users:
Unique Users: these are users who are not seeing ads in any other channels
Shared Users: these are users who are seeing ads in other channels
2 Types of Value
Depending on what type of user an ad reaches, it delivers a different type of value from an attribution standpoint. In reality, the ads are doing the same thing – influencing a purchase decision. For the sake of attribution, it’s important to understand if the channel was in a media mix influencing conversion, or if it acted alone.
Incremental Value: this value is where a channel reaches and converts Unique Users. Seeing Incremental Value lets you understand what channels have the potential to scale based on efficiency to reach net new audiences.
Media Mix Lift Value: this value is where a channel reaches and lifts conversion rate to Shared Users. Seeing Media Mix Lift Value lets you understand where channels lift conversion rates in a media mix. An example of this would be delivering Programmatic retargeting ads to users who saw Social to increase conversion rate from 1% to 2%.
Think of how you’re getting attribution data now. Can you see both the Incremental and Media Mix Lift value of each channel and campaign?
If you use Last Click or Last Touch attribution, you can’t really see either.
Media Mix Models and Multi-Touch Attribution will only show the Media MIx Lift Value, but ignore Incremental Value.
Incrementality Testing will show Incremental Value, but ignores Media Mix Lift.
The first step to getting a better view of attribution is to ‘uncover’ the other half of the picture for Complete Attribution. Complete attribution will show you both types of value a channel delivers.
If you choose a tool or methodology that only shows you one, you could be missing out on huge opportunities. Here’s an example: most agency clients are skeptical of programmatic advertising because it performs poorly on Last-Click. That’s because programmatic is not an acquisition channel. It delivers low incremental value, but programmatic can deliver immense Media Mix Lift Value. We’ve seen campaigns where adding programmatic to search or social will boost conversion rates by 3x-10x. If you had an incomplete view of attribution, you’d probably pull budget out of programmatic and lose out on all the performance gains from Media Mix Lift.
How You Measure Matters – Seeing Accurately
The second step to getting a better view of your attribution is to get rid of the ‘blur’ – making sure the data is accurate. Attribution methodologies that rely on statistics have estimations (inaccuracy) baked in. Inaccuracies get compounded when users are not deduplicated.
Here’s how each attribution methodology adds inaccuracy:
Last-Touch and Last-Click don’t consider any other channels, so they can’t show an accurate picture of attribution. They rely on siloed data that doesn’t consider what’s happening in the other walled gardens.
Media Mix Models don’t ingest enough detail to produce specific reporting. Add in the assumptions that make the statistical models and you get ‘directional’ reporting that is zoomed out too far.
Multi-Touch Attribution doesn’t deduplicate data that comes from ad platforms and has all the same issues of assumptive statistical modeling.
To get an accurate view of attribution, you need a measurement methodology that doesn’t rely on statistical modeling and deduplicates users.
Seeing the whole picture accurately requires an attribution methodology that can show both types of value channels deliver: Incremental and Media Mix Lift. In addition, the way you measure matters – you shouldn’t add in guesswork where you don’t have to.
In the next wave of marketing analytics, more and more attribution solutions will be able to show both types of value advertising delivers with greater accuracy.