Marketing attribution is a hot topic in the eCommerce world, especially post-iOS14, when in-platform metrics have become less reliable than ever. And that’s a major reason why we developed Triple Whale: to demystify what parts of your marketing spend are truly driving your results.
Most marketers’ introduction to attribution is Google Analytics last click-attribution. The reason? Most brands launch with GA as their primary analytics source because it’s free and easy to implement. But last-click never tells the entire story, and it over-simplifies a complicated situation.
At the other end of the spectrum, you have some large brands that work to build their own custom attribution models. Investing in a custom model can make sense if you are spending multiple millions on advertising each month across a number of digital and “IRL” channels. But the unfortunate side-effect of custom modeling is analysis paralysis: organizations become so obsessed with finding the perfect answer that they forget to move forward.
If you’re reading this, you probably sit somewhere in the middle of that spectrum. So it’s a perfect time to take a step back and dig into attribution from a first-principles perspective so you can make informed decisions.
There is a famous marketing quote that goes something like this: “I know that half of my media spend is wasted. I just don’t know which half.” This sums up the attribution puzzle pretty well. And, ironically, there are two potential sources for the original quote. Another attribution puzzle.
Incrementality is the holy grail of marketing. The reason that marketing exists is to create demand for a product or service that wouldn’t exist in absence of marketing–and do so profitably.
Last-click attribution measures customers’ interactions with a specific campaign. But it doesn’t measure what would have happened if you didn’t send the campaign. So last-click analytics are essentially useless for understanding the incremental impact of your marketing.
This is where attribution modeling comes in. Through a variety of approaches, attribution attempts to determine which marketing campaigns were most responsible for a conversion, without forcing the marketer to run a holdout test on every single campaign.
Some marketing activities push net-new demand into your business. Other marketing activities capture existing demand; they wouldn’t drive any conversions in the absence of existing demand. Branded search is a great example of a demand capture channel. If someone is clicking on a branded search ad, they have to type your brand name into Google to trigger the ad.
Demand capture channels are closer to the bottom of the funnel because they require existing brand awareness and intention to purpose. For that reason, they often produce great last-click results, but their ability to drive incremental conversions is questionable. At worst, these campaigns are simply collecting a toll from you each time you make a sale.
Understanding which of your marketing activities generate demand vs capture it is the foundation of instinctual attribution. If you only have a single demand gen channel running, and your brand is relatively new and unknown, almost 100% of your sales can be attributed to that channel.
You can’t solve attribution problems purely by gut instinct, but this understanding will help you sense check any attribution reporting you run.
Most marketers use a metric like aMER to get a daily read on the efficiency of their marketing spend. aMER is total dollars spent on acquisition divided by total revenue from new customers in the same period of time.
As your acquisition increases, your ad efficiency typically decreases. This is because you’re reaching broader audiences who are less likely to have heard of the brand and need more convincing to make a purchase.
But looking at incremental spend on a blended basis obscures the true efficiency of each additional dollar you spend. That’s because every dollar and every transaction is averaged out. To get a true understanding of the efficiency of each additional dollar, you have to limit your analysis to the conversions you won with that dollar.
Each customer you acquire has some future lifetime value potential, although many of your customers won’t return after their first purchase. That future lifetime value potential is rarely factored into attribution modeling, but it’s an important tradeoff to consider.
Let’s say that the average new customer you acquire spends $200 in their first year with your brand, places two orders, and has a 30% probability of returning for a second purchase. That gives each new customer a future value of $60.
In this hypothetical example you have $30 to spend. That $30 will buy one new customer conversion or three returning customer conversions. You would want to be very confident in the incrementality of those returning customer conversions before deciding to invest in them. New customer acquisition benefits your ability to sell in the future.
Now you should have a better idea of the outcomes that are likely to result from your digital marketing decisions. But that doesn’t mean that your marketing mix will never throw you a curveball. That’s what attribution modeling is for. Here is an overview of the three most common types of models and how they work.
This is one of the most straightforward attribution models on the market, popularized by Google Analytics. Essentially, the last channel that a customer interacts with before visiting the site and converting gets the “credit” for the conversion.
So if a customer receives an email, and that email prompts the customer to visit your brand’s Instagram page where she sees a new product, and then that customer searches “[your brand] + [product name]” on Google and clicks on a search ad, the search ad would receive credit for the sale.
In this example, it’s easy to see how last-click attribution dilutes the importance of demand generation channels and overstates the importance of low-funnel demand capture channels. That doesn’t mean that last-click attribution is worthless. It can still be used as a way to track baseline performance over time. But to be successful, you need to hold demand gen and demand capture campaigns and channels to different standards.
Multi-touch attribution gathers a user’s entire history of views and clicks, then applies statistical models to all of your customers’ data to determine which channels “caused” a conversion. MTA does a much better job than last click of measuring the impact of demand generation channels. This is especially important when a brand expands beyond its first paid acquisition channel and determining which channel kicked off a purchase journey becomes trickier.
Of course, no model or tool is going to capture 100% of every customer’s data. But a good MTA model will use statistical techniques to account for data that may be missing. MTA is also unable to capture channels that purely exist “IRL”, like billboards. But in some cases it is possible to use cell phone location data to get an approximation of who viewed out of home creative.
Instead of relying on individual customer data, Marketing Mix Modeling collects historical data about when ad campaigns were running, how much budget was spent, and how many conversions took place. A statistical model is built using this data, as well as seasonal and historical trends.
The MMM approach is able to capture non-digital campaigns without special considerations. But like any statistical model, there are limitations. MMM can capture the impact of outlier situations like COVID-19 in hindsight, but not necessarily while they are occurring. And the conclusions drawn from these models are only valid if the conditions powering the model remain relatively static.
Humanity has yet to develop a statistical model that can predict the future. Attribution models can give you a clearer understanding of what happened in the past, and why. But they can’t completely de-risk your decisions. Models will tell you what’s working and what isn’t. But they won’t tell you how to solve your problems if your current approach isn’t working.
Here’s an example: let’s say your brand is running ads on Facebook, Pinterest and Google. After running an attribution model, you discover that the Pinterest spend is more profitable than you thought, and the Facebook spend is just about breaking even.
Using this information, you can see what happens when they pull back on Facebook budgets slightly and invest those dollars in Pinterest. The model won’t tell you what to do once Facebook and Pinterest have been scaled to the point where both channels have reached break-even. And it won’t tell you what to do if Facebook suddenly becomes unprofitable when you haven’t changed your investment or campaign strategy.
Data doesn’t solve every problem. There is still a need for creative problem solving in marketing, and there is value in “getting in the reps” and experiencing different challenges for yourself.
If you’re looking for an attribution tool that will help you make better decisions, faster, then you should give Triple Whale a try.
Supercharge your growth with a purpose-built ecomOS for brands and agencies.