Most marketers believe that the only way to get attribution data is with models, but few realize that accurate attribution data can also be obtained with direct measurement.
Specifying attribution data from a model or measurement might sound like a difference without a distinction, but the truth is that modeled and measured attribution couldn’t be further apart. Let’s break down the difference between modeled and measured attribution and discuss when you should use each one.
The Way It’s Always Been
The first attempts to show attribution data – showing how ad spend turns into revenue and crediting channels for conversions – happened in the late 1950’s. At the time there were only a few different ad channels a company could run: print, broadcast radio, and broadcast television.
There was no technology to connect a store’s customers to those who were exposed to advertisements. This necessitated mathematical models to fill in the gap and estimate the impact a brand’s advertising has on sales.
How Modeled Attribution Works
Statistical modeling uses mathematical models, takes sample data from a larger data set or population and applies statistical assumptions to approximate what’s happening in the real world. In the case of attribution, this means taking a small sample (5%-10%) of advertising delivery and/or engagement data and applying a set of assumptions around buyer behavior to give credit to ad channels for sales. This is how almost all attribution companies will generate data.
Leveraging statistical modeling for attribution makes a lot of sense when there’s no good way to measure when or where your ads are being served. Using modeling also lets marketers inject external data to forecast and predict performance like seasonality, weather, or macroeconomic conditions. Another notable advantage of modeled attribution is the simplicity of the outputs: fractional credit. Attribution credit is easy to understand for stakeholders who may not understand the nuances of how advertising works.
Modeled attribution often falls short in creating trust with marketers – most attribution models feel like a black box and you don’t really know what assumptions are ‘feeding’ the model. Another weakness of attribution models is the answers given often lack the depth needed for marketers to truly understand performance and optimize campaigns.
What About Measured Attribution?
Using direct measurement to get attribution data challenges the assumptions of the past. Instead of blindly accepting the status quo, measured attribution is rooted in the present instead of the past.
Measured attribution directly plugs into ad platforms and records every impression, engagement, and conversion. Next, all that data is taken for cleaning and deduplication to show campaign performance. This means after the process is complete, about 70%-80% of all campaign activity is being reported on.
One of the main strengths of measured attribution is there’s no need to trust the data that’s being shown. There’s no estimations involved so it feels much more like direct reporting. Another benefit of measured attribution is how detailed the reporting can be. Because there are no models, measured attribution reporting can show more specifically what’s working, what isn’t, and how to improve.
The main drawback of measured attribution is the fact that there are some ad channels that can’t be measured whether that’s closed ecosystems like Amazon or Youtube, or analog advertisements like OOH or broadcast radio.
When to Use Each
Your media mix and conversion event types will largely determine when and how much of each type of attribution to use.
If your marketing is heavy in channels that can’t be directly measured like broadcast television or if your conversion events aren’t reported back to you like CPG, using modeling for attribution will be able to approximate answers that measured attribution can’t. While these answers won’t be 100% accurate, you should still be able to garner insights that allow you to spend media dollars more effectively.
When you’re using digital channels that can be measured or you have online conversion events, measuring attribution data makes the most sense. If you’re mostly using channels like Search, Social, Programmatic, CTV, Streaming Audio, or Direct Mail, measuring attribution data allows you to skip estimations that come with modeling for more definitive answers.
In the end, modeled and measured attribution reports are very different. While everyone has access to the same data, how you process the data can make a big difference.
