Part 3: What Your Attribution Model Isn’t Telling You

The many names and faces of digital attribution. 

The first two parts of this series covered the different tracking implementations available and the attribution models used to assign credit across channels. This final post is about what to do with that information once it exists - specifically, how to map actual value onto different stages of the funnel in a way that produces a more complete picture of ROI. 

Before getting into the methodology, it's worth addressing something that sits upstream of it - setting the right goals for each campaign type. Attribution data is only as useful as the objectives it's being measured against, and one of the most common ways it gets misread is by applying the wrong success metric to different stages of the funnel. An awareness campaign exists to generate reach and introduce a brand to new audiences – trying to judge its performance by the number of direct conversions it produced is going to give you a hard time. A retargeting campaign exists to convert existing intent - measuring it on reach would be equally pointless. The goal is to define what success looks like at each stage before the campaign runs, and to resist the pressure - internal or otherwise - to track every activity against the same bottom-line metric. Doing this properly is what makes the funnel valuation framework below meaningful rather than cosmetic. 

The basic principle 

The idea is straightforward enough: if there's a known average value attached to a conversion, it's possible to work backwards through the funnel and assign a proportional value to every interaction that contributed to it - clicks, impressions, micro-conversions, and anything else being tracked along the way.  

When you’re trying to justify the spend level of an upper funnel campaign by reporting how many direct conversions it’s produced, you’re going to have a hard time – this method turns abstract engagement metrics into something with a number attached to it, which tends to make conversations about upper-funnel investment considerably easier to have and also works hand-in-hand with value-based bidding models. 

It's worth being clear that this isn't something that can be done from a standing start. It relies on aggregated historic performance data from comparable campaigns - conversion rates, click-through rates, average CPMs, customer lifetime value - and improves in accuracy over time as more data is collected. The first iteration will be a rough approximation. That's fine. Every subsequent campaign makes it more reliable. 

How it works in practice 

The process is similar to budget forecasting. Starting with a known value – typically a conversion value, or ideally an average customer lifetime value - and reverse-engineers the funnel from there. Using historic conversion rates gives the number of clicks needed to generate a single conversion. Using historic click-through rates gives the number of impressions needed to produce that volume of clicks. Combining this with average CPMs gives a rough indication of the budget required to hit a target number of conversions. 

Running the same calculation but focusing on value rather than volume gives an estimated rate that can be applied at each stage. The example below illustrates the basic version of this.

To further expand the picture, this can also be applied to leads, micro-conversions - newsletter sign-ups, specific page visits, product saves, social engagement - and adjusted to reflect different campaign types at different stages of the funnel. A brand awareness campaign measured purely on conversions will always look like it's underperforming. Measured on the estimated value of the impressions it generated, in the context of what those impressions are historically worth to the bottom of the funnel, the picture changes. 

A note on what this is and isn't 

This method doesn't produce a definitive ROI figure. Nothing in digital advertising does, regardless of what a platform's reporting dashboard might suggest. What it produces is a structured and consistently applied estimation - one that gets more accurate over time, that can be stress-tested against actual outcomes, and that gives stakeholders a rational basis for funding activity higher up the funnel rather than concentrating everything at the point of conversion. 

The number on the dashboard is rarely the whole story. The goal isn't a perfect figure - it's a consistently honest one. 

Wrapping up the series 

Across these three posts, the picture that emerges is a fairly consistent one: digital attribution is imprecise, the tools available to measure it all have trade-offs, and the platforms responsible for reporting on it have a financial interest in their own numbers looking good. None of that makes attribution useless - it makes it something that needs to be approached with a clear understanding of what each method can and can't tell you. 

The practical takeaway is this: use the most complete tracking implementation available - server-side where possible. Choose an attribution model that reflects the full shape of the campaign rather than only flattering the last touchpoint. Build a funnel value framework from historic data and iterate on it over time. And when a platform's reporting looks suspiciously clean, it's worth asking who wrote the marking scheme. 

Attribution will never be a 100% certain. The best version of it is an honest question, asked consistently.