Programmatic Attribution Modeling – Don’t Just Measure, Execute

The goal behind an attribution model is to understand how all of your marketing touch-points fit together, showing you which dollars are working the hardest, and subsequently allowing you to make smarter decisions next time around.

Sounds simple, right? Actually, it’s very far from simple, and is actually the wrong goal to be aiming for.

Firstly, most of the barriers to building an attribution solution are extremely hard to overcome, and others are simply impossible to accomplish. Technical limitations mean we can’t tie together all of our digital touch-points, human limitations mean we can’t get everyone to agree to how the model should work, and simple life realities (such as real face-to-face communications) mean we can’t account for every day-to-day interaction.

Programmatic Attribution Modeling is a solution that moves us beyond these barriers, and focuses us back on the idea of taking action based on data, not just making sexy graphs. But first, let’s understand why this fresh approach is really needed.

Technical Limitations

Where do we begin?! To make headway with an attribution model, we must be able to account for as many touch-points as we can, with at least a reasonable sample size of the targeted population.

Some of these are simple. It is not uncommon to bring together all display buys under one cookie by using a buying platform for real-time media, and that same cookie can be served on premium/direct display buys, too. This could be your ad server, or it could be a demand-side platform (DSP), a data-management platform (DMP), or better still, a PMP (Programmatic Marketing Platform) that combines all these technologies.

And, that cookie can also be matched against the clickers from your SEM program, with the same tools, or with on-site analytics. Tag container companies like BrightTag, TagMan, Tealium and others can help further, and some take you to a deeper level still by helping to bridge more vendors that might be on your plans.

The next step would require pure, sneaky “smarts.” Using Pinterest or Facebook in your marketing? Use your PMP to look for incoming traffic (and cookies) from those sites and add that individual’s interaction to their universal profile, knowing they have pinned or friended your brand.

At this stage, we are still missing everyone’s offline exposure to the dataset, as well as interactions with real-life influencers. In the future, perhaps the presence of RFID chips in phones (and in people ) will help overcome that; but, until it becomes widespread and measurable, we need to settle for broad, geographical trends based on the investment of advertising at the DMA level to get a partial solution for this piece.

Human Limitations

Even though there’s been progress made on the technical side, your organization must agree on what to do with that data. Typically, CMOs find this a bigger barrier than lining up the technical aspects described above.

Imagine the meeting to determine an attribution model between the CMO and the leads for Display, Analytics, SEO, PPC and Social. Display typically buys on a CPM and measures with view-through; PPC buys and measures on a CPC; SEO and Social are both producing content and monitoring their effectiveness with clicks and organic visitor growth; and, Analytics can only monitor on-site behavior and incoming click traffic, ignoring influence from exposure elsewhere. And, let’s hope no one invites the Traditional media buyer, too!

Each party’s motivations differ from the others’. Some favor a last-click situation getting credit for what brought each conversion over the line; others spend their time further up the funnel driving demand, and as such, want first touch to be the deciding factor. And, channels like Social play throughout the process and argue that organic, earned media always outweighs the value of paid media and should, therefore, be given more credit.

Get Something Done!

It’s an impossible scenario for the CMO to solve, but the worst thing to do is to do nothing. I have seen many models used over the years, and the one that seems to have the fewest limitations is Forrester’s (PDF) model, shown below.

In such a model, each touch point has value — the one further back from the time of conversion gets the least credit, the one closest gets the most. It also has the added value of being from an independent third party; and, if presented to the Impossible Group of your heads of channels, it’s hard for them to propose a much better alternative that considers each other’s programs.

There are companies that will help you with such things, too. At Chango, we have seen our clients have success with the likes of Adometry and ClearSaleing; just don’t expect them to be able to work magic if you haven’t started solving the technical problems.

But It’s Still The Wrong Point!

If an attribution model is there to help us make smarter decisions, then let’s create data that result in immediate action. The outcome of all of this hard work will tell you which channel should get more budget and which should get less, but the experience for each individual won’t change, and neither will the investment in your most valuable prospects.

Worrying about an exact division between channels shouldn’t be the focus – instead, concentrate on the now, and concentrate on the exact individuals you are trying to influence.

Programmatic Attribution Modeling is another benefit of the big data movement (“big data” simply being lots more data than you use now, and “programmatic” being the mechanism to make it actionable). In such a case, the media buy itself is programmatic, meaning that at the exact point the decision to buy is made, all the most recent data is queried and an assessment made as to whether that person needs another touch point — and, if they do, what format should it be and with what creative execution?

Programmatic Attribution Modeling only works if your buying platform is processing all of the above types of data; and if it is, then you can execute without the typical wastage, building an attribution model of sorts for every individual, not just the broad population.

Today, the executions are typically display media, with some mobile and tablet thrown in. But, PPC is around the corner and billboards, radio, TV and the like will follow suit in the future.

Go Away And Do Something

Now, it’s your turn. Go look at what you are doing today. Is every channel optimized to individuals, or to broad decisions about media planning, PPC budgets and your latest social campaign? Are there areas where you can make a more precise decision about an individual today, rather than waiting for a population-level model in the future? Cross-channel is cool; but as marketers, we are speaking to people, not spreadsheets.