'Harnessing Attribution: Breaking Through Silos Via Models & Algorithms', by Jon Baron, CEO, TagMan

Among the channels within the media mix, digital marketing is the most measurable and offers the most potential for effective analysis of what is and isn’t working within your marketing. Every dollar can be reported on for ROI and this has enabled search campaigns to be optimised, with strong keywords being identified and exploited. It enables display campaigns to be reported on, to determine if a publisher really does have a suitable audience for your company or product.

Break free from the silos
The problem is that this measurement is only done in silos, and consequently any optimisation that results can only ever be partial and one-dimensional. To date, a lot of measurement is based on the “last click wins” attribution model, meaning the final touch point before a sale receives the credit, largely undermining many other channels’ valuable contributions to a sale. Also, many companies come to rely on audit tools which measure performance that includes that company’s own campaign performance, therefore, gaining truly independent and impartial reporting is the foundation to ensure optimisation is channel-agnostic and reaches the full potential possible.

Hidden Value
Attribution enables marketers to see performance as a whole, rather than as a set of linear channel results. By giving all channels their own value, marketers can clearly see the hidden value of their various investments, for instance, if you knew certain display ads were more influential at starting high-value conversion paths, what would you do? (Hint: you would use more of them and would also reallocate budget to spend less money on poorer performers. Also, understanding the kinds of attributes those ads have that makes them special provides a creative blueprint for future campaigns). In short, attribution provides intelligence, performance and competitive advantage.

So, how do marketers get started with intelligent attribution? To harness the power of attribution, marketers need to change approach and look past the skewed “last click wins” model to gain serious value.

Modelling vs. Models
There are two main paths that can be taken when it comes to attribution.

Attribution modelling is about reporting and insights. The goal is to drive better understanding of the influence and value of each channel and campaign. It can include offline data, competitor activities, weather, macro conditions, etc.,as well as digital marketing data, such as the study of audience behaviours. It gives marketers the chance to try out different models by using past data, and without disrupting the entire marketing mix.

Alternatively, attribution models are business logic that is applied in real-time with the goal of controlling the delivery of tracking tags. This affects in reality which marketing vendors get attributed credit for every sale. Implementing an attribution model is vital to de-duplication and, in particular, ensures that marketers don’t end up paying a CPA to multiple affiliate networks for the same sale.
At TagMan, we see how much money is saved through de-duplication, and we see clients reasserting control over their data and ensuring that only the correct data is being transmitted across the web to vendors. This is achieved with an attribution model that governs control over tags and the data they receive and when they should fire.
How do I decide which model to use?

How do you decide what is right for your business? It is not as simple as ordering off a menu, as eventually every business and its needs are different.

Attribution modelling in this sense can be implemented to derive insights in order to better understand your marketing mix, and then fine-tune the attribution model that your Tag Management System will eventually implement.

More and more advertisers are now simulating a variety of attribution models over their data in order to see performance re-calculated; true performance shown in a completely new light.

For instance, if you evaluate your display campaigns by which creative drives the most first clicks in conversion paths then you are measuring them on a level playing field. You are measuring display campaigns for the job you are hiring them to perform, i.e. driving awareness. This may sound simple but it’s a great first step in moving beyond last click wins reporting. It provides immediate actions you can use to optimise activity.

Understanding channel synergy also unlocks opportunity. When you can report on which channels are most effective at influencing rather than closing the sale, you can determine how to move budget around so that you ramp up the activity that is vital to pushing users towards a purchase.

Understand your consumer behaviour
When you can demonstrate that a channel clearly contributes to assisting and influencing purchasers, then you can ensure you invest your time and resources in those areas. Measuring performance on ‘last click wins’ will keep you in the dark, and you’ll miss out on the opportunities your competitors are seizing.

Let’s say you have some campaigns you are using to drive new customer acquisition. If you could prove that the users you’re reaching are already consistently clicking through SEO Brand, then you know you can reallocate that budget, as they have effectively already been paid for.

What about the algorithmic approach?
The alternate approach within attribution modelling is the algorithmic approach. This approach is more of a black box that you feed your data into, which in turn crunches the data and produces optimisation recommendations. It’s a machine-learning approach that interprets all the attributes of each interaction and the bigger and deeper you go with it the greater the accuracy of its output. The approach can be likened to how music players such as Last FM and Pandora have worked with their users in the past: the more you skip a certain type of song, the software’s algorithm remembers your taste and serves up better music results, which are tailored to your liking.

But remember, algorithmic attribution only works as well as the data you feed into it. Getting a solid foundation in place and using high quality data are key to unlocking the real potential of attribution.
Algorithm-based attribution modelling is certainly a more heavy-duty, expensive and advanced approach but it effectively harnesses the power of today’s processors to build full and comprehensive optimisation plans.

Which is the right one for you?
Firstly, ensure you are tracking all campaign activity under one, unified system. Just getting all the data together so you have a single view of the consumer’s journey is a huge first step.
Secondly, ensure you are de-duplicating sales and are not reporting back the same sale to multiple vendors. Using a tag management system with robust attribution functionality is important.

Thirdly, decide which approach to modelling is the right one for your business. If you are new to attribution, then take the ‘test and learn’ approach: embark on your first steps into rule-based attribution modelling where immediate benefit can be unlocked. Discovering a number of optimisation opportunities becomes inevitable from that point. rom there, hand it oFver to the machines…