As analytics and data capture in general have improved over the years, more focus has shifted on capturing and understanding the performance of marketing campaigns in order to capture the return on investment (ROI) for each marketing channel. It became possible within most analytics tools, such as WebTrends and Omniture SiteCatalyst, to capture the referring traffic into a web site; site owners became more intelligent in the way they captured online sales information. Tying the two together enabled sales to be plotted against different campaigns. But this was not without its challenges.
The arguments start when you consider that a visitor may have previously been to the site using an alternative marketing channel. Suddenly, the question shifts to which marketing channel was the one which was really responsible for the sales generated on your website. Was it the first to send a visitor to the site? Or the most recent? The truth is that it is almost impossible to tell, since in most cases a purchase will be as a result of a number of visits, each as important as the last. Thus we turn to attribution modeling in an attempt to credit each marketing channel with the correct sales and revenue amounts.
Start to Attribute Campaign Value
In truth, the above examples are themselves basic attribution models, with first-click and last-click models still widely used to attribute sales to individual campaigns. However, we have moved on to attribute credit across all (or most) visit sources across a customer’s lifecycle. This allows all campaign sources to be considered for ROI credit, enabling us to fully understand the impact of all of our channels.
Consider for example a customer who searches for your brand online. They choose to visit the site via PPC, on a brand term, and then leave without purchasing. They then visit again, having used an affiliate site to compare prices for your product. Having found your brand to offer the cheapest price, they then leave to do more research. Finally, they enter via natural search, having already identified your brand as the place to buy their product. In this case, first-click attribution would tell us that this was a PPC sale; last-click attribution tells us that this was a natural search sale. But surely the affiliate comparison site had a part to play in this sale?
In light of this, we have moved to a number of multi-attribution models; these are models whereby the credit is given proportionally to multiple referral sources. Initially, it is possible to allocate credit uniformly across all marketing channels. Thus, if a visitor saw 3 different campaigns before buying, each of the campaigns would receive credit for a third of the revenue. This is only a starting point however, as we move towards more bespoke attribution modeling.
Attribute Campaign Value Based on your Bespoke Requirements
The best attribution model to use is always one based on your specific requirements. By this, we mean a model which attributes value based on the key goals for your website. There are some common attributes we can use for this, including:
- Visit recency
- Visit duration
- Number of pages viewed in the visit
From here, the key actions are to supplement this information with site specific actions.
A model we created recently captured this theory superbly; The website itself was a transactional sales site, so the attributes above were essential within the attribution model. However, to supplement this, we added product view (but no purchase) information, registrations and newsletter sign ups to help us identify where the true value existed for each sale. As a result, the client can now understand in much more detail the true value of each marketing campaign. Whilst revenue is the ultimate indicator of ROI, adding a weighting and subsequent value to each of the other site actions provides a more accurate understanding of where the true value lies.
How Would That Model Have Looked?
In order to attribute the revenue as required, a model was built for each sale to share the revenue across the referring traffic sources. This model was based on the key attributes associated with each visit in the stack, allowing us to attribute the value in detail to each campaign source. The results seen were similar to those shown below:
As a result, there are significant differences to be seen based on these different models. The differences are significant, as highlighted in the graph below, with the graph showing the values attributed to each of the campaign channels based on the models discussed.
Looking at these differences, it is clear that the model used must be the right one for your business, based on the attributes important to your website.
So How Do I Capture This Information?
A number of attribution products currently exist in the marketplace, but most are based on a number of pre-set models which can’t be altered in any way. Therefore, they quite often do not allow the level of customization required to show the models mentioned above.
For which reason, we work with our partner SiteTagger to quickly implement functions to capture the information required and report it accordingly. Primarily, this involves the creation of a marketing channel ‘stack’, stored as a first party cookie, that allows us to capture the full referral history at point of sale. Given the bespoke nature of this approach, we can then attribute different values for individual events against each member of the stack, allowing us to build up the information required for even the most advanced attribution model. Similarly, we can control quickly the cookie expiry, number of referrals and other such control mechanisms within the coding provided.