The best thing about digital marketing and display advertising in particular is that everything is trackable and measureable. You can see the impressions of the ads, the clicks that are being driven to the site and the revenue, hopefully, pouring in.
Performance can be broken down by publishers on the plan, Remarketers and Prospecting, Facebook and Display. But the questions we need to ask are “What channels are really working well? Which channels are bringing in the revenue, and what part of your budget is being lost into the advertising ether?”
Although the below looks like a weirdly odd barcode, these are Google images for attribution modeling.
Last touch, first touch, linear and weighted. These are some of the most common attribution models used to judge what is driving the revenue and what is driving…well, not a lot. It must be noted that a click always trumps a view. So if a customer clicks a Publisher 1 ad then sees a Publisher 2 ad, then goes on to convert, the sale will be attributed to Publisher 1. Finally the default look back window is normally set at 30 days, anything further than the 30 days after seeing or clicking an ad, no conversion will be attributed. Simple? Then let’s continue.
Last touch attribution is arguably the most commonly used metric. This gives the publisher who had the last touch the sale. But is this the most effective way to judge sales? Across a digital marketing plan there may be a couple of retargeting publishers and the majority will be prospecting publishers.
Searching out new customers that have not previously visited the site and driving them to the site and pushing them on to conversion is always going to be harder to do than targeting users that have already been on and browsed the site, then pushing them on to conversion.
Therefore, on a last touch attribution the prospecting publisher is never going to perform as well. On top of this, the problem is exacerbated by the fact that if a customer is brought to the site by a prospecting publisher, they will no longer be able to be targeted by the prospecting publisher if they leave the site and will only be targeted by the retargeting provider.
If the customer does not convert on the first visit they will most likely be retargeted and the conversions go to the publisher that does the retargeting. Is that the fairest attribution when the prospecting publisher originally drove the customer to the site and introduced them to the brand?
The CPA therefore for retargeting should always be the lowest on a last touch basis.
First touch attribution at first sight solves the issue for prospecting activity. The first ad that was seen or clicked before the user went on to drive the conversion is relatively simple to look at. However, is the first ad the most relevant? What if a users is shown one ad from prospecting publisher 1 and then 28 days later sees ads from prospecting publisher 2 then goes on to convert after 1 day.
Surely the sale should be weighted towards publisher 2? Also using this method, retargeting could possibly not end up getting the full attribution that it deserves.
The path towards a conversion for display is also a lot more complicated than originally thought. To make things trickier customers can be shown numerous ads from different publishers multiple times along the conversion funnel. So perhaps a linear attribution is more helpful for the larger media plan? For example if a customer has seen 6 ads on the way to a conversion, the 1 conversion is divided by those 6 impressions and assigned 0.17 of a sale.
But really, with no weighting or time scale involved in this for example a users seeing an ad 30 days ago being given the same weighting as an ad seen 1 day before purchase, this does not really seem the fairest attribution either.
What is key is that the attribution model that is used is understood. The effect that the attribution model has on retargeting, prospecting and brand display activity is always going to be different so it is worth thinking about the effects before it is decided whether partner A or B is performing particularly badly or well. It can be worth running two different attribution models together and comparing the two. It can be surprising to see the difference that each attribution model makes.
Finally one thing to make sure is that there is enough data to make a decision. There is no point in number crunching 5 sales for 5 partners and looking for the best performer. This is simply not enough data to get any results. A couple of sales more or less per publisher would wildly swing any performance either way. With digital data it will have to be good quality and high volume.