The origins of multi-touch attribution (MTA) were in the digital space, as a result of advertising spend transitioning away from traditional “offline ads” to digital media and channels which were deemed to be more accountable. Journeys within a client’s website, or between websites, could be stitched together and the resulting orders joined back to customers and their orders.
Now that the removal of third-party cookies is going to remove much of the stitching between websites, unless you are working with a collaborative data solution that allows this to take place, you are left just with customer journeys within your own website in full view.
These journeys may however include affiliates, referrers, digital ad campaigns, PPC, and direct search so you will at least know where the visitors came from if not the ad impressions they had been served to get them there.
However, we believe that there is still a great deal of merit in MTA, but not when it is restricted just to online events. (There is also the huge consideration that some of the large analytics platforms use sampled data and not 100% raw data which means the more you dig, the less you see).
Estimates vary about how much of advertising spend is digital, but the consensus appears to be around 55% currently, and that leaves 45% non-digital which is clearly far too much to ignore, much as Google would like us to. We also suspect that with the removal of some of the programmatic advertising volume, the digital proportion is likely to reduce down, perhaps to around 50%.
There is absolutely no reason why MTA should ignore the non-digital channels; but it means that you require the technology to joint it all together at a customer and order level. This is most effectively achieved using a customer data platform which is specifically designed to join browsing activity with off-line into a single customer view.
The non-digital ‘touches’, we prefer to call them ‘events’, can for instance include emails opened, text messages, call centre contacts, retail visits, and direct mail. These are all direct events, but on top of these are non-direct advertisements such as TV, which we discuss below.
There is a lot of unscientific opinionizing about the best approach to weighting events before an order. We are confident that we have found a reasonably good statistical solution for this. It uses a mix of Markov chains and survival curve statistics to give the weightings to any specific set of events. This approach does not presume anything about first or last touch, but rather looks at the evidence presented when the events have all been joined together in a single customer view, together with your customer and order data.
To deal with the non-direct channels like TV, the more ambitious will also want to build econometric models which reveal the overall effect on demand of all channels, direct and indirect, when working in combination. Econometric models often get a bad press as being unresponsive to short term changes in consumer behaviour, and not being granular enough in their spending recommendations, but they are the best tool we have to give the non-direct media their fair share of the credit for sales made.
Techniques now exist to align econometric models with multi-touch attribution so that, in effect, value initially credited to direct channels can be reattributed back to the indirect channels; this usually has a significant influence on the overall share of demand given to the direct channels.
So, to present what we have been describing diagrammatically, a full attribution process is going to look like this:
One of the often overlooked, and we believe very significant benefits of MTA, when it is sitting on a single customer view, is that it can be cut by customer type. The simplest cut is to distinguish between what is bringing new customers versus existing. But the cut can be for any customer segment, like high value versus low value customers, or purchasers of particular types of merchandise.
Multi-touch attribution can tell you a lot about how your marketing works, but only when you look at all of your online and offline channels in combination. And for many advertisers using indirect channels like TV, then it becomes important when possible to align MTA with econometrics.
In so far as we only look at events prior to a sale we will learn nothing about what doesn’t generate a positive outcome; however, if we take a look at all browsing events, we can start to examine the probability of an event leading, or not leading, to a sale.
There are two reasons why this is valuable information, although unfortunately often ignored. First, because knowing the probability of say a Facebook advert leading to a sale brings a sense of realism about advertising there, but also because serving people adverts in which they are not interested does damage to your brand.
Back in the heyday of direct mail, people were so fed up with the quantities that kept on arriving that they called it junk mail, and often had stickers on their post boxes asking for it not to be delivered. (Unfortunately, the postman had no choice but to pop it in their box).
PPC Protect estimates that in 2021 the average person (we assume in the US) will see up to 10,000 ads per day, whereas in 2007 estimates were only at 5,000 ads per day.
Common sense suggests that this must be way over the top of what is either necessary or enjoyable, and people will increasingly assert their objections to it.
Clearly brands that focus on the relevance of their advertisements will create a much more favourable impression than those that just focus on volume.
We have started to investigate browsing behaviour in terms of its likelihood to lead to a sale, with the following result:
To explain how this table works (and it was built using actual online and offline event data) it shows the probability of a person moving either from one event channel to another, or to a sale. So, if you start with picking a channel on the Y or vertical axis, you can then move along the row to view the probability of a customer moving to the next browsing state. For instance, someone coming to your website from a social network has a 96% probability of doing nothing further, and a 0.55% probability of being converted to a sale without engaging with additional channels. They also have a 0.67% chance of moving next to a search engine, whence they will have a 3.3% probability of making a purchase. However, someone receiving a campaign has a 6.5% probability of conversion without using other channels, and a 6.7% probability of moving next to a search engine.
So, in conclusion, we suggest that there is a strong role for multi-touch attribution, post third party cookies, with or without econometrics, and another new role for data science in investigating what we might call dark advertising, the stuff you see, but which makes little or a negative impression.
Read more about how Unifida’s marketing attribution works and what it can deliver.
UniFida is the trading name of Marketing Planning Services Ltd, a London based technology and data science company set up in 2014. Our overall aim is to help organisations build more customer value at less marketing cost.
Our technology focus has been to develop UniFida. Data science business comes both from existing users of UniFida, and from clients looking to us to solve their more complex data related marketing questions.
Marketing is changing at an explosive speed. Our ambition is to help our clients stay empowered and ahead in this challenging environment.