What is the role of attribution modelling?
Metrics to demonstrate the impact of an advert are increasingly redundant in a time of social media but attribution modelling may fill the gap
Historically, one of the biggest problems facing brands was that they didn’t know who their customers were and what was driving them to buy. They didn’t really understand who was watching their adverts or how each customer responded. This hampered their efforts towards the ultimate goal of encourage them to return time after time.
Fortunately, payment technology and the Internet have made consumers less anonymous. Customers can now be identified in stores via loyalty and credit cards, as well as customer relationship marketing programmes, while search engine optimisation and search engine marketing have long been added to the marketing lexicon.
Metrics such as cost per click (CPD), cost per thousand impressions (CPM) and cost per action (CPA) have meanwhile helped measure the effectiveness of investment in online channels. A whole new industry (and dictionary of acronyms) has been created.
If they see an advert, then search for something and then sign up for your email, instead of ascribing all the value to that last event, attribution modelling tries to work out what other elements contributed to that email being signed up for or to something being bought and to allocate value up the marketing chain
Despite all this, brands still face a modern version of their age-old problem. Search engine technology may tell them how customers arrive at their online checkouts. But they don’t provide brands with much knowledge about all the steps customers take along the way or whether the investment that retailers put into this is an efficient or effective use of resources.
To solve that riddle, another technique has been developed, under the buzz phrase of attribution modelling. Defined as a method of identifying a set of user actions that contribute in some manner to a desired outcome and then assigning a value to each of these events, it has been around for about four years and is gaining in currency and use.
‘Most forms of digital marketing have traditionally been measured on someone clicking on something and doing something online: the single interaction that someone takes,’ explains Farhad Koodoruth, managing director of digital and public relations agency Threepipe.
‘What an attribution model enables you to do is to measure and allocate a value to each step that someone takes. If they see an advert, then search for something and then sign up for your email, instead of ascribing all the value to that last event, attribution modelling tries to work out what other elements contributed to that email being signed up for or to something being bought and to allocate value up the marketing chain.
‘It’s about trying to get a much better understanding of the exact impact of online advertising and what it does, rather than just trying to measure a single channel or a single set of key words.’
As with most things to do with measurement, the answer is rarely simple. Because all consumers’ buying cycles for different products are different, there is no set attribution model, so a great deal of analysis has to evaluate what models make sense for individual brands, how to weight between different priorities and where to ascribe value that produces an improvement in results.
The easy way has been to focus on the information provided by Google and by search engine optimisation firms, since this is simplest to measure.
However, experts say online searches are increasingly becoming the last step consumers take when they buy something, having already got some notion of where they want to find it, whether through an advert, personal recommendation or interactive marketing. Brands end up knowing little about what actually drives their online sales.
‘One of our findings is that most people over-invest in search because they don’t analyse it properly,’ says Koodoruth.
‘They keep pouring more money into paid search advertising because they think it’s driving incremental value but there’s overspending in that channel and brands get better value by investing in things that are harder to measure like online video, display content or advertorial approaches.
‘Search is easier, big and a bit more transparent. There are natural reasons why it accounts for something like 60 per cent of all online advertising spend but a lot of people are doing it because they don’t know how to measure the other things properly.
‘Display advertising and online video do not drive direct sales as much as search but the uplift they have on other channels is significant and they’re the only way of really driving increased volumes of new customers to the brands.’
Attribution modelling is neither a cheap or quick fix, however, with Koodoruth admitting that it takes ‘a good year’ to get to a point where you have really robust data that tells you what’s really happening and realistically probably starts at between £60,000 to £100,000 for the technology alone.
Nevertheless, Threepipe says it has resulted in sales increases of up to 25 per cent for retail clients including Laura Ashley, Reiss and Sweaty Betty.
A different type of attribution modelling can also help brands know what personal recommendations on social networking websites are driving their sales, where they are coming from and how to influence them.
A recent Gartner report found that recommendations from friends are now the number one reason that people buy online but such recommendations are not as transparent as they seem. While any company with a Twitter monitoring function can find all the mentions of their brands, it is harder to track recommendations on some of the other social network sites. And recommendations posted on Amazon.com, for example, don’t give retailers much information on the people posting them or why consumers have bought the products.
We can show users what products are trending, who their top advocates are, where their products are being shared and ultimately what’s being bought because of recommendations
Peter Janes is managing director of Shopa, a new social shopping website that has signed up 400 retailers and is aiming to revolutionise the way that people shop, having spent 18 months developing attribution modelling technology.
‘We have four patents on attribution technology. It’s the premise of everything we do,’ he says. ‘Although we’re a consumer-facing marketplace, the back end that the business is built on is all about attribution modelling. ‘What’s increasingly happening in retail is that sales are coming through recommendations on social media but it’s impossible for retailers to share a banner advert or Twitter or attribute affiliate links through Facebook.
‘They can use Google Analytics to find out where the traffic is coming from and the quantity of that traffic. What it can’t tell them is why that traffic has come to them so they don’t know which part of their marketing has been effective.’
Shopa, which holds no stock inventory itself, pulls in products from its retailers on its portal and allows the brands to know where the social media recommendations for their items are coming from.
‘We have the ability to associate an individual with a recommendation and to be able to track that recommendation across any platform, whether it is Facebook, Twitter, Pinterest or something else, on any device in real time,’ says Janes.
‘We can show users what products are trending, who their top advocates are, where their products are being shared and ultimately what’s being bought because of recommendations.
‘We can attribute an individual with a referral, which has not really been done before. All this sort of attribution modelling technology is now invaluable to brands. We can see what is being promoted and where it’s happening and filter all that back to retailers or businesses who work with us. And we do it all in real time.’
Attribution modelling is also happening in online publishing, where business magazine Forbes has been pioneering a new type of journalistic model with contributors who post five articles a month and are paid based on the number of views generated. (Disclosure: the writer of this article is a Forbes contributor).
My favourites are dwell time per post, total dwell time for all posts, the percentage of author pages that readers scroll, scroll velocity and scroll depth
Lewis DVorkin, chief product officer at Forbes Media, says the latest measurement technology can attribute not only how many page views and unique visitors each post generates but also how often 50 per cent of an advert is viewed for one continuous second.
He says advertising groups are using such metrics to reduce advertising rates in a development that threatens to chop one-third off annual digital revenues of media companies.
Forbes has added its own metrics to help it compete. ‘For each of our 1,500 authors,’ DVorkin wrote in a recent post, ‘I get to see 11 data points, per post and in aggregate.
‘My favourites are dwell time per post, total dwell time for all posts, the percentage of author pages that readers scroll, scroll velocity and scroll depth.’
DVorkin foresees the day when Forbes might remunerate its contributors based on such metrics, ushering in what he calls an ‘era of super-journalists’ who excel in such measurements and are better bets for publishers who want to access readers.
That’s enough to send shudders down the spines of many writers and some retailers are also viewing attribution modelling technology very cautiously.
Matt Roberts, digital marketing controller at Home Retail Group, the retailer company that owns the Argos and Homebase chains, says: ‘Argos has successfully deployed SEO and SEM [search engine marketing] technologies for a number of years.
‘We are aware of new technologies for attribution modelling and are currently evaluating our options in this area. No commitments have been made at this time.’ The evolution of the super-retailer in the information age has much further to run yet.