Attribution modeling or path-to-purchase analysis? These concepts are often used in the same context -- and, at times, are confused with one another. The purpose of both is to establish optimal touchpoints for consumer interaction with the brand and to optimize individual online channels to achieve the best cost savings and customer experience. So, what is the difference between the two?
Attribution is ad-centric. That is, the unit of analysis used is an ad placement.
Marketers use the data generated from tagging these placements to understand the degree of influence each ad has on the consumer. The main outcome of an attribution analysis is apportioning of the credit for each conversion among the ads. In other words, they look to analyze a consumer that was exposed to an ad prior to converting and then used that information to change strategies accordingly by re-allocating media spend to the ads that had the greatest impact.
Path to Purchase analysis is consumer-centric. The unit of analysis here is an individual consumer, or rather, the string of his or her online actions.
This analysis concentrates on the sequence and intensity of each relevant behavior that may eventually impact a consumer’s decision. Exposure to an ad or a click on a paid search ad are all accounted for, but so are the visits to a competitor's site, a review site, the use of a search engine on an aggregator site, seeing a product placement in a YouTube video, etc. The approach is exploratory - no prior assumptions are made about consumers' behaviors. The clustering algorithms applied to the multitudes of individual paths establish how many distinctive behavioral patterns exist in the mix.
So which approach is better?
This depends on what you value more: promptness or depth.
Attribution management, at its basic level, could be almost real-time, with the results of users' exposure/interactions with ads being immediately processed, weighted, translated into algorithms and fed into media optimization engines. This is a giant leap forward from the traditional "last click or view wins" mentality. But it also presents the risk that one might err in a big way.
In today's diverse and fragmented media ecosystem, paid ads are hardly the main factor steering consumers to purchase. Earned media, online word-of-mouth, price fluctuations, affiliate marketing, promotions, competitive activity (that obviously, cannot be tagged), baseline brand preferences and even seasonality usually remain beyond the frame of attribution analysis. Credits can only be allocated between the entities that can be tagged. This makes a marketer's decision-making process easy - but will it lead to a right decision? Re-allocation of spend between paid-ads will not improve the situation when the right solution would be to take some money out of a paid media budget and apply it to somewhere else.
Some of the most advanced attribution management providers attempt to correct this bias by bringing all the missing factors into their attribution models. The relevant data, however, is not immediately available, it has to be collected, cleaned and matched to the advertising campaign data. This turns the entire process into another media mix modeling exercise - only on a more granular level than traditionally done -- and slows the process to a halt.
An alternative to turning the exercise into a yearly science project is to identify where and when consumers are most open to marketing messages during the purchase process - and then strategically optimize advertising at the stages of consumer decision making when ads matter the most.
The Path-to-Purchase approach serves such a purpose. By looking at the entire industry rather than at an individual advertiser's campaign the path-to-purchase analysis is able to discover previously neglected touch points that are resulting in a loss of prospective customers to the competition. With this, one can than re-allocate marketing resources to remedy the situation.
For example, when analyzing consumer shopping for an expensive household electronics product, we extracted a typical path followed by 65% of all shoppers. It was ripe for effective marketing intervention, as four out of five consumers on that path went through only three steps. Influencing even one of those steps could really tip the scales for a brand. This was especially true as the average time on the path was not short at all, indicating that in-depth researching was going on. By staying on the top of consumer's mind during that time and by applying an intelligent creative strategy, a marketer could tip the scales for any brand.
On the other hand, influencing consumers who are embarking on the next most popular path was not up to the banners or paid search keywords. On that path, retailer sites drove continuous traffic to original manufacturers' sites, with little opportunities to serve the ads in between. It was feasible however to shorten and ease the path for consumers by creating “brand stores” at the aggregators (e.g. a “Samsung” store on BestBuy.com) to speed up the conversions.
So, attribution modeling can be very effective -- but only when the larger picture of consumer behavior is known.
We find no problem with clients who use "last click" and "overall exposure" measures - both have their merits. But I'm puzzled by how you could possibly analyze a system but make "no prior assumptions" about consumers' behavior. Seems to me this gives you an impossible task because there are literally thousands of variables that *might* affect customers - global warming, politics, phase of the moon, etc.
Yaakov, Your definitions for “Attribution as ad-centric only” is incorrect. In fact, path-to-purchase is subset of attribution. A good attribution methodology looks at wide variety of attributes such as consumer behavior, ad performance, changes to marketers business (e.g. new product launch, bad press etc.), competitive pressures, exogenous variables such as Pete’s examples (weather, politics, global warming etc.). Seasoned attribution systems can figure out what attributes are important to each particular marketer and what are not, before modeling. For example, weather is important to one advertiser but not to other. Data can tell all these. This way, you don’t have a never-ending-project, but you have a short, cost effective, time capped project that can produce results that you can use it right away. I am speaking from my experience with a wide variety of agencies and advertisers where I have evaluated most attribution providers out there. The best place to look for strengths and weakness of attribution providers, you can read the Forrester wave reports published on Digital Attribution and Cross Channel Attribution. Hope this help.
Great post, Yaakov! Very insightful.
Pete, the clients who are satisfied with the "last click" approach should probably be told that it holds no merit whatsoever - beyond, perhaps, the sweet bliss of ignorance. There is simply no honest alternative to analyzing complex dynamic systems with multiple factors affecting consumer decisions. Fortunately, there are very powerful techniques to separate signal from noise, despite the multitude of potentially important variables.