Online customer behaviour keeps changing and single visit purchases become more and more unlikely. This creates a particular challenge in attributing the steps prior to conversion to the correct traffic acquisition source.

This sounds both drastic and very complicated. With a brilliant phrasing from Avinash Kaushik a particular solution referred to as “first click attribution” has the following downside: “First click attribution is a bit like crediting your first girlfriend for your current marriage.” Sounds odd and funny, but hits the nail on the head.

Similar counter arguments have been brought up for last-click attributions: they ignore any attempts to create Reach and brand awareness, and although quite a lot of money is spent on brand awareness, these campaigns are not even expected to directly contribute to revenues up-front.

There are yet two more problems with the concept of revenue attribution – and the first one lies straight in the nature of the term itself: “attribution” comes from the Latin word attributionem (nom. attributio) and denominates an assignment. In other words: the act of attribution points towards a person (or a group of persons), not to the core or the object behind. It is (and always was) an anthropocentric concept, nothing objective or measurable.

Let’s have a closer look at the second problem. As we’ll see: any attribution logic is tying itself to another particular (though: hidden) attribution. And this leads to an interesting effect:

Let’s assume somebody needs a hard drive with a set of two particular specifications. A very common approach for “scanning the market” is to insert the needed specs together with the term “hard drive” into a Search Engine.

Besides particular manufacturer sites, offering an exact match on all three terms, it’s very likely that one or more retailer sites show up. Depending on the prevailing investigation mode the user has in mind both clicks are equally likely to occur (for the sake of simplicity let’s say: 50 % probability for each).

From here the subsequent path bifurcates: the person arriving straight on a manufacturer’s site may not come to a purchase decision during the initial visit and is rather repeating the initial search to get comparable products from other manufacturers. The person looking at the portfolio first may select two or three products to scrutinize them further on the manufacturers’ sites.

In both cases a visit on a manufacturer’s site happens as an result of a Search Engine visit. One prospect may arrive via a branded term (manufacturer/model number), the other may arrive via non-branded terms (i.e. a size or feature specification).

Both visits on the manufacturer’s site don’t lead to a purchase, but the interesting point is that both visits on the manufacturer’s site are ending in opposite directions: the first visit ends towards a Search engine to investigate alternatives, the second visit may be geared towards a retailer site to perform a price and availability comparison. Both visits may as well just end without a dedicated exit direction, just with a session expiry.

Crediting the acquisition source “Search Engine” regardless of the “search mode”, seems to be yet another way to distort results from revenue attributions. Even worse: conceding that information retrieval patterns on the web may have predetermined stages, but arbitrary sequences, duplicates the original attribution problem.

The original sequence of A(ttention)-I(nterest)-D(esire)-A(ction) can easily be re-written as I(nterest)-A(ttention)-D(esire)-A(ction) under today’s conditions, as information retrieval via Search Engines starts with a dedicated “pull action” in all cases; and each stage of the sequence can even be repeated ad infinitum (particularly when we consider smart shopping, aka bargain hunting).

In all cases, current revenue attribution models are desisting from the arbitrarity of the prospects’ original intent. First-click/last-click attributions are jumping too short in all cases, as they overemphasize only their particular logic.
Unfortunately, linear attribution models are equally ill-suited, as intermediate steps may occur on third-party web sites which are unaccountable for any revenue contribution models.

To wrap up: the idea that a refinement always is more elaborate and precise than any previous attempt is just another attribution which doesn’t hold a reality check. Users’ laziness (or then: the infamous “Google Instant”) may lead to a visit via a site’s home page. Crediting this visit (last click) for the purchase ignores any previous visit which may have started from a more elaborated search or a deeper landing page.

Having the revenue credited to this deeper landing page doesn’t take users’ intents into consideration. Depending on the phase of the consideration cycle the first visit may only have served the purpose of a primal market scan, and no matter how badly or perfectly the information may have been present on the site, have never resulted in a purchase (as the user simply wasn’t ready to buy in any case).

It is of course tempting (as we tried earlier) to make up “engagement scores” – pages deeper in the navigation (i.e. the “Technical Specifications” page for a product) would get a higher score for “prospect engagement” than a site’s home page. Downloading the specs as a PDF could be considered a micro-conversion and increase the score even further – but we still can’t take a look inside a consumer’s brain to attribute his/her intention correctly.

I am still not a big fan of the concept of “behaviorism” – as initially formulated by Thorndike or Watson, later refined and radicalized by Frederic Skinner – but next I will give Peterson’s and Carrabis’ paper on “Measuring the immeasurable: Visitor engagement” a deeper look.
Maybe they have collected some thoughts about this topic.