Re-Thinking The Attribution Model: Isn't It Time Attribution Was Based On Statistical Analysis?

Robin Davies is Country Manager UK at Mediaplex. Here he discusses why we should be re-thinking attribution and why the industry should be basing models more on atatistical analysis.
Digital marketing is supposed to have solved John Wanamaker’s question: “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” But the reality is that while digital media is strides ahead of “old” media in terms of accountability, there is still some way to travel before we can give brands a wholly accurate account of the multiple variables that influence people’s online purchase decision-making.
It is little wonder then that last click versus cross channel attribution, or “fractional attribution”, is one of the most hotly debated topics in digital marketing.
That the industry should focus on the appealing simplicity of last click attribution, which directly aligns spend with sales, was perhaps inevitable in order to achieve a standard around which it could unify and grow.
It’s also natural now that digital marketing is maturing, and accounting for a large proportion of client spend, that we should seek to move on from the blunt instrument of last click.
Online marketing targets consumers at all stages of the buying cycle and there is a growing agreement that we should aim to measure the influences from the first click to the last and all in between.
The problem is that the wide variety of attribution techniques on offer have not moved the game on far enough.
There are two major issues with fractional attribution. The first is that the significance attached to brand touch points on a consumer journey is all too often entirely arbitrary. A basic principle of data analysis is often forgotten—correlation is not the same as causation.
It’s noted that nations that add fluoride to their water have a higher cancer rate than those that don’t. A solid connection? No, it’s utterly misleading.
The simple explanation is those nations that add fluoride to their water are generally more developed and so more health-conscious. More of their citizens live long enough to develop cancer, which is, to a large extent, a disease of old age.
Just because a conversion event correlates to a click or impression does not mean it influenced or contributed to a sale and should not be attributed for doing so; it may be simply circumstantial.
The second problem with fractional attribution is that, whilst it generates an understanding of what the online marketing mix looks like, simply identifying footprints along user paths is too simplistic. It does not account for how much value should be weighted to different forms of online marketing.
For example, the different qualities and merits that should be attributed to display, “push” media, compared to search, which is much more “shop shelf” and represents consumers actively looking for something they want to find.
Typically, “flat attribution” models are used which present an overview of what drove a sale and the path a visitor took, and provide an equal share reward across digital sub-channels. However, this model does not provide any advice on which sub-channel performed best and therefore how marketing strategy should be changed and marketing spend reallocated: it simply echoes back what it was told.
So where do we go from here? The answer is a more rigorous approach, using statistical analysis to qualify causal inference.
Data must be crunched down as far as possible whilst still remaining statistically viable. Causality is inferred at the level of single variables such as, time of day, day of week, creative placement combinations.
This approach yields statistically significant indicators of incremental media value. An actual number can be generated that quantifies the contributing effects of elements in a media plan so that spend can be more confidently reallocated accordingly. Note that by considering media elements that didn’t correlate to sales you also take into account media investments that didn’t result in sales.
There will always be a degree of assumption that cannot be avoided; qualifying the effectiveness of a display impression will never be a black and white issue, so we will never answer Mr Wanamaker’s question with 100% accuracy.
As with anything in life, sometimes the best solution isn’t the easiest. Many of the current fractional attribution systems on offer do add value when accompanied by expert interpretation, but there’s a danger that digital clients will end up accepting canned solutions that do not actually do what they claim on the tin.
It’s high time the digital industry aimed for a more definitive solution to identifying and evaluating the influences on consumers’ digital journeys.