Analysis Paralysis

Asked to turn in a few hundred words on the phenomenon of analysis paralysis, I did what most right-thinking people would do and looked it up on Wikipedia. It struck me as strange that the entry there uses Rodin’s Le Penseur to illustrate the concept: a solid, unmovable manifestation of inertia, one literally cast in bronze. In this understanding, analysis paralysis is defined as an inability to make a decision due to the potential consequences of making the wrong one, with this pressure causing the individual to stop entirely and be unable to move forwards with their project: ultimately to sit down, elbow on knee and have a good long ponder.  

In our trade, analysis paralysis has the capacity to be markedly less passive. Replete with the considerable array of insights tools afforded us by the modern age, analysis becomes an active pursuit of meaning through doing rather than an exercise in aimless reflection. For this reason, there’s a great potential for this phenomenon to be a very vigorous way to waste time; a means to create inertia through the frenetic dissipation of energy and resources, in pursuit of a universal truth that can’t ever be grasped. Or perhaps one which even when grasped, is of only inconsequential value. Within most marketing functions each individual decision may lack the gravity of consequence required to reduce Rodin’s risk-averse thinker to his familiar pose, but the over-preponderance of decisions and the massive proliferation of data points on which to base them can certainly achieve the same effect.  

To demonstrate this point, I’ve used Google’s Ngram viewer to chart the rise of the phrase analysis paralysis through the literature of the last 222 years. The graph isn’t illustrative of anything at all really, but that lack of value – coupled to the fact that I’ve nonetheless bothered to look it up and then to include it here and waste your time with it – is wholly apposite. If you would all kindly spend a lot of time studying it in some detail, that would be even better. I did try to repeat the exercise with the Esperanto translation (analiza paralizo), but a lack of printed sources to use as a base proved an issue here and my petitions for Google to include constructed languages among their list of Ngram literary corpuses have so far been unsuccessful. 



Considered in this way, as the pursuit and accumulation of data that only obfuscates the issue at hand and doesn’t push a decision or (better still) an action forward, analysis paralysis is perhaps a situation in which a moment or two’s quiet reflection might be very beneficial: in effect, the opposite of Rodin’s underdressed contemplator. Put another way, it’s all heat and no light: a dearth of contemplation is very much part of the cause of contemporary analysis paralysis while, for Le Penseur, contemplation was itself an obstacle to getting things done.  

So if that’s the highfalutin thesis, what does it look like in practical terms? The opportunities to waste time in over-analysis or to grind to a halt amidst a morass of marketing data are as diverse as they are numerous, but they often have one common thread: a lack of context or proportionality informed by basic common sense. This proportionality may among other factors concern either the relative insignificance of a data point or the lack of statistical reliability of the value it indicates; whether that’s giving undue consideration to the bounce metric on a page which has almost no content and is designed simply to give directions to an outlet, or overestimating the importance of answers given by a group of respondents which is far smaller than √n. In either case, the risk for marketeers and agencies is not merely basing decisions on inconsequential data points or on data points of dubious certainty, it’s sinking resource into building slide after slide around these “facts” and fabricating from them conclusions and proposals that further nothing or, worse, simply admiring the facts and the graphs that propound them, drawing no conclusions, agreeing no actions and then moving on to marvel at the next fact. 

Attribution modelling presents a particularly rich seam for over-analysis and, counterintuitively, is also an area where under-analysis and an over-reliance on heuristics can present its own problems. Both situations arise from the same lack of proportionality and judgment. Without proper reflection or experience it’s tempting and easy to attribute all of the value of sales or conversions to the last touchpoint prior to that engagement. Given the complexity of human psychology and the innumerable actual and potential influences upon it, the idea that the final touchpoint before a purchase is solely responsible for generating the value therein is (as any agency type who has so much as left a coffee ring on a Kahneman book is desperate to tell you) a diamond-encrusted example of survivorship bias. This is perhaps more true for long-tail purchases like property, motors or finance but is also still the case for FMCG. I didn’t decide to buy this house rather than that one solely because a Facebook advert sent me to a website; likewise, while a perfectly positioned poster for a particular brand of cider might feasibly send me gasping into the nearest off licence, would it have been quite so impactful if I’d never heard of that brand before or hadn’t had that weird dream about apples again the night before?  

Whilst a moment’s reflection and analysis can reveal to us the fallacy of over-valuing the last touch, there’s also a risk of over-analysis here. In discarding the last-touch model as we ought, we may disappear too far up our own attribution funnel by trying to divine the “true” split of the value of each sale and attribute a fair share to each touchpoint real, hypothetical, and purely imagined. Typically, this is done within digital marketing by applying a differentiated weighting to the first, last and intervening visits to a website and then splitting the value of the basket in pro to this weighting. While this model is likely in many cases to be significantly more accurate than a last-touch model, it is also a largely arbitrary split and one which inevitably overemphasises the engagements that can be measured and disregards entirely those that cannot. This is an availability bias: we’re basing judgments on those events that can make their way into our data set, and we’re disregarding any that cannot. Such limitations are fine if they are well understood and their impact on the veracity of any conclusions is factored in, however analysis paralysis assumes that arriving at a “correct” answer is possible, invites refinement of the attribution model, and requires us to fret endlessly over the proper ratios of our attribution model: to add bay-windows and parapets to a castle we built in the sky, measure the windows and then invest heavily in curtains of exactly the right size.  

While this may read as a manifesto for gut feel or a paean to the lost art of intuition it truly isn’t. Data absolutely has its place, but the insight derived from it must not be some random emergent property of the numbers shoved in; it should instead be subject to the rigorous scrutiny of common sense and take its place among a wider narrative that makes sense within the real world. Lacking the clarity of contextualised understanding, we risk paralysis through over-analysis, chasing a true understanding of issues that cannot be truly understood, which could be approximated to within an acceptable margin of error with an educated guess or which simply don’t matter. We should draw generalities from specific data points where the data is robust, yes – but we should also generalise from lived experience and from logic.  

In many respects, knowing what to ignore is at the heart of making good commercial decisions. Given the overabundance of data available today, this has never been more true. The judgment concerning what to ignore and the confidence to do so comes from experience within a discipline and from contextual understanding of a sector, but it also comes from a business already having a direction of travel: from having a properly articulated agenda, growth plan, roadmap or strategy into which new insights can be assimilated or which can be modified or corrected by new learnings as they arise. Ultimately, it’s the thinker just standing up and getting on with it, moving forwards, reassured by analysis that progress in any direction is better than none.