One of the most common requests from my clients is for Actionable Insight; a desire to receive simple, understandable output from analytical activity which provides a clear indication for what they should do. As a request it’s entirely understandable, reasonable and a great focus for the majority of analysis; after all, analysis is most valuable when it directs action and results in improvement. There are, however, two clear cases when the desire for Actionable Insight oversteps the mark and misses the point:
- At it’s worst, the Actionable Insight request transmutes into a Fix My Business Button – an unreasonable expectation for a magic box which absolves the user from having to think; and,
- There are times – limited, admittedly – when the most important insight is that which is not immediately actionable, let me explain…
Effective analysis needs to take into account the context in which it is performed – it is usually performed within constraints, of which the analyst should be explicitly aware, such as:
- Timeframe; most analysis needs to produce a conclusion within a defined period
- Operating conditions; commercial insight is only insightful if the organisation’s operational constraints are understood – market, operating practices, systems, processes, resources etc.
- Data; completeness, quality and availability of suitable, relevant data constrain most analysis
It is the second of these which can over-constrain analysis, by preventing the analyst thinking beyond current conditions; there are occasions where effective analysis should challenge the consensus view, and move beyond the status quo limitations.
This has been brought into sharp focus for me recently on an analytical project for a large UK retailer, where the organisation goal is a reduction in the cost of waste – a noble aim, and one shared by many UK grocers at present with food waste very visible in the news. In its pursuit of this goal, the retailer in question is looking to improve its understanding of shopper behaviour, review and optimise its store clustering approach, adapt its product range (to shopper behaviour) and generally get the right product in the right store. Our analysis is helping to identify how and where products perform, in order to inform these activities, but there is an elephant in the room.
The elephant’s name is Store-Specific Ranging – the ability for a retailer to tailor the products in a store to meet the needs of its shoppers. This presents a challenge for many retailers; their processes, computer systems and supply chain logistics are not geared for this level of customisation, and so stores are grouped together and treated as similar in behaviour. This similarity is reasonable, in aggregate, as in many ways shoppers do behave in predictable ways… but this is another example of the Average of Averages problem discussed in a previous piece.
Visual Analytics offers an extremely effective means of exploring the detail within such aggregated models – for example, plotting all stores on a map, and then sizing each plot by the sales of a selected item or brand, and colouring by variance from the national, regional or cluster average, will identify local and regional clusters instantly. What’s more, the resulting image is immediately and easily communicated to colleagues; there is no need to explain the workings or rationale for an algorithm, when the algorithm is already wired into our brains.
The insight yielded is powerful; predictably, cider is more popular in the South-West of the UK than the rest of the country, for example, but:
* National brands over-perform there as well as local /regional brands – this informs macro space planning within BWS (beers, wines and spirits) as ciders should be given more space
* Some regional brands do work right across the region, but many are far more localised…
… treating the South-West as a single region will mean that Cornwall ciders run out of stock in Cornwall, but generate waste in Somerset, and vice versa
* Across the rest of the country there are individual stores which massively over-perform on 2L bottles; closer inspection reveals these as University towns
How much of this insight is immediately actionable depends greatly on the retailer’s merchandising systems and supply chain processes. It’s possible that nothing can be actioned immediately – perhaps all of these observations require systems change before they can be acted upon.
Does this lack of actionability render the insight redundant? Absolutely not; how would the retailer know to consider systems and process change without such insight?
This is the crossover between the realm of the “known unknown” and the “unknown unknown” – in looking for actionable insight, exploring possible answers to a known problem, we have stumbled across a plethora of subtleties which offer a better understanding of shopper need, and identified a range of unrecognised opportunities which we *could* use to our advantage if our operations could cater for them.
Sometimes the most valuable insight is not actionable, but directional.
One thought on “In praise of *unactionable* insight”
Reblogged this on Data Animator.