What is Actionable Insight?

What is “Actionable Insight“? Let’s break that question down: what is insight and when is it actionable?

“Insight” is typically presented by business intelligence (BI) and analytics vendors as in the middle of the continuum of hindsight, insight and foresight.  From a temporal point of view this suggests that insight is neither about what happened in the past nor about what is going to happen in the future.  In other words, insight is concerned with what is happening right now.  There is some truth in this: for example, a microcontroller in a wind turbine monitors atmospheric conditions and adjusts the angle and tilt of the turbine to maximise electricity generation.  Or, your logistics system determines that there is a traffic jam between junctions 8 and 10 on the highway and it directs your driver to turn off at junction 7 and take an alternate route, which will be more efficient.  Or your call centre software suggests the “next best action” for you when a potentially churning customer calls in.

However, while this temporal idea is a neat suggestion from a marketing point of view, it fails the test of logic.  Is an archaeologist digging up historical remains providing hindsight; or is he or she providing insight into the past?  Clearly, the latter.  The hindsight-insight-foresight paradigm also suggests that foresight is not providing insight into the future which, of course, it is.

I would argue that insight is what tells you how you could or should react to whatever circumstances you have detected.  If we take the examples quoted, insight into current conditions is driving some particular action that is designed to optimise performance.  In practice, there are two ways that this insight may be derived.  Consider the wind turbine: the positioning of the turbine blade is based on an understanding of atmospheric physics, which has been determined by research.  In the logistics example, you need a knowledge of distances, average speed, congestion and so on.  In both cases, insight is embedded in the relevant software which tells the turbine or driver how to react. Both of these are essentially empirical.  On the other hand, call centre software is based on an understanding that, historically, particular groups of people will tend to act in a certain way given appropriate incentives.  So, sometimes insight is based on historical analysis of what happened and sometimes it is based on observation and experimentation.

Now consider another scenario. Suppose that your stores in the South-West under-performed compared to stores in other regions: why is that?  Were sales affected by the weather?  Or were there staff shortages?  Is it a question of poor management?  Perhaps there is particularly strong regional competition?  Possibly it is a demographic issue?  Hindsight can give you the base facts about under-performance.  What it can’t answer is why.  And I would argue that understanding why is, at least sometimes, an important aspect of insight.  In this particular case you want to know how to rectify the situation but you need to know why things have gone wrong first.  This will require further research.  You may have to do some further analytics on weather patterns, you may need to investigate management and governance (“oversight” in this context) and you may want to look at what the competition is doing (sometimes called “inside-outsight”).  In any case, you need to gain genuine insight into what has been going on.

So, insight is any information that has the potential to allow you to determine the best way to proceed.  If you are an archaeologist, insight gives you more information, to allow you to continue your research; if you are a business person, it lets you know how you might optimise a particular function, or that there is a threat or opportunity that you could sensibly react to (perhaps after further analysis or research).

However, there is one further caveat.  How reliable is your insight?  Is it based on trustworthy data?  How can you be sure that all – and not just some – relevant data has been included in the analysis from which your insight has been drawn?  And, similarly, is the underlying data up-to-date?  Even if the answer to all these questions is yes, insight is never entirely reliable.  It is possible, for example, to draw false conclusions; or to reason in favour of correlations that are merely coincidental.  In other words, there is a risk that any particular insight is actually a false one.  In addition, there is the question of the reliability of the data being analysed.  This has led to some experts suggesting that insight (or the data associated with it) be colour coded.  So, if it comes out of playing around with an Excel spreadsheet on data derived from an Access database then it is probably coloured red while, at the other extreme, if it is coming out of a traditional data warehouse that has high standards of data quality and where analysis goes through formal processes of verification, then this is probably gold standard.  Nevertheless, remember that this is only a measure of the data quality and still doesn’t necessarily imply accurate insights.  In any case, what we are really assessing here is risk: what is the risk that this insight is in fact erroneous and will lead us down the wrong path if we act upon it?

Okay, so much for insight.  When is it actionable?  Technically, “actionable” means the ability to action something.  This isn’t very useful.  To begin with it raises the question of whether insight is ever non-actionable?  I can certainly think of insights that are not actionable within a particular context: for example, some skunkworks project that has discovered something really interesting but which is totally irrelevant to the organisation.  Leaving that aside, if we make the assumption that any insights uncovered have at least some relevance to your company, then I think they must all be actionable.  At least in principle.  In practice, any action may be deeply impractical given the current culture within your organisation. Thus the fact that insight is actionable doesn’t necessarily mean that you will take any action, but, in theory, you could do if you wanted to.  Readers are invited to submit any examples of relevant insight – insight not mere facts – that could not, at least in theory, be actioned. If I am correct in thinking that all insight is potentially actionable then what is the point of qualifying insight as “actionable”?

Another question that needs to be addressed is that, if insight is actionable, does that mean that it is necessarily valuable?  Cassandra provided lots of actionable insight but nobody believed her, so her predictions (insights into the future) were not acted upon.  Clearly, her forecasts of doom had potential value but, in practice, they were of no value whatsoever.  Thus the fact that insight is actionable does not mean that it has value.

To consider this further, some people would argue that if you uncover some insight and you email the details of that insight to your boss, then that insight is “actionable”.  One thing that might happen next is that your boss might simply ignore your insight, which, following this line of argument, means that the insight is actionable from your perspective but is not actionable from the point of view of your boss.  Does this mean that the insight wasn’t actionable in the first place?  Or does it mean that the insight has no value?  Or that your boss an ignoramus?  Or, perhaps, that you are happy to live with the idea that the ability to take action can be paradoxical?

There are other possibilities: your boss might consider your insight in detail but determine that because of time and resource constraints there is nothing the company can do about it, at least for the time being.  In other words, your insight is valuable but not valuable enough.  And, thirdly, s/he might take some specific action to leverage your insight, in which case we can conclude that your insight probably has real and genuine value.

In other words, simply “doing something” with your insight does not make your insight any more actionable than it was in the first place and it does not make it any more valuable.

To sum up this discussion of what “actionable” means: firstly, all insight is actionable, so the concept has no useful meaning.  Secondly, even if it does have meaning it might have no value, it might have some theoretical value (but not enough) or it might have a lot of value.  It is even possible that this insight has a negative value: because it is, in fact, erroneous.  A better way to put it would be to say that insight may have potential.  Even assuming that your insight is deemed to be valid, whether that potential is realised in terms of providing corporate value depends on whether your insights are actioned, not on whether they are “actionable”.  There is therefore a distinction between (actionable) insight, which could be leveraged; and actioned insight, which is being leveraged.

However, the truth is that we are stuck with the term “actionable insight”.  How might we redefine “actionable” so that this concept is useful in a real-world sense?  The best option, I think, is to equate “actionable” with “to be actioned”.  This might be tomorrow, or next week, or even next year, but if the business is going to ignore (i.e., not action) insights that it has invested time and money in discovering, then it might as well go home.  Which brings up a further point.  How do you ensure that actionable insight actually gets actioned, especially where action may take some time to come to fruition?  The answer is that you need some sort of governance processes – this is probably worth a whole separate discussion –  that monitor the implementation of important business insights.

So, what does actioned insight look like in practice?  The logistics and wind turbine use cases previously described are both examples of actioned insight whereby insight has been embedded into software processes, removing the need for human intervention at run-time.  This is also often the case with applications based on trend information.  For example, you can derive patterns of behaviour from trend data and those patterns can be formalised, so that pattern recognition software can identify when those patterns recur or, conversely, when exceptions to those patterns arise.  Good examples would be anti-money laundering and fraud detection respectively.  Alternatively, you may be able to derive rules from trend information and these rules can be exposed through decision management software and embedded into operational processes.

Alternatively, actions based on insight may be almost entirely manual.  For example, a doctor examining a patient may need to decide to recommend surgery; or to order blood tests; or an MRI scan; or that the patient is a malingerer.

And then there are hybrid environments.  Suppose, for example, that you have implemented predictive maintenance and are monitoring a piece of equipment that has an expected value for whatever it is you are measuring and tolerance levels between which the device should be operating.  If, over time, operational readings for this particular metric begin to oscillate between those tolerance levels, and if the amplitude of that oscillation is gradually increasing then you can reasonably deduce that something is going to go seriously wrong sooner or later, and probably sooner.  At this point the software raises an alert suggesting that an engineer be sent to the site to repair or replace the component that is acting up.  The relevant supervisor will still have a role to play, because he or she will need to authorise the relevant work order.  Thus there is a business process around acting on this insight.

One further point that I should make is that insights are not static but mutable.  We are not dealing with eternal truths here.  For instance, trends – especially when they deal with human behaviour – change over time.  It is therefore not sufficient to just derive some rules from today’s trends and then take appropriate actions.  You need to go on monitoring these trends, recognise when they diverge from previous expectations, and generate new patterns or rules that reflect today’s trends rather than yesterday’s.  From a business perspective this is non-trivial.  Changing to meet changing market conditions requires what Bloor Research refers to as mutability (see, for example, the introduction to David Norfolk’s blog here).  The mutable enterprise has the ability to change and adapt as circumstances change.  But the point is that insight should never be treated as static: things change and, as they change, the insight derived from data will change too, and so must the actions taken on the basis of that insight.

Actionable insight – at least as we have defined it – provides value from big data and, indeed “small data”.  The analysis of any sort of data, in any sort of quantity, has no value unless it provides actionable insight.  As readers will have surmised, I am not overly thrilled with the term “actionable insight”.  This is what the marketing departments of various IT vendors have lumbered us with and whatever I might say isn’t going to change that.  Historical analysis (insight) into marketing speak suggests that we should not be surprised that we have meaningless terminology that doesn’t adequately describe what we are doing.  Nevertheless, the fundamental idea has merit and is important.  Insight is valuable if, and only if, it has business value.  That value can only be realised if insight is used effectively to inform business and operational decision making, which is what “taking action” is all about.