How does Analytics fit in the Industrial Internet Reference Architecture?

The Industrial Internet Reference Architecture - published by the Industrial Internet Consortium (IIC) - was released recently.  Given my focus on analytics, that's really the only topic that I was interested in (my ARC colleagues cover the other areas).  But then again, I would argue that unless you're going to apply robust analytics to Industrial Internet data there really is no point spending time and money to collect it.

So how did my pet topic hold up in this first version of the reference architecture?  At first take, the word analytics occurs an impressive 67 times in a document that is essentially 89 pages long.  But, what did those 67 words actually say?  Surprising little to be honest.  There is a section on "Analytics and Advanced Data Processing", and a sole page dedicated to "Advanced Analytics" (more on that later).  There are a few high level themes throughout which ring true, that we're already starting to see play out in industry:

• There's the notion that analytics will be pervasive.  Way back in time, business intelligence and operational intelligence was something of an afterthought.  Although data was used in the here and now to run operations, deeper analysis typically occurred much later, long after the operations had happened.  Data generated by the Industrial Internet of Things (IIoT) will enable (or even demand) analytics to be embedded throughout operations.  Getting analytics deployed so widely will be a technology challenge for many industrial companies.  Above and beyond that though, it will present a monster cultural challenge for many too, as it implies that decision making becomes more democratized and is pushed down to more junior employees.


• Analytics will be pushed to the edge (decentralized).  We saw examples of this at the ARC forum back in February, with companies optimizing their use of assets or improving productivity by pushing analytics closer to the data source.  As the reference architecture also notes, this is going to be important if companies are going to avoid overwhelming their network bandwidth.  Analytics at the edge will require different technologies, such as complex event processing (aka event streaming) as I noted here, or predictive analytics as noted here.


• There's a tangential reference to learning - "analytics can detect both imminent failure of a component, but also circumstances extant across a fleet of components when a component failed. This enables global learning...Mechanisms must be in place to communicate knowledge up the chain of command based on actual incidents".  That style of learning is always a good thing of course.  In reality though, many analytics systems currently in place have not been strong in this respect, so that's an area of potential development.

Beyond those high level concepts, there is definitely room for improvement.   I may be reading the industry wrong, but I feel many people won't find the 5 page  "Analytics and Advanced Data Processing" section very useful (page 84 for those following along at home).  Advanced Data Processing - as far as I can tell - is (near) real-time operational intelligence.  Two technologies in particular are referenced - complex event processing (CEP) and "advanced analytics".  CEP we are starting to see (case study referenced above) and the Advanced Analytics section  does use the terms popularized by Tom Davenport - Descriptive, Predictive and Prescriptive.  (ARC has extended that framework a little to include newer developments).

However, things go a bit astray at this point in the reference architecture:

• The reference architecture document states "These advanced analytics fall into four major categories".  However, only three categories (not four) are listed - descriptive, predictive and prescriptive.  A simple proof-reading error perhaps - it is on page 87 after all...


• More pointedly, descriptive analytics are simple reports and dashboards - the style of business and operational intelligence that has existed for more than 30 years.  In my view,  descriptive approaches cannot be described as advanced analytics because - well, because there's nothing advanced about them.  To be fair, advanced analytics is a term that's crept into use over the last few years.  It's never been well defined, but the clue is in the word "advanced".  To selectively pull from the Merriam-Webster dictionary, advanced means "being beyond the elementary or introductory", or "greatly developed beyond an initial stage". In other words, "advanced" means something that is beyond reports and dashboards!  Descriptive isn't advance analytics and never will be in my opinion.

Beyond that, a number of desirable characteristics of analytics are listed at this point, including:  Automated, real-time, streaming, active, causal-oriented and distributed.  That's all fine.

I understand this is a reference architecture - and a first version at that.    But for me, this is too high level, too abstract and - dare I suggest - a bit too muddled to help people define and implement a strategy for IIoT analytics.  There's just nothing actionable here for analytics.