Terminology: Operational analytics
It’s time for me to try to define “operational analytics”. Clues pointing me to that need include:
- The term investigative analytics has gotten considerable traction.
- I generally contrast “investigative” and “operational” analytics, for example in the last line of the post linked above, or in my recent introduction to Odiago WibiData.
- It’s clear that I’m conflating several different things in the term. (See for example the operational analytics sections of my posts on eight kinds of analytic database or definitional challenges for 2011.)
- I’m pretty negative about the utility of alternate terms such as “operational BI”.
But as in all definitional discussions, please remember that nothing concise is ever precise.
Activities I want to call “operational analytics” include but are not limited to (and some of these overlap):
- A lot of things that happen at the time of customer interaction:
- Ad serving, web page personalization, and so on.
- On-the-fly fraud assessment.
- In general, much of what might be called “Next Best Action”.
- Most of automated price re-setting.
- Most of automated risk analysis.
- Algorithmic and/or high-frequency trading.
- Human beings taking specific actions as part of reasonably well-defined business processes, in response to exception conditions that they notice via something resembling a business intelligence tool.
However, I think it is usually unhelpful to stretch the term “business process” to include technical control kinds of activities (whether human or wholly automated) such as network security, network operations, power plant operations, etc. So I’m inclined to leave those areas out of “operational analytics” as well, although I’m not sure whether that distinction will hold up — and even it does, there surely will be border cases that are hard to exclude.
In simplest terms, what I mean by operational analytics is analytics being done on the fly as part of operational business processes. By way of contrast, investigative analytics is done at the speed of research, not the speed of operational business processes. But naturally there are border cases in this version of the dichotomy too, such as when the analytics are highly urgent yet otherwise investigative in nature.
To get to a somewhat more rigorous version, let’s start by recalling my definition of investigative analytics as
seeking (previously unknown) patterns in data.
The key concept there is that the patterns aren’t known until you investigate and find them.
By way of contrast, most of what passes for cognition and programmed behavior alike is the recognition of and response to known patterns. Indeed, when we consider what happens in microelectronics and neurons alike, it’s “pattern response all the way down.”* Machines and humans alike monitor situations, detect exceptions, and make decisions based on known patterns. So I’m going to say that the essence of operational analytics is
analytic pattern response included in operational business processes.
*For some years I’ve believed “pattern response” is a better way of putting the concept than just “pattern recognition”; the latter phrase says both too much and too little at once. Otherwise, I was referencing the meme “turtles all the way down”.
For now, I’ll take the Justice Stewart approach to defining “operational business processes” — we know them when we see them, and no further elucidation is needed. Even so, it might be helpful to observe that the processes can be:
- Wholly automated — examples included automated trading or web page ad serving.
- Primarily automated — for example, software might tell a call center employee what offer to make next.
- At considerable human discretion — for example, a human being automatically notified of an exception could have freedom to decide what to do next.
The technologies for identifying a pattern match are also varied. You can write an ordinary database application. You can use a “rules engine”. Something like PMML (Predictive Modeling Markup Language) can be in the mix. Or, especially in human-discretion cases, various forms of business intelligence tool (broadly defined) can be involved, either standalone or integrated with more transactional applications.
So what do you think? Is the definition of “operational analytics” sufficiently clear? Is it accurate? Is it useful?
Comments
6 Responses to “Terminology: Operational analytics”
Leave a Reply
Morning 🙂
I liked the read, but it raised an eyebrow on this side of the screen.. I think the word(s) you’re looking for instead of “investigative analytics” is data mining? therehttp://en.wikipedia.org/wiki/Data_mining
On the “operational analytics” note:
Personally, I’m used to using simpler words like reporting and queries but thats mostly cos the clients find it easier to understand.
Hi Ivana!
If you click through to my definition of “Investigative Analytics”, it comprises more than data mining.
I think your definition is good, but it would be more clear-cut to define operational analytics in terms of its input. Specifically, I think operational analytics must depend upon real-time data from current “operations”. Investigative analytics normally only deals with learning from static historical data. The two can intersect when you want to answer a question like “What normally happens when conditions are like they are now?”
Brian,
While operational analytics depends on reasonably fresh data, “real-time” feels like an exaggeration.
For example, it’s operational analytics for L. L. Bean to personalize their offers to me, but if my cookie-suppression plug-in is working as intended, they may not have a lot of data to go on that’s fresher than Christmas season of last year.
(Yeah, I know, that’s overstated — there’s Twitter, my general credit card bills, and so on. But I hope it establishes the general point. Besides, I’m not sure that L. L. Bean really captures any of that ancillary data …)
[…] focused on investigative analytics, although some of its existing users seem to be more focused on operational analytics. Most specifically, SAND is trying to focus on “people data” — customer loyalty, […]
[…] definition of operational analytics Categories: Actian and Ingres, Aerospike, Basho and Riak, Cassandra, Clustrix, Couchbase, […]