Business strategy and technological innovation are on convergent paths; some would say they’re more or less inseparable already. CIOs and CTOs find themselves focusing less on the operational and more on the strategic: how can the technology that pervades their business be used to drive insights, reduce risk and create disruptive opportunities?
This shift has mirrored my own career. I’ve always been focused on physical manifestations of technology. Eighteen years ago you would have found me out in the field (often in an actual field) working on sensors for monitoring weather and the like. Pretty soon my focus shifted to making the data from those sensors more accessible – essentially putting it on the cloud. This was long before the term “Internet of Things” gained currency. Today, with the better analytic tools we have at our fingertips, it’s all about finding the actionable insights hidden in that sea of ones and zeros.
Those insights have always been there, of course. But, with a greater number of off-the-shelf algorithms and more affordable compute power, we can have access to them in something close to real time – not in a dense management report that takes several weeks to compile. We’re seeing data come to life at the hands of these analytic tools. It’s pretty exciting.
It can be pretty daunting too. Data volumes are burgeoning, so identifying which datasets to analyse – whether they're from your own operations, publically available resources or some combination of the two – can be an interesting exercise in inspiration and lateral thinking. Selecting the right analytical tools is no picnic either. And, just to add to the pressure, your competitors – not to mention an increasingly powerful cohort of cybercriminals – are seeking out their own strategic edge from exactly the same data (or at least the bits they can succeed in getting their hands on).
As an industry, though, I think we’re guilty of contributing to this sense of complexity. And part of that is down to some of the terms we bandy around so freely.
Data is always big
The minute you start talking about analytics I can guarantee a fair proportion of your audience has already assumed you’re talking about correlation between extensive, sprawling datasets. It’s applications like targeted advertising and cancer diagnosis, after all, that always hit the headlines. These are the poster children of the big data revolution.
But data doesn’t have to be so big. I worked on a project recently that analysed the behaviour of domestic fridges and freezers. We were sampling power usage every ten minutes: not huge volumes of data, but ultimately more than enough to tell people if they need to get something fixed or that it’s time to buy a new fridge. Getting a text to say your freezer has broken? That’s potentially hugely valuable. And all from very modest data.
Analytics isn’t always cognitive
There’s a tendency right now – of lumping any and every kind of statistical algorithm under the “cognitive” banner. (And of using the word “cognitive” as a noun. Don’t even start me on that!) The reality, though, is such techniques exist on a spectrum, only one end of which justifies the term “cognitive computing” at all.
At one end is the analysis of simple statistics: maximum, minimum and average readings, for instance. Having spent time worrying about which cognitive model would best suit my needs for the fridge monitoring project, I realised that all I needed to do was to compare average power consumption from one week to the next. I didn’t need AI; Excel could have done the trick.
In the middle of our range we have model-based analytics. Essentially building a “digital twin” of a real-world system and running tests and simulations on the software model rather than on the real thing. What would happen if we ran this machine for another five years? Or doubled the throughput on that one. A lot of our work in predictive maintenance is based on models like this. But it’s not really “cognitive” computing in any true sense of the word.
It’s only at the far end of the spectrum, when we get into AI (officially “artificial intelligence”, although at IBM we prefer the A to stand for “augmented”) and machine learning, that our models and tools really become cognitive. And it’s certainly powerful stuff. We’ve built a system for Kone, for instance, that can make hypotheses about the causes of anomalies it detects in lift performance, in order to better inform and equip engineers to deal with the problem – often before the problem has even manifested itself.
Systems like this - that can understand, reason, learn and interact - are undeniably powerful. Even more so when, like Watson, they can show how they arrived at their recommendations.
But CIOs shouldn’t fall into the trap of assuming that the only way they can deliver useful insights is through implementation of a complex, cognitive system.
And they shouldn’t worry how big their data is either.