Jon Ingham and I agree on a lot of things but we’re not aligned on analytics. I believe that the heart of this is due to a common misconception about what is analytics, and how the modern analytical approach works.
At the heart of all we do is probability. Our work, like most other analysts is about trying to stack the odds in our, and our clients’ favour. We know the future is uncertain but, as Sherlock Holmes used to say
“when you have eliminated the impossible, whatever remains, however improbable, must be the truth. ”
In his recent post on the CIPD’s new framework Jon comments that:
Business and Commercial Insight, and Analytics. Oh dear, oh dear. Firstly, yes, we need business and commercial insight, but again, that’s not going to make the difference. Much more important than this is insight into people and organisational culture. That’s going to help change our organisation’s much more than developing the same skills our business colleagues already have! Similarly, with analytics – yes, we need to use whatever tools and approaches we can to further our insight about our people, but it’s what we do with the analytics that will count. Most of that is still going to come from good, calibrated intuition, and enhanced imagination!
We’ll address the last point first because I believe that it is the very core of what good analytics is.
On the importance of intuition
In his book on the great french mathematician Pierre-Simon Laplace, Charles C. Gillispie notes that “Laplace took probability as an instrument for repairing defects in knowledge.” The key part here is that we must start with our own intuition and knowledge to be able to repair it, or as most analysts would probably say, to update our understanding as additional information is acquired.
The starting point for the majority of analytic projects is the intuitions of the people most likely to have first-hand knowledge of the situation we’re trying to understand. The analyst seeks to gather anecdotal evidence and beliefs which they’ll use to guide their analysis.
The first use of this knowledge is in what is often called feature engineering, that is turning the raw data into something meaningful for analysis. As the BigML blog notes
“70% of the project’s time goes into feature engineering, 20% goes towards figuring out what comprises a proper and comprehensive evaluation of the algorithm, and only 10% goes into algorithm selection and tuning.”
From this, arguably 90% of the time of any machine learning project, or even any analytic project, goes into activity which is defined by domain knowledge. OrganizationView have strong knowledge of the analytical techniques but our key differentiator is having these techniques and the HR domain knowledge. Good analytics doesn’t ignore those intuitions, it embraces them.
Edwin Chen wrote one of my favourite posts that go some way into explaining the approach that a good analyst follows. He describes the way that the winners of the Netflix prize frame the project, and by it the way the data has been created. Extending this to the HR field it is clear to me that having a set of models that can be used to explain employees and their behaviours (which is how the data is created) is essential to be able to perform insightful analysis.
The second way that we use of intuition or knowledge in analytic projects is by using data to update our understanding. If you want a good, complete view of what we’re doing start here. Most folks of our age who did stats at university didn’t get taught this way. It just turns out that this method is particularly useful for analysing HR data because we tend not to have huge data sets, but have a relatively small number of observed (not experiment-generated) instances. In technical terms we call our intuition ‘a prior’. It’s what we update and we need to explicitly describe it.
Why analytics encourages imagination
HR folks are, as a whole, notably conservative and risk averse. Many commentators describe an approach defined by ‘telling you what you cannot do rather than what you can’. Analytics gives you the ability, even (especially?) if you are risk-averse, of being confident to be led by your imagination.
With a modern analysis we can start with a large number of possible explanations and find the one most likely to be supported by the data. We’re not looking at single points but at ranges of probabilities. What this gives us the ability to do is to consider multiple ideas, some of which may be deemed unconventional and to understand them in terms of how they are updated with the available data.
What I’m describing is a framework, using analytics, to explore our imagination and gain confidence through how our ideas are supported by the data. Analytics doesn’t quash our imagination – in contrast it gives us the platform to explore new, fresh ideas and understand how credible they are.
Why HR might still be scared
HR is often portrayed as scared of analytics. The typical story is that it requires a new set of skills which aren’t common in HR folk. I think there is some truth in this but I don’t think it’s why HR should be most scared.
The HR profession has adopted a set of ‘best-practice’ approaches and methods which, if one explores with the data, aren’t particularly resilient. An example would be competencies where there seems weak evidence of any relationship between competencies and objective performance.
Data forces us to be humble, by being willing to update our views as new data is available. Nate Silver calls this the prediction paradox:
“The more humility we have about our ability to make predictions, the more successful we can be in planning for the future.”
As data and analytics causes several of our core beliefs to unravel how many of us will be humble enough to say “we got it wrong”?
What excites me as the analytical HR specialist is that, by linking what we do to a data-driven understanding of its relationship to business performance I believe we’re making the biggest difference HR has ever done.