How Data Analytics is Revolutionising Baking and Hunting
Big data is all well and good, but the true analytics pioneers are applying rigorous modeling to the raw numbers. you'll be amazed by what they find, say Morice Mendoza.
Swiss-Thai asparagus quiche. It’s a prospect even the most determined herbivore might struggle to salivate over. And yet, according to a predictive analytical tool from IBM, it’s scientifically delicious.
The technology giant has developed a “recipe generation engine” that is attracting interest from world-leading chefs and food manufacturers. It works by breaking meals down into their component parts and running the numbers on what makes certain ingredients work well together. It’s also an instructive example of how predictive analytics is adding nuance and insight to the straightforward crunching of data which, thanks to a hike in computational power that makes Moore’s Law look like an exercise in fence-sitting, has flooded our world with reams of information.
The IBM “cognitive cooking” algorithm allows for the intangible elements that make a meal work, which it has quantified as the degree of “surprise” in each dish. It knows you need a counter-intuitive smidgen of lemongrass and curry powder to offset the reliable but uninspiring asparagus, or that Gruyère (not mozzarella and certainly not ricotta) is the only cheese that will work.
There are few areas of life that such principles cannot transform. Thanks to the Hunt Predictor app, for instance, anyone stalking deer, turkey or waterfowl can benefit from a rigorous analysis of local weather conditions, lunar cycles and terrain that tells them exactly when to leave home to maximise their chances of bagging their prize. Conversely, the South African government is exploring whether similar inputs, aligned to an index of past incidents, can help it allocate its resources more effectively to protect the country’s dwindling number of rhino from mercenary poachers it has so far been unable to outfox.
Predictive analytics – the branch of data mining that uses mathematical techniques to predict future events or behaviours – is neither as new nor as impenetrable as popular narrative would attest. In retail, it has been picked over for decades, ever since Tesco engaged analytics consultancy Dunhumby, which it now owns outright, to create Clubcard, a loyalty scheme the supermarket integrated with its own marketing, pricing and supply chain operations to supplant intuition as the key indicator of future customer behaviour.
Today, US department store Target can predict when a customer might be pregnant – whether she knows it or not – based on changes in her purchasing habits. Amazon has just patented an application it thinks will tell it definitively what you want to read based on who you are, not what you’ve read before.
Yet while business is no stranger to this brave new world, it has yet to truly impact the way we manage employees inside organisations. Here, it could be even more revolutionary, particularly in areas such as recruitment, but its burn has been slower, says Luk Smeyers, founder of the European HR predictive analytics consultancy iNostix. Around 10 years ago, he says, people data was used to hone processes such as learning, with organisations focusing on data integration.
Today, we are concentrating squarely on deriving insight and adding value. If we can make employee metrics as transformational as customer metrics, business will never be the same again.
Google, predictably, is at the forefront of deriving such insights. It knows the optimal number of interviews per successful hire and the best way to on-board a new starter (it says an ‘agenda’ for the first four days leads to a 15 percent increase in productivity over time). It uses performance data to influence organisational design and structure, and staffing arrangements within particular departments.
Bank of America, another convert, had an issue with employee churn at some call centres. Consultants Sociometric Solutions asked employees to wear badges with sensors that picked up and recorded their locations, movement, tone of voice, volume and method of communication. Their comfortingly low-tech conclusion? People needed to feel more social at work; encouraging them to take their breaks with friends is credited with reducing the turnover rate from 40 per cent to 12 per cent.
Any analyst will tell you that asking the right questions is the key determinant of successful prediction. Karen O’Leonard, vice president with responsibility for talent analytics and benchmarking at Bersin by Deloitte, begins each assignment by hypothesizing with her client why an issue is worth examining, and what the real reasons lying behind it might be. She gives the example of a pharmaceutical company she worked with which was experiencing alarming churn rates among its Chinese sales staff. Management held “some pretty firm views on what was causing the turnover”, assuming it to be around compensation.
The data told a different story. “It netted out that compensation wasn’t significantly related to turnover and that actually there were other variables that were significant predictors of turnover,” says O’Leonard.
Time spent in role turned out to have particular relevance. Three years of data was applied to the current sales force to produce a predicted score of who was most likely to quit. Armed with this insight, the company was able to turn things around, targeting interventions around the di# cult two-year point of employees’ tenure: “Their manager would hold career discussions.
Stretch programmes and job rotations were put in place. Even if an individual didn’t get a promotion at the two-year mark, maybe they could be moved to a different territory or a slightly different role… Within nine months of rolling out a number of different talent initiatives, turnover dropped to single digits.”
Large companies in high-churn industries are the low-hanging fruit of analytics. Michael Housman, chief analytics o# cer at the hotly tipped young US data analytics company Evolv, has conducted research into employee turnover in call centre and sales roles. “I was surprised when the analysis came in based on hundreds of thousands of employees,” he says. “We found that applicants being job hoppers or being unemployed for long periods of time had almost no bearing on how likely you were to stick around at the next job you were hired for.”
He worked with companies including AT&T, Kelly Services and Xerox to integrate such insight into hiring systems. “We were responsible for hundreds, if not thousands, of individuals getting back into work when they probably wouldn’t have otherwise been able to find a job.”
At RBS, Greig Aitken, group head of people strategy and a notable analytics evangelist, has looked at 370 anonymised business units to spot correlations between engagement, business results, leadership capability and customer service. The result revealed a chasm between different areas: performance in the top 10 per cent was double that of the bottom 10 per cent across many measures, with measurable fi nancial losses as a result. Units in the lower quartile were also 60 per cent more likely to have had more than one unit manager in charge over the 12-month period of the study.
In performance management terms, the leading and best business units tended to focus on developing potential for the future. The worst performers were oriented towards the past, overly results-focused and lacked a coaching culture. Understanding these issues has helped make change management more impactful, says Aitken: “We have undergone a lot of organisational change to support the business. Analytics has been invaluable in helping us understand the extent to which we’re taking staff with us.”
Still, predictive analytics is an intellectual leap for many business functions, HR included. The first issue is around capability – it’s not a good idea to push HR business partners, for example, into a data-crunching role. “They didn’t come into HR to be in the numbers business,” says Smeyers, who advises that change needs to be incremental.
“HR hasn’t had a strong history of analytical skills and really understanding data,” agrees Peter Cheese, CIPD CEO. “And most organisations seem to really struggle with having coherent data about people.” Addressing these issues, he suggests, represents an opportunity to redefine and extend HR’s value to the business, but it relies on the function being proactive and owning the data agenda. Nick Holley, co-director of the Centre for HR Excellence at Henley Business School, says: “The problem is that most people in HR are so far away from analytics, let alone predictive analytics, that in general I don’t see that many organisations that are truly doing it.
The danger is that they are focusing on data and analytics within the HR function, whereas what actually matters is the application by HR of predictive analytics that can help it create value for the organisation.”
In this way, suggests Professor John Boudreau of USC Marshall School of Business in Los Angeles, HR can act as a conductor, using analytics and outside expertise to orchestrate internal experts to achieve their goals. He collected data from 30 HR leaders in 11 organisations, and asked them their views on trends such as big data and gamification. The more open to future ideas the HR department was, he concluded, the more they reached out to others both inside and outside the business, thereby increasing their wider value and connections.
To do this, you need to know which data to draw on, how to organise it and ultimately how to derive real insight. This comes back to knowing the business issues you want to solve, says Holley: “People are analysing a load of data and looking for insight. But the issue is how you understand the challenges the business is facing, then identify the insights you need to address them. HR may have data that isn’t actually of any interest! to the business, around profiling levels and sickness absenteeism.
Whereas actually, the business is interested in profitability, productivity and building strategic capability.” HR, in his analogy, is like someone scrambling around on the floor at night: “You ask him what he’s doing and he says ‘I’m trying to find my car keys, I just dropped them.’ You ask, did you drop them over here? And he says ‘No, I dropped them over there but I can’t see over there.’ We are often looking in the wrong place for the wrong things.”
Some of the most important data we have is unstructured, not held in a bucket somewhere, says Andy Campbell, HCM strategy director at Oracle. “Somebody leaves, they have an exit interview, and their reason for leaving is for more money. However, if we look at the unstructured data such as performance reviews they might have been hacked o" at the lack of opportunity.” All this contextual data is effectively lost, says Campbell. “These are little nuggets which are potentially much better indicators as to why the person left and if we can get hold of this kind of data we might have been able to prevent them leaving.”
Even if you can provide insight, there is still the issue of communicating it in a compelling way to the CEO and other business leaders, says Holley: “They can see the rigour of what you are doing, but also how it can make a difference to what they’re being judged on in their roles.”
Issues of governance also need to be considered, says Smeyers, largely related to being sensitive around privacy issues. James Hayton, professor of HRM and entrepreneurship at Warwick Business School, warns of the knock-on effect for fairness too. “One of the interesting issues is if you’re making decisions that give people either promotions or pay rises or opportunities of some kind. If other people are not getting those things, you’ve got to be able to explain why they didn’t get them.” Pointing to the data may not be good enough, he says. “When you’ve got these complex algorithms it becomes very opaque and if you can’t explain it, of course it undermines trust.”
Get it right and the implications are profound. IBM, while a naturally analytics-friendly organisation, is now beginning to look at the deeper contexts among its employees. Through an internal social media site known as IBM Connections, its 75,000 employees can share thoughts and ideas through forums, blogs, wikis and status updates. Using data mining software, IBM creates anonymised ‘senti-maps’ showing what employees are talking about. They are used in predictive modelling, most immediately to spot trends in attrition, but could also help alert the business to potential problems or untapped markets. Interestingly, IBM’s research appears to show a correlation between high levels of social media usage and workplace performance, though the reasons for this are as yet unclear.
The possibility that we may be able to capture, document and dissect almost everything that goes on inside our organisations, eradicating the vagaries of chance or ill-fortune, takes us into uncharted territory. Such a quest for certainty is an intrinsic element of the human condition. It’s why the Greeks mythologised the Oracle, a cavebound priestess whose barely coherent ramblings were said to contain fundamental truths about the future, and why Nostradamus’s predictions have captured our imagination for centuries. It’s also why Ronald Reagan, despite direct access to some of the finest analysts available, employed an astrologer throughout his presidency and even moved the date of his second inauguration so the stars were better aligned.
But there is a danger in over-estimating what mathematical models can predict, says Cheese: “We live in a VUCA (volatile, uncertain, complex and ambiguous) world. Business isn’t – and probably never was – a scientific, deterministic discipline. It certainly isn’t now.
In the end, black swan events happen.” With economists beginning to issue grim predictions of a possible return to global recession, this is a pertinent point. As Nassim Nicholas Taleb, author of The Black Swan, puts it: “Our predictors might be good at predicting the ordinary, but not the irregular, and this is where they ultimately fail.” Extrimistan – Taleb’s word for the world of black swans – is “where we are subjected to the tyranny of the singular, the accidental, the unseen and the unpredicted.” In this light, the best mathematical models cannot predict anything with absolute certainty. Pre-2008 employee hiring strategies, for example, would have been rendered redundant as the financial crisis unravelled.
Cheese stresses that business is about human beings and intangible elements that cannot be entirely deterministic. He quotes Einstein: “Not everything that counts can be counted and not everything that can be counted counts.” There are too many variables and possible causes of people-related events for absolute causal links to be found. “I really struggle with anybody who suggests there is a causal link that you can absolutely show between engagement, say, and profitability.
There’s a whole industry built up around engagement measurement where you will find organisations trying to assert that if you increase your engagement score by 5 per cent you’ll increase your total return to shareholders by 0.5 per cent. That doesn’t stand up to the rigour of analysis because the reality is there are so many intervening variables.”
Holley agrees. Too much predictive analytics hasn’t got a strong enough methodology behind it, he says. “A lot of it tries to equate correlation with causality. It doesn’t necessarily hold up.” In the 19th-century United States, he says, there was a perfect statistical correlation between increases in arrests for public drunkenness and the increasing number of travelling Baptist preachers. “Did one cause the other? No, it’s probably just pure coincidence. So am I going to start arresting Baptist preachers? If you stand back and use the common sense test it’s clearly not true.”
Bias and false assumption are also dangers when it comes to analysing data. For starters, there is the known problem of over-confidence bias, says Morten Kamp Andersen, a partner at the Danish management consultancy Proacteur, who used to work in financial predictive analytics before studying psychology. In short, it suggests you will exaggerate the accuracy of your theory. “According to the theory, and what I have experienced many times in reality, when we work with data over a stretch of time... we are slowly developing a story in our minds, a working hypothesis if you will. And slowly we are finding data which supports our hypothesis. This is a natural inclination.”
To deal with such biases, Holley says HR must ensure it knows enough about statistics, regression analysis and other mathematical phenomena to smell a rat. “They should know enough to be able to listen to the data analysts and know what true causality is. The success of predictive analytics depends not on correlation but on causality: this and this caused this and not this and this happened together.”
The next frontier may well be prescriptive analytics – knowing enough about individuals or groups of individuals to prompt action such as recommending training or offering a menu of customised benefits. As Boudreau says: “I can see a future where HR will routinely come to employees and based on data analysis say, ‘we think you’re leaving, you don’t know it but we do and our computer told us offering you this package will keep you’.”
Getting to that point will require businesses to truly interrogate the data in a way they have so far abjectly neglected to do. It will also be trickier than the quest for the ideal risotto. But it ought to be even more rewarding.