The possibilities for prescriptive analytics in HR

Is prescriptive analytics the holy grail for HR data, or is it simply something the function has always done?

It’s a familiar scenario. You’re about to switch the telly off and turn your attention to something more constructive or edifying (reading a book, doing the dishes, conversing with your child…). But what’s this? A ‘for you’ recommendation on Netflix.

It’s not your usual thing, but it does have a strong female lead and brooding Scandi setting. Before you know it, you’ve settled back in on the sofa.

Such is the lure of streaming platforms’ fiendishly clever algorithms, which now learn your viewing preferences from highly nuanced factors, such as where and when you watch, what you watched last year compared to this year, and whether you enjoy ‘Visually striking nostalgic dramas’ or ‘Understated romantic road trip movies’ (genuine Netflix categories).

It comes as little surprise to hear, then, that we now spend twice as much time binge-watching Netflix as we do bonding with our families (71 versus 34-37 minutes, according to Streaming Observer analysis). But most will be unaware of the technical term at the heart of these highly addictive recommendations: ‘prescriptive analytics’.

And yet it’s a concept that powers a surprising amount of modern life. Shopping on Amazon, for example. Or navigating traffic, where your phone can combine datasets including roadworks and other travellers’ whereabouts to recommend the best route.

It’s something also already used extensively in some areas of business. And it is now making its way slowly but surely into HR.

Data in HR:

Putting people on the analytics map, part one

How to lead with data in times of crisis

HR not collecting enough data to understand employees

The holy grail

But what exactly does prescriptive analytics mean? The received wisdom goes that prescriptive analytics is the next stage on the journey after descriptive and predictive.

“In the chain of value, descriptive is just getting access to the data in a manner you can deal with,” explains Ankur Modi, CEO of StatusToday. “The next layer is around diagnostics, which is understanding key trends and why something might be happening. The third is predictive analytics, which looks at ‘now that I know that trend, can I predict what’s going to happen?’

“The final stage, and arguably the holy grail, is prescriptive analytics. In simple terms it’s the action plan based on the data.”

When it comes to HR, Modi concedes that while many in the profession will have heard the term more of late, few have gotten to grips with it. “In general, HR analytics awareness is low. And prescriptive analytics is hard for technical people, let alone for HR people,” he says.

“This is in its infancy in terms of the ways it’s being used by HR at the moment,” agrees Olly Britnell, global head of workforce analytics and HR strategy at Experian. “But it could get very sophisticated very quickly.”

In terms of potential applications, both Britnell and Modi cite the same key area optimised by its predecessor, predictive analytics: retention. Extending analytics capability from predictive to prescriptive here means HR professionals can draw significant competitive advantage not just from foretelling which employees may soon resign and when, but also how to prevent this.

“One of the predictive models we’ve developed in HR at Experian which will lend itself well to the prescriptive end of things is around attrition,” says Britnell. “We have a predictive model that flags who’s high risk and the factors driving this – such as that they’ve had two supervisor changes in the last month.

“We’re using machine learning to track interventions such as changing the team structure or offering more training, and then tracking which ones are having an impact. The idea is that in a year’s time we’ll be able to give our HR business partners (HRBPs) insight into who’s high risk but also, based on what we’ve seen in the business, what to do about it.”

Another area Experian is applying this to is D&I. “Like most organisations, a challenge we face is around gender diversity, and we’ve done lots of analysis around ‘where do we have the challenge?’, then modelling around the levers at our disposal,” says Britnell.

“To me, that is a prescriptive piece of work because effectively you’re outlining what the future entails, providing the data around what’s needed to achieve a target, and giving the business the tools to execute around the right strategy.”

Paul Cutler, group HR director at Travelex, adds workforce planning to the mix. For a highly seasonal business like his, it’s critical that HR can prescribe the volume of staff needed in stores and airport stands according not just to historic customer demand and passenger flow data, but by also predicting the impact of new, emerging events.

“So it’s what impact does Brexit have on summer travel numbers in the UK and how does that lead to staffing recommendations?” he says. “Because if you were operating in a purely descriptive analytics world, you would be basing that on last year.”

Something and nothing

The objection forming for some, however, might be that this sounds very much like what HR has always done – or, perhaps, should always have been doing. After all, as an HR professional, if you’re not handling and analysing data in a way that leads to action, what – to put it bluntly – are you doing?

Britnell confirms this is a fair challenge. The danger is making prescriptive analytics sound overly new-fangled and complicated, he says.

“I’m not a massive fan of the term because it is just what we should be doing as a function,” he says, adding that, as such, while it might be the holy grail of analytics, this doesn’t mean it should be the last stage on an HR team’s analytics journey.

“Models on HR analytics can be a bit misleading. Because they suggest you have to do one thing first, then another, then another – they’re either a pyramid or a curve. But actually some of the prescriptive stuff is a bit simpler than predictive.

"So it’s being tactical about what outcome you want and what methods you need, as opposed to thinking you need predictive and prescriptive across the whole lifecycle and a dashboard and all the rest of it,” says Britnell.

He adds how unhelpfully varied definitions of prescriptive analytics can be. “There isn’t a recognised global view of what it is, whereas predictive is a bit cleaner and clearer; you’re predicting an outcome.”

Cutler agrees that deciding on a course of action is “surely what analytics is all about”. But he feels there is not necessarily any harm in the term if it helps HR professionals recalibrate their approach to data.

“I think the risk is people have just got swamped in data and have become completely overwhelmed with where to focus. Sometimes the result has been analysis paralysis,” he says.

“So for me prescriptive analytics is where analytics can finally begin to be useful. Because what differentiates it is it’s applied nature. So it’s not just analysis of trends and using that to predict. It’s actually going: let’s cut straight to the outcome.”

“I guess it’s how broad is your definition?” muses Craig Stanton, manager, Workday practice at PwC UK. “If your definition of prescription is ‘it’s completely brand new insight’, maybe it doesn’t qualify. But, if prescriptive is ‘I need help making a decision, what do I do?’, [then maybe it does].”

He adds: “For me, humans are pretty bad at making decisions even in the face of data that points to an outcome. So prescriptive analytics says: ‘Look, here is all the data that’s telling a story, and here’s what the system is telling you to do.’”

This is particularly helpful for smaller businesses, adds Modi. He agrees that prescriptive analytics should be demystified so that people realise they have effectively been doing the ‘low-tech’ version with their brains for years.

“When you go to an expert and tell them about a problem, they ask a few questions and come up with a recommendation. That act is a data-driven analytic. That’s why prescriptive analytics can be so powerful for small businesses. Large firms have the resources to hire experts and effectively do this manually.”

He adds: “The beauty of these recommendations is that, unlike in a traditional case where a consultant could come up with generic recommendations, these are specific to the data.”

True analytics?

But it’s on this latter point that the experts begin to seriously part ways – with many sceptical of technology’s ability at this stage to recommend action truly based on specific datasets. What the tech is actually doing in most cases, some assert, is offering generic best-practice advice.

It is crucial to be clear on whether it is the system itself telling you what to do or the professional extrapolating the best course of action based on the data, stresses Keith McNulty, global director of people analytics and measurement at McKinsey & Company.

The danger, he explains, is that some tech vendors use the label ‘prescriptive analytics’ to suggest their software does something it in fact does not.

“If you give it a formal term and someone’s offering it as a part of a product then you have to assume they have something in there where the software tells you what to do,” he says.

“But often they’ve just put a few simple rules in which say ‘if the result comes in like this, tell the user to do this’. There’s no actual number crunching involved. It raises the question of whether the word ‘analytics’ should be attached.”

HR's analytics journey at IBM

“I do believe that use of analytics will increase exponentially over the next few years. To me this is the ‘sweet spot’ that will help businesses make more informed decisions," says Stephen Kelly, VP and CHRO at IBM Global Business Services.

"At IBM we’ve dramatically transformed our own HR function with our AI platform, Watson. AI, or ‘augmented intelligence’ as we prefer to call it, is about providing deeper insight by looking at multiple data sources and factors that allow managers/leaders to make more informed decisions around critical actions, processes and workflows, to solve pervasive talent issues.

"In our case we have leveraged AI to help better understand our skills, prevent employee turnover, match employees and external candidates with career opportunities, support managers with better salary investment guidance, and create a platform for employees to learn.

"This transformation has driven more than $300 million in benefits to IBM and, just as importantly, created significantly better candidate, employee and manager experiences. Today, eight out of 10 IBM employees have skills of the future compared to just four out of 10 five years ago.

"We need a new breed of HR professional – one that is data-driven, business savvy, comfortable with ambiguity, and capable of thinking and communicating strategically and using insight to inform decision-making. In this way, data will elevate HR to a different level in many leaders’ minds.”

Sally Winston, head of EX solution strategy EMEA at Qualtrics, agrees that people shouldn’t be under any illusions about the software’s current capabilities. Qualtrics works by identifying problem areas within an organisation and then “connecting back with ideas for action”, she explains. “At the moment we’re not in a place where robots create the advice.”

In light of these limitations, prescriptive analytics should currently always augment rather than replace human judgement.

“We most certainly need human intervention to oversee the predictions and prescriptions produced by advanced analytics techniques,” agrees Padmashri Suresh, principal data scientist at O.C. Tanner.

“Human subject matter experts who understand both the problem area and the advanced analytics techniques are required to ensure the data that these algorithms are using, and the algorithms used, do not result in any inadvertent bias.”

Ian Cook, VP of people solutions at Visier, explains that whereas prescriptive analytics has worked unproblematically for years in recommending when to buy, hold and sell stocks, for example (so that “the expertise of an investment analyst is being reproduced by the analytics”), recommendations that relate to people are a whole different ball game.

“The circumstances when it comes to people, where there’s one clear answer, I’ve yet to find,” says Cook. “There’s a group advocating that there should always be a human in the loop with any kind of AI, so you cannot absolve yourself of accountability.”

As such, application of truly prescriptive analytics is perhaps best confined at present to very straightforward, operational areas of HR, he says. Which will typically consist of the employee themselves being “nudged” to do something.

“So with things like benefits claims or questions about your experience in the organisation, there are opportunities to say ‘you haven’t signed up for that yet’ or ‘you should probably do that next’,” explains Cook.

Netflix for L&D

Which brings us to what many feel could actually be the most compelling application of prescriptive analytics in HR: Netflix-style recommendations within L&D.

“Prescription in learning is where we’ve already seen some really cool applications,” says Cook. He explains that algorithms can create a tailored dashboard of L&D recommendations based on what an individual is working on, their career aspirations, and what training people like them have done.

The same principles could be applied to making connections across the workforce, says Modi. “You can empower employees to say ‘these are the people in my network who are maybe the super-influencers or have similar interests to me, I need to catch up with them’.

“Or it’s telling people how they should optimise their productivity, showing them the difference between their average day and the average day of someone who’s hyper-productive,” he adds, explaining that such capabilities could be incredibly powerful in future in democratising and decentralising HR. “There’s no reason things like performance have to be so top down,” says Modi.

The danger though, warns Cook, is that such recommendations will only ever be as good as the data behind them. “You could have a system that says ‘you’ve not talked to Jane in two weeks, you should catch up’.

That’s a great example of where it could be helpful, or extremely annoying – because I sit next to Jane so I interact with her in person all the time.”

Quality issues aside, the quantity of data needed also poses a challenge. “The issue is that no organisation will have sufficient data to refine those recommendations to a point where a human would say ‘that is adding value’. Even thousands of people won’t be enough. Even with hundreds of millions of people – as Netflix has – you get some wonky recommendations,” says McNulty.

Which perhaps spells out the case for prescriptive analytics to be fuelled not just by data collected by one employer, but numerous external datasets.

“There are so many datasets becoming more available that you can now draw on,” says Ed Houghton, head of research and thought leadership at the CIPD.

“For example, in workforce planning: you’re a London-based business but you know from your data that many of your employees are based in Brighton, and that the train line will be disrupted. So you can use macro-economic datasets to plan around whether you plant someone in a certain location or put people up in hotels.”

“The art of the possible is huge,” agrees Britnell. “For example, we’ve seen vendors recently that look at publicly available data on Facebook and LinkedIn so you know as a recruiter the best time to pick up the phone.”

But the issue for McNulty is that “once the data has moved into centralised, aggregated portals, it becomes much lower quality”.

“For example, I get endorsements on LinkedIn from people who have never met me and never worked with me,” he says, adding that legislation such as the General Data Protection Regulation (GDPR) has put a serious dampener on the amount of personal data freely shared.

Data definitions: Analytics glossary

Descriptive analytics – The use of historical data to describe things today. Many organisations already do this for things like absence rates, skills level, and pay and remuneration – even if they refer to data generated by HR systems simply as ‘reporting’.

Predictive analytics – The use of statistical techniques to understand current or historical facts to then use them to make future predictions. Some regular business flows can be very predictable, such as absence and the time it takes to hire people. Others are less predictable, such as when someone will leave an organisation. It’s in getting to grips with these that HR can deliver significant strategic advantage, many feel.

Prescriptive analytics – Examining the outcomes of computerised modelling exercises for predictions using different variables – including data from outside the HR function – to recommend the best course of action.

The other barrier is HR skillsets, says Mark Judd, VP of HCM product strategy, EMEA at Workday and former head of HR operations for shared services at Rolls-Royce. “We know that even among those HR people familiar with this world, still only 1% actively use data engineering,” he says. “Most often they’re pulling it into an Excel spreadsheet then presenting it to their exec team.”

Which brings us back to how important it is for HR to get stuck into decision-making based on data – whether spurred on by a shiny new term or not.

“If HR is doing analytics full stop, that’s a positive move in the right direction,” says Cutler. “I don’t think the categories matter too much.

“What matters is: are HR people numerate? Are they fluent in the language of the business? Are they able to leverage tech to interpret multiple and diverse datasets, combine that with understanding of the business and their people, and then make strong, well-backed-up recommendations?

“If you’re doing that, you can call it whatever you like.”

This piece appeared in the October 2019 HR Technology Supplement