It’s January 2nd. You’re the talent manager and your boss has just asked you to estimate likely hiring needs for the year ahead. What was once a bit of a chore is now really easy. You just login to your ATS and press the ‘predict hiring needs’ report. Via a clever algorithm that you have helped programme, it then tells you how many people are likely to leave this year and crucially which departments/areas you are going to need staff for and this allows you to start planning way ahead instead of the normal rush to replace a leaver.
For those not sure what an algorithm is, well it’s a predictive model that will produce a set of results based on metrics you plug in and algorithms run the world. No seriously they do. Google’s success is down to 1 thing: a very, very good algorithm that converts a search query into results by analysing what it thinks you are trying to find out then showing you websites it thinks answers that query. Want to know how insurance companies work out what to charge you...…..an algorithm. Modelling traffic flows at any particular time of the day…..yup, an algorithm. They are everywhere and they are coming to an ATS near you soon, or at least they should do. Here’s how they work for modelling the weather. Clever PhD meteorologist builds a model to predict next week’s weather. The model says something like…’if the temperature last year was x then this year it will be x, + or - 4% points. Multiply that figure by yy if the wind is z mph and coming from zz direction, cloud cover is xxzz you add 2c and after plugging in every conceivable variable and giving that variable a weighting, they then run the model and it predicts what the temperature will be normally with astonishing accuracy. The holy grail of course is predicting stuff you can make money on, sports matches, shares, movements in raw material prices but from the HR community’s perspective how could an algorithm help?
Well imagine if you could accurately predict not just the number of people that will leave your company, you could even predict which departments/areas and, wait for it…..imagine if you could accurately predict who is likely to leave. Think about that for a moment, on your ATS or HR system the report you just run gives you this result:
Churn rate this year 14.3%
You then click on each department result and it then flags up the at risk people i.e. those the model is suggesting are most likely to leave.
Think that is complicated to do? Actually you might be surprised how easy it would be to build this: add a dash of historic data (this many left in 2016,2017 etc broken down by department, and even further by specific job titles) and already you’ll get a reasonably accurate figure. Then you add in the analysis you have done sampling why different types of people leave, average length of stay at previous jobs, and hey presto you have a basic algorithm that could theoretically then flag up specific people as ‘at risk of leaving’.
All of which is all well and good but how could this help you? Well if you know at the start of the year approximately how many are going to leave from each department and have a good handle on what roles they do you can start planning your hiring before they leave. Whilst it may be difficult to actually start contacting people on Linkedin until there is a specific vacancy at your company you can start to build talent pools in these areas before someone leaves so you are fully prepared with a hit list of suitable applicants when the inevitable vacancy appears. As in any area of life, the more prepared you are for worst case scenarios the less negative impact such a scenario can have when/if it does happen.
Speaking of algorithms the next way they will be used is actually predicting how likely it is that someone will be a success at your company. In a way you use an algorithm now, you just give it a different name. It’s called the hiring team. They evaluate the data presented…..resume, test results, interview, references and make a judgement call on that candidate. The difference between this approach and an algorithm is bias and predictive success. Bias is pretty obvious: recruiters (like anyone else) are inherently biased through no fault of their own, that’s just human nature but hirers don’t make a judgement on how successful the person will be, they just say yes or no. Imagine if the decision was largely taken out of a recruiter’s hands and was left to an algorithm to make the call. There would be no bias and it could then even give them a rating of likely success…...from ‘this one’s a slam dunk’ to ‘keep them on a short notice period’. Would you even need to interview the person? Just plug in the data into the model and the model decides on whether they should be hired. Think that’s too far fetched? I’m not so sure. AI is increasingly getting good at replicating what humans do. Robots can replace us on the production line so why not an algorithm make the call on whether to hire or not?
Why don’t more companies try to build such a predictive model? I imagine they don’t know where to start but if you are seriously trying to win the war for talent, and which company isn’t, then you need to start modelling which applicants turned out to be superstars, which ones were turkeys and which ones just did ok. When you look at who turned out to be a superstar gradually you’ll start to see patterns emerging…..”superstars scored very highly on these technical tests pre hire, exhibiting these personality traits, had a leaving salary above the industry average, worked for a direct competitor/market leader, marginally above average grades at Uni/college etc etc, all had strong numerical skills, regularly promoted, average length of job tenure was 4.3 years”...”Duffers had the exact opposite” etc etc.
When I look at bad hires we’ve made or duffers I have worked with, undoubtedly they had some shared traits: regularly changed job, mediocre academics, low salary for their level of experience…..plug that data into our hiring algorithm and it would have red flagged them immediately.
Algorithms are coming to HR. This is not an if but a when. Google is driven by algorithms. Amazon……..here are 5 other books you might be interested in or the search results they find when you search for anything, it’s all driven by an algorithm. Netflix…...films you might like to watch etc etc. Job boards, Linkedin search results, etc…….all driven by algorithms so start building yours now and see the difference it can make. So does this mean our hard working Talent / HR Manager will no longer be needed if AI makes all the calls? No, not at all but the superstar HR/resourcer should develop skills in AI, data analytics and yes, if possible learn how to build an algorithm. For the HR community to lead not follow the coming data driven changes in the recruiting and talent space, they will need to adapt / develop a new set of data driven, numbers based skills. Worth putting that variable into your hiring algorithm next time you need to hire in your HR team: applicant has xyz skill level with data driven metrics.
Coming next week, I’ll give you an actual model we use to predict likely success of our hires.
Following on from my article last week about how you can use data or more specifically an algorithm to predict how successful a person will be if you hired them, we’ve put together a very simple, very basic algorithm you can use. Simply allocate more points depending on the weighting you give to each variable so if ‘they came from a referral’ has proven a great predictor of success in the past you would allocate more points from that variable as it has a highter weighting than another variable.
Work on this scoring system.
Strong hire: score of 35+
Medium hire: score of 20 - 34
Anything under 20 points - not hired
You plug the data into the model and hey presto you get the decision made for you.
Salary relative to expected market rate 1 - 3 points
Length of time within each previous company: 1 - 3 points
Speed of progression within previous companies: 1 - 8 points
Reputation/market leadership of all current/previous employers: 1 - 5 points
University / education better than average: 1 - 4 points
Personality profiling testing results: 1 - 3 points
Technical testing results: 1 - 8 points
Source of applicant: 1 - 3 points (more points allocated if source has been proven to be successful in the past or if a referral from employee: 3 - 8 points (depending on quality of employee doing the referring/previous track record of referring people)
Has worked for a company like ours before: 1 - 3 points
Lives nearby: 1 - 4 points
Been part of exactly the same project/type of work previously: 1- 5 points
Known by someone in the team 1 - 5 points
Overall rating from hiring team: 1 - 8 points
Number of Linkedin positive references 1 - 5 (1 point for each reference)
Estimated cultural fit: 1 - 4 points
Now all of these may or may not be relevant to your organisation. You need to do a detailed study of the shared characteristics of your superstar employees vs the duffers / B players. Your Holy Grail should be to try to identify markers that point to likely success at your company or put simply: what did our best employees have in common when they applied? Not all of the above may be relevant and you may wish to give a higher weighting to certain variables, so for example we give a higher rating to employees referred to us as we know that on average they tend to be more successful than non referrals, but this may not be the case for you. You just need to adjust the weighting as you see fit.
By building a predictive model like this we largely take most of the unconscious bias out of the hiring process and focus on the variables we know define our best employees. Try doing the same for your company and you might be surprised not only how accurate a well built algorithm can be, but also how it will often produce a different result to what just an interview alone might yield.
Looking for an ATS - iKrut can help. Go to www.ikrut.com, sign up and start using it free of charge.