The four fundamental pillars of Predictive Recruitment
By taking interest in predictive recruitment, you are part of HR-related precursor analytics. This innovative subject tends to dominate in focus groups with HR actors. Like any innovative trend, predictive recruitment will by monopolised by a multitude of HR service providers who are disappointed not to have burst the “HR 3.0” and other “Generation Y” management advice.
At AssessFirst, we want to put forward expertise that involves a precise methodology. Also, any recruitment officer can claim to practice predictive recruitment.
Predictive recruitment must follow four principles that determine the quality of results obtained in terms of recruitment, as well as part of its application.
#1 – What do you want to predict? (You have the right to a SMART response)
Before answering, predictive recruitment raises questions that every company should ask when recruiting.
One of our favourite questions is “What do you want to predict?” because behind it obvious appearance, it raises the real issues of an organisation. The first answer that comes to mind is “the future success of the candidates”.
Ok on principle, but what is success, in your company, in that position?
What you should keep in mind to answer this question is that you can choose to act on two types of factors.
Factors related to performance:
- Training success
- Number of new clients
- Delivery delay
- Quality level
- Number of bugs fixed
- Level on a “nine box grid”
- Or simply the percentage of goal achievement. Almost every position has key indicators that monitoring performance.
Factors related to risk:
Selecting a priority criteria to maximise during your selection process is one of the key conditions for success in predictive recruitment. It is, indeed, rare to be able to act to a full extent across a set of variables.
It is the evolution of this criteria that will be followed over time to measure the effects of predictive recruitment in terms of results.
#2 – Collect and analyse the data
Predictive recruitment has value in the data on which it is based. Therefore, the collection of information is crucial in assuring a quality approach.
A pitfall to avoid: relying on declarative information. Some people associate predictive recruitment with the capacity to anticipate skills required for a position. They will, for example, collect the description of successful employees and apply them in their selection. This practice is not predictive, it simply resembles classic recruitment. The value of predictive recruitment is precisely in overcoming the pitfalls of subjective information.
In the end, to carry out sound statistical studies, two questions arise; that of information quality and that of information quantity.
To ensure the quality of information collected, you have to stick to three identified levels.
1) You must be able to measure what you want to predict, and measure it in a timely manner. Information taken at one point may be the result of a random event, but it occurs over a significant period, it is no longer by accident.
2) We then find information that can explain that these outcomes differ from one individual to another. These individual characteristics are principally personality, motivation, problem-solving, education, experience, commuting distance etc.
3) You must also identify all the contextual variables that could influence the results obtained by one person, for example, seniority, location, industry, manger etc. These variables ensure the performance share related to the work conditions and the proportion related to individual characteristics mentioned above.
From a quantitative point of view, there is no perfect threshold for implementing predictive recruitment. What is important is the available information. It is better to have information about 80 people in a population of 100, rather than the information of 300 people in a population of 5000.
Lastly, do not trust a service provider under the pretext of the use of the word “correlation”. Here is a significant correlation (the divorce rate and the consumption of margarine in Maine).
Should we draw the causal link as such?
Data analysis highlights the characteristics and variables that impact the criteria we want to predict and those that have no effect. Thus, some companies have found that education and experience has no impact on job success and we should not take them into account in the selection criteria.
#3 – Establish a predictive model
When speaking of predictive models, we are talking about the formula that will lead to a recommendation in terms of recruitment.
This model consists of the previously identified variables. Each candidate will be compared and at the end, estimate their probability to perform within the target criteria.
Thus, the compatible model/candidate can give rise to two types of outputs:
- A class estimation. Example: low, medium or high performance probability.
- A linear estimation in the form of a score or percentage. The higher the score, the greater the probability of success.
You can estimate the quality of a predictive model by comparing it to people on the job of whom you know performance level. This will result in the following table:
A type 1 error corresponds to people whom the model evaluates as having good potential for success, when this is not the case in reality.
A type 2 error corresponds to people whom the model evaluates as having low potential for success, which these people succeed in the position.
Remember that this table is valid for any recruitment system. It happens that the candidates who were not recruited are found to be good performers on the job and, similarly, people recruited then proved to be in a situation of failure.
Keep in mind the extend to which the predictive model can improve the quality of your current decisions. Once this estimation is made, confirm (or not) the implementation of a predictive model in your selection process.
#4 – Measure the impact of predictive recruitment
It is neither the perfect nor the eternal model. Predictive recruitment involves working repetively. A reviews is periodically carried out in order to continue to improve it.
Measuring the impact consists of monitoring the indicator that we want to maximise.
A question may arise: how can we know if the effect on performance is more due to contextual factors or recruitment of different profiles?
There are two ways to answer this question:
– A company never recruits just the most compatible profiles with the model. So you can compare the results obtained by people against the model and those recruited in line with the model.
– If there is a contextual effect, the whole target population will benefit as you can smooth the effect by measuring people who were present at the presentation of the predictive model.
“Machine learning” is also part of the new technology that will optimise the predictive models through real-time updates of the criteria.
Predictive recruitment will gradually establish itself as a reference in the best HR practices. As with any technique, certain precautions must be taken, especially to appeal to actors who master the subject. The presentation of these four founding principles aims to lay the foundations and best practices that should be the establishment of predictive recruitment by an organisation.