Machine Learning (ML) is among the trendiest technologies nowadays. The machine learning engineers are in high demand, the investors are ready to pay a premium if advanced machine learning algorithms have been deployed, and entire industries are in fear of being disrupted and jobs being cut.
For some time, the recruitment industry was focused on sourcing machine learning engineers. After all, what a recruiter has to do with algorithms and how can they change the recruitment process?
A simplified recruitment process looks like this:
There is a role that needs to be filled.
The role’s requirements, responsibilities, etc. are defined and approved.
Candidates are sourced through various channels – job boards, staffing agencies, head hunters, referrals, etc.
All CVs, references, etc. are reviewed, systemized and documented (depending on the internal procedures).
A short list is built and the candidates go through several rounds of interviews.
Eventually, an offer is extended and accepted and the role is closed.
The process takes one to six months to complete and so far, has been automated to some extent by recruitment software solutions and applicant tracking systems. Some recruitment software solutions rely on matching algorithms but are these matching algorithms and machine learning one and the same thing?
By definition, ‘’ Machine learning is a field of computer science that often uses statistical techniques to give computers the ability to "learn" (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed.
The name machine learning was coined in 1959 by Arthur Samuel. Evolved from the study of pattern recognition and computational learning theory in artificial intelligence, machine learning explores the study and construction of algorithms that can learn from and make predictions on data – such algorithms overcome following strictly static program instructions by making data-driven predictions or decisions,:2 through building a model from sample inputs. Machine learning is employed in a range of computing tasks where designing and programming explicit algorithms with good performance is difficult or infeasible; example applications include email filtering, detection of network intruders or malicious insiders working towards a data breach, optical character recognition (OCR), learning to rank, and computer vision.
Still wondering how this definition relates to the recruitment process described above?
Let’s start with the matching algorithms which are easier to understand. Here is a case scenario:
Mr Jones, a recruiter, is in charge of sourcing a CFO for a SaaS company.
A good recruitment software can match the applicants based on the number of skills, salary expectations, location, willingness to relocate, etc. making it easy to filter the candidates, check the experience and referrals of the top 10 or 20 candidates only. A short list is created in hours instead of in days allowing to work on a number of roles within the same timeframe that was previously required for just one role.
At this stage, the recruitment process is already optimized but still depends heavily on human judgment and prone to bias. It is entirely in Mr Jones hands to decide if a candidate lands on the short list if the applicants happen to have a similar number of matching skills, experience and salary expectations. Quite often, this bias results in a ‘’bad hire’’.
According to a new CareerBuilder survey, companies lost an average of $14,900 on every bad hire in the last year, and it’s a common mistake — nearly three in four employers (74 percent) say they’ve hired the wrong person for a position.
74% is a striking number. A big organization will survive and compensate the lost productivity, low quality and lost customers resulting from ‘’bad hires’’. The negative impact on revenue and EBITDA will be there but it will not matter as much as for a startup or a small organization.
According to CBInsights, ‘’running out of cash’’ and ‘’not the right team’’ rank as number #2 and #3 among the reasons for startup failure.
Removing the recruitment bias can save tons of cash and lots of startups will not cease to exist.
So how can machine learning help?
Imagine that there are 5 candidates having similar skills, experience and salary expectations. All advanced in their careers, no career gaps, lack of references or other objective reasons to help a human to objectively rank the candidates. Mr Jones is obviously facing a challenge to not only select the 3 candidates that are to be short listed and interviewed but also to provide the rationale behind his decision if the company ends up with a ‘’bad hire’’.
Well, there are just 5 candidates in front of Mr Jones’ eyes but there are thousands or even millions of candidates in the database having similar profiles. A proper machine learning algorithm will ‘’learn’’ from the data and predict the probability of success of each candidate.
If this still sounds surreal, take a look at the example below.
In the database, there is data regarding the career path of 3 000 people whose title is ‘’CFO’’ and the industry is defined as ’’IT/Software’’.
75% of them are chartered accountants – CPA, ACA, ACCA, etc.
60% have started their careers with a ‘’Big 4’’ or ‘’Big 20’’ consultancy company
35% have experience with McKinsey, AT Kerney, etc.
55% have M&A experience
14% have an IPO experience
45% have their entire career path within the IT industry
7% have led a company to an exit of USD 1billion or more
1% have built their careers with one employer only
25% have changed their employer every 3 years
…and so on…
Combining all the criteria, assigning ‘’weights’’ based on the importance, etc. defines the profile of the candidate that is most likely to be successful, and if properly ‘’trained’’, the machine learning algorithm can predict the probability of success of each candidate on Mr Jones’ list.
Although the above is an extremely simplified example, it still illustrates the power of technology and its potential if deployed in the recruitment industry. Many experts argue that AI has its limitations and can’t take into account factors such as luck, being at the right time at the right place, social skills, change management and crisis management skills, etc. However, even if a person was lucky enough to get a job, staying with that job for 3+ years suggests that s/he has relevant skills.
The taxonomy is also a challenge as a startup with 10 employees is likely to have CFO while an industry expert working for an organization having 10 000+ employees may still be an Assistant Finance Director. To make it even more complicated, the industry classifications are not always harmonized and vary from country to country.
What about the ‘’creative’’ titles that emerged in the tech industry in the last decade including ‘’SEO guru’’, ‘’QA ninja’’ and the like? Will there be a solution removing the bias originating from comparing apples with pears aka titles and responsibilities having nothing to do with each other?
Will AI and machine learning cut jobs in the recruitment industry, and if so, how many?
Transformify is a CSR Recruitment Platform leveraging on HR-tech, fintech and AI to help businesses access talent, enter new markets, transfer secure payments and boost sustainable growth.