Not content with dissecting your CV, the machine now evaluates your speech in every detail. A fabulous time saver for recruiters, one more requirement for candidates. In any case,to use in moderation…
To catch the rare pearl in a tide of CVs, companies are increasingly en relying on tools doped with artificial intelligence. According to a study carried out in France by the recruitment firm Robert Walters, this technology would indeed mean saving time for 41% of recruiters but also 39% of candidates.. This practice is developing mainly in large groups facing large volumes of job applications. Sure, “the machine doesn’t get tired.
She now reads a CV better than a recruiter by being able to analyze thousands of documents and establish matches between applications and the criteria of the ad,” says Laurent Cervoni, doctor of computer science and director of artificial intelligence at Talan. This mega count, coupled with machine learning, more or less efficient, allows many recruitment consulting firms to promise employers a sorting of candidates “objective and scientific”. Could a recruiting robot be more effective than a human? A round-up of the world.
1/ Increased recruitment
In this digital far west, three “smart” intelligents recruitment models are emerging. The first, called “increased,” is to analyze the correspondence between a candidate’s profile and the average profile of those hired in an equivalent position. Example: Smart Recruiters offers employers,via its platform,to manage their recruitment. The start-up registers online applicants and CVs received. “We have 40 million profiles created per year; we follow those who have passed the maintenance en stage, those who have spent two, those who have been recruited. This volume allows for a real analysis,”assures Jérôme Ternynck, director. For a single en offer, the machine sifts through an average of passe 200 applications. She reads them, summarizes them and assigns them a score of 0 to 100 that corresponds to the level of suitability with the position to be filled.
So far, nothing new. But the algorithm “increases” this first analysis by comparing the candidate’s experience, skill and level of training with the thousands of others who have already passed the selection process. This gives a probability of matching between a career and a position. The program also assesses the potential of a degree. It details, for example,the skills provided by the HR Master’s at Paris Dauphine University and establishes in what type of position this diploma is privileged. Result: “In 90% of cases, the person hired is one of those we had initially recommended for an interview,” adds the boss.
- The plus: A sorting by skills with a comparison of factual data.
- The -: Does not assess the candidate’s human qualities.
2/ Predictive recruitment
Predicting a candidate’s success capabilities rather than staying in a descriptive approach to his or her profile is the credo of AssessFirst. Psychologist of training work, its founder Simon Baron publishes since 2002 psychometric tests to evaluate candidates. Tests completed, from 2012, by the creation of predictive models for each of its customers. In practical terms, when a company seeks to recruit on a given trade, AssessFirst determines with it two types of indicators. The former are related to the operational performance of employees who already hold this position in the company concerned. For example,the percentage of targets achieved,the deadlines for completion, etc. The latter evaluate the HR data of the people already in office: their success rate during training, their commitment, their involvement…
All these executives also answer an online questionnaire, inspired by research in occupational psychology (with no less than 90 questions! ) that defines their personality, their qualities and their relationship to others. This data contributes to the machine’s development of a model composed of dominant personality traits to succeed in this given position. “We then keep eight specific criteria that relate the qualities of the framework to these quantitative indicators,” he said. Armed with this “ideal”sketch, all that remains is to compare the profiles of the candidates. In the end, AssessFirst delivers to its client only “people who get more than 60% of correspondence with the eight pre-defined criteria”.
The problem is that the method suffers from many risks of bias. Coming,of course,from the data that feeds the algorithm. In addition, in this case, artificial intelligence is programmed to recruit the same profiles as those who are the best performing in the company. So passe what happens if men perform better than women in this position? “The only data we use is test scores. The criteria of age, sex, origin and even hobbies, which can be gendered, are not integrated into it,” argues Simon Baron.
- The plus: Requires recruiters to identify relevant indicators to reflect the reality of a position (do they do it?)
- The -: Prefers social reproduction and the hiring of clones.
3/ Recruitment of”personality”
“HR is hungry for tangible data, such as diplomas, but at the same time shows a desire to benefit knowledge and soft skills, which is a bit contradictory,” says Pierre Trippitelli,managing partner Europe of Perpetual, an HR consulting firm. On the model of the American HireVue, the hexagonal company Itwapp has set itself on the niche of deferred video maintenance with AI. His premise? Help recruiters sort candidates based on the data suggested by their facial and oral expressions. Two thousand companies would use the InterviewApp app, which invites candidates to film themselves in front of their webcam by answering a series of selected questions. Not only do employers save 70% of time on making appointments and performing interviews physically, but machine analysis offers them a codified sort.
“Our platform evaluates up to 200 criteria on image, audio and semantics, but it’s up to recruiters to choose the ones that are important to them,” warns Julien Dargaisse, director. The choice of criteria does not fall from the sky: it is inspired by the five major personality factors (Big Five), a classic of research in behavioral psychology. . The goal? Identify by analysis of body and oral language, openness (curiosity), conscientiousness (control,discipline…), extraversion, agreability and neuroticism (negativity)of the candidate.
The machine classifies the elements of the candidate’s language, the richness of his vocabulary, the intonation,the modulation of his voice,the length of his sentences, revealing his ability to synthesize,his speech flow (X words per minute), etc. Once all this data is collected, it is correlated with the famous Big Five test… according to an algorithm that Julien Dargaisse prefers to “not detail”.
“The best rated candidates are generally those with the greatest grande lexical diversity, which denotes their mastery of the subject,and whose discourse is fluid,” says Mickael Cabrol, founder of Easyrecrue. The start-up says it does not use facial analysis, as the technology is far from perfect. For the time being, it proposes to its clients to transcribe the entire speech of the candidate to evaluate it according to five criteria: prosody (accentuation, intonation), fluidity of the subject, verbal skill, speech flow and verbal content. It takes less than three minutes for the machine to pour analyze these criteria and propose a rating between candidates who have responded to the same offer. Criteria that question Laurence Devillers. For this professor of artificial intelligence at the Sorbonne University, “there is a huge cultural dimension in the way we express, ” ourselves. What do we do with stuttering people, those who express themselves naturally slowly, those who have an accent? “
Leading a research team at the CNRS on the emotional and social dimensions in spoken interactions, she adds: “Detecting emotions in the voice or the meaning of silence remains very random.” For Cecile Jarleton, PhD student in occupational psychology at the HR Lab, “the analysis of all these factors does not necessarily presume competence. There is a risk of a still frequent amalgamation between personality traits and performance.” To ensure the seriousness of the rating criteria used by its algorithms, Cabrol recalls that they were developed in collaboration with a scientific committee composed of researchers from CNRS, Telecom ParisTech and CentraleSupélec. “
But it is true that we still have to adapt to a business corpus and integrate the corporate culture to create dedicated versions of our algorithm.” The introduction of artificial intelligence as a decision-making tool will force recruiters to define their evaluation criteria for a given position in a very detailed way. Maybe that’s the good news for the candidates!
- The plus: Objectivizes the interview by placing all candidates under the same conditions in the face of the same questions.
- The -: Lack of transparency about what is assessed and the actual suitability with the required skills and the culture of the company.
Practice convincing the machine
During a delayed video interview, your time is running out. You will have to answer 5 or 6 questions in one or two minutes each. Can’t start over or re-record your response. Once started,the process does not stop. Adecco offers on its website a handy simulator to practice and review your performance. Knowing that the machine will scrutinize your speech, work your answers with a stopwatch.
Before each answer,the program usually leaves you between 30 seconds and one minute of reflection. Without knowing the questions in advance, train your oral expression. You can list the lexical field related to the function for which you are applying. List the terms,the variety of expressions, so you remember them during the interview. A silence greater than 2 seconds will be taken for a hesitation… Courage!