Predictive Recruitment: fact or fiction
- 12 dec, 2017
On the Adver-Online Gotomyjob.com e-recruitment platform, you can put together your own campaign. The platform helps with campaign suggestions to select the most relevant channels for each type of vacancy, in a smart, fast and how to reach your target group. We also call this predictive recruitment. But what exactly does such a predictive recruitment campaign suggestion entail, how does this come about and how does it work exactly? We talk to Lukas Heeren, Data Scientist at Adver-Online, about predictive recruitment and the role of machine learning in this.
What is machine learning?
Machine learning is the process of allowing software to learn data or information almost autonomously. Contrary to popular belief, machine learning algorithms have been around for more than 60 years. In 1952, the first machine learning algorithm was written by Arthur Samuel for a chess game. In the 1990s, the algorithms were rewritten and the switch from knowledge-based algorithms to data-based algorithms was made. From then on it was possible to have large data sets analyzed and to apply models to them, and people no longer had to enter the rules themselves.
If machine learning was known for a long time, why has it become important right now?
This is due to the restrictions at the computers of the time. Some may know Moore’s law. Gordon Moore made a prediction in 1965 that the number of transistors (and therefore the calculation speed) doubles every 12 months. Today, this law has been revised to 24 months, but it indicates that it is going incredibly fast. To imagine this quadratic relationship, ask yourself how often you need to fold a 1mm paper in half before reaching the height of the moon. The answer is only 39 times! In fact, you are already far past it. The technology is therefore advancing at an unimaginable speed and with it the possibilities for machine learning to analyze more and more data. The computers are now fast enough to analyze Gigabytes or sometimes Terabytes of information and to extract actionable actions from it at the same time.
Companies such as Google or Amazon make supercomputers available to do these calculations and they make extensive use of it.
Everything can be calculated or predicted. An insurance company can calculate how likely someone is to cause damage, health insurance can optimize the healthcare premium, psychologists can better analyze the causes of depression, one can convert written text to digital text and Adver-Online can predict how many candidates one can expect for a certain recruitment campaign.
Why an e-recruitment platform? How dit it orginate?
If you can automate something, you have to do it. Otherwise, you are wasting money. With the e-recruitment platform we automate the process of package composition and we optimize expectations management for the recruiter. The recruiter indicates what kind of person they are looking for and the system shows the optimal campaign composition and the expected results in number of views, candidates and interviews. This is based on three data / information sources: The first source is legacy data from Adver-Online. This involves looking at the completed campaigns. This data is used as a source for the expected number of views, candidates and interviews and is predicted with the most recent data and machine learning techniques. All recent and most successful machine learning techniques, such as the Deep Learning Artificial Neural Network technique and the most successful Regression Trees, have been compared.
The second source is knowledge of the account managers of Adver-Online. These compile the optimal configuration of distribution channels and check the results of the predictive machine learning models. We do not blindly trust our models and put a lot of effort into validating the results. We use the knowledge of the account managers in such a way that we focus on quality over quantity. Models quantify theoretical constructs, but do not convey any substantial meaning. Because quality is hardly quantifiable, the models focus on quantity instead of quality. So if we determine everything by our models, we will get more candidates, but the quality of these candidates will not be taken into account. By combining the knowledge of the account managers with the predictive power of our machine learning models, we achieve the optimal result.
The third source is Social Media. You cut yourself in the fingers these days if you don’t use social media in your campaigns. We work closely with Wonderkind (formerly Recruitz) to put together optimal social media campaigns. Job boards only feature active job seekers, but almost everyone is on social media. Through active and accurate targeting we can reach the latent jobseeker and in this way attract more, but above all good quality, candidates for the campaign. All data from the social media campaigns have enriched the data of Adver-Online.
The second source is knowledge of the account managers of Adver-Online. These compile the optimal configuration of distribution channels and check the results of the predictive machine learning models. We do not blindly trust our models and put a lot of effort into validating the results. We use the knowledge of the account managers in such a way that we focus on quality over quantity. Models quantify theoretical constructs, but do not convey any substantial meaning. Because quality is hardly quantifiable, the models focus on quantity instead of quality. So if we determine everything by our models, we will get more candidates, but the quality of these candidates will not be taken into account. By combining the knowledge of the account managers with the predictive power of our machine learning models, we achieve the optimal result.
The third source is Social Media. You cut yourself in the fingers these days if you don’t use social media in your campaigns. We work closely with Wonderkind (formerly Recruitz) to put together optimal social media campaigns. Job boards only feature active job seekers, but almost everyone is on social media. Through active and accurate targeting we can reach the latent jobseeker and in this way attract more, but above all good quality, candidates for the campaign. All data from the social media campaigns have enriched the data of Adver-Online.
Is machine learning and artificial intelligence the holy grail
Yes and no. Both can do a lot of work for us by automation and helping us making decisions. But there is still much to be achieved in the recruitment sector and the e-recruitment platform is a good start.
However, one should never blindly rely on machine learning models. Machine learning and artificial intelligence often only scratch the surface. A good intuitive example is the strongly significant correlation between the number of ice creams sold and the number of child deaths from drowning. If we were to blindly trust the results, we would immediately remove all ice cream from the shelves. However, there is a clear underlying variable here, namely temperature. When the temperature goes up, people eat more ice cream and therfore swim more. This is a very simple example, but errors like this can be less obvious. At Adver-Online we have our managers who can avoid mistakes like this, by merging their business rules with predictive models and actively validating the outcomes.
However, one should never blindly rely on machine learning models. Machine learning and artificial intelligence often only scratch the surface. A good intuitive example is the strongly significant correlation between the number of ice creams sold and the number of child deaths from drowning. If we were to blindly trust the results, we would immediately remove all ice cream from the shelves. However, there is a clear underlying variable here, namely temperature. When the temperature goes up, people eat more ice cream and therfore swim more. This is a very simple example, but errors like this can be less obvious. At Adver-Online we have our managers who can avoid mistakes like this, by merging their business rules with predictive models and actively validating the outcomes.
Machine learning has a lot of potential, and in that respect it has the opportunity to become the holy grail. But only if it is provided with a healthy portion of common sense and well-funded knowledge.
Share on linkedin
Share on twitter
Share on facebook
Share on email
