AI Screening of Candidates

No doubt, AI will make a big impact on TA/Recruiting teams, however, it will take some time if ever, it will replace recruiters.

We already have multiple vendors shouting about their technology in screening applicants/CVs/resumes, let's look under the hood of these systems

The type of AI used to screen applicants CVs/resumes etc for the best candidates is typically a form of natural language processing (NLP) and machine learning (ML) technology.

Here's how it generally works:

• Natural Language Processing (NLP): This technology enables computers to understand, interpret, and generate human-like text. In the context of resume screening, NLP helps in extracting relevant information from resumes, such as skills, qualifications, and experience.
• Machine Learning (ML): ML algorithms are trained on large datasets of resumes and corresponding hiring outcomes. These algorithms learn patterns and relationships between the attributes of resumes and the success of candidates in the hiring process. Over time, the system becomes better at predicting which candidates are more likely to be successful based on historical data.
• Keyword Matching: Some systems use keyword matching to identify specific terms or phrases related to job requirements in resumes. This can be a simple rule-based approach or part of a more sophisticated algorithm.
• Semantic Analysis: More advanced systems use semantic analysis to understand the context and meaning of words and phrases in a resume. This helps in identifying not just keywords but the actual relevance of skills and experiences to the job.
• Ranking and Scoring: AI systems often provide a ranking or scoring mechanism to prioritise candidates based on their fit for the job. This can help recruiters or hiring managers focus on the most promising candidates first.
• Continuous Learning: These systems are often designed to learn and adapt over time. As new resumes are processed and hiring outcomes are known, the AI system can refine its algorithms to improve accuracy and effectiveness.

Development

The development time for an AI system to screen applications for the best candidates can vary significantly based on several factors. These factors include the complexity of the system, the depth of functionality, the size and expertise of the development team, and the availability of relevant data. Here are some general considerations:

• Scope and Complexity: The more features and functionalities you want in the system, the longer it will take to develop. For a basic resume screening system, development might take several months. More advanced systems with features like semantic analysis, continuous learning, and integration with other HR processes could take a year or more.
• Data Availability and Quality: Access to high-quality and diverse training data is crucial for developing an effective resume screening system. The time it takes to gather, clean, and preprocess this data can impact the development timeline.
• Algorithm Development: Creating and fine-tuning machine learning algorithms can be time-consuming. The complexity of the algorithms, as well as the need for continuous learning mechanisms, can extend the development time.
• User Interface and Integration: If the system needs a user interface for recruiters or HR professionals, or if it needs to integrate with existing HR software and databases, additional time will be required for design and development.
• Testing and Iteration: Thorough testing is essential to ensure the system's accuracy and reliability. Iterative testing and refinement are common in AI development to address any issues that may arise.
• Regulatory Compliance: Depending on the industry and region, there may be regulatory considerations, especially regarding data privacy and fairness. Compliance with these regulations can add time to the development process.
• Team Size and Expertise: The size and expertise of the development team can significantly impact development time. A well-coordinated and experienced team may be able to complete the project more efficiently.

So would each different role need to be trained, I hear you cry.

• Yes, typically, AI systems for applicant screening would benefit from role-specific training. Training an AI system involves providing it with labelled data, where the labels indicate the desired outcomes. In the context of screening, the labeled data would consist of applications along with information about whether the candidates were successful in their applications or hiring processes.
• Each different role or job position may have specific requirements, skills, and qualifications. Therefore, training an AI system to accurately screen candidates for a particular role helps tailor the system to the specific needs of that position. The system can learn patterns and associations between certain keywords, experiences, and qualifications that are more relevant to the success of candidates in that role.
• For example, the skills and experiences sought in a software developer position might differ significantly from those in a marketing manager role. By training the AI on data specific to each role, you enhance its ability to understand and prioritise the attributes that are most important for success in those positions.
• It's important to have a diverse and representative dataset for training to avoid biases and ensure fair and effective screening. Continuous monitoring and retraining of the AI system are also crucial to keep it updated with evolving job requirements and to adapt to changes in the workforce landscape.
• Some organisations opt for a more generalised approach, especially if they are screening for a wide variety of roles. In such cases, the AI system may be trained on a broader dataset that covers a range of positions. However, for the highest accuracy and relevance, role-specific training is often preferred.

Conclusion

Off-the-shelf AI solutions will be lightweight and generalistic. I doubt if any organisation will hand over or will be allowed to hand over data (GDPR) that will enable the various models to be trained well enough to be of significance.

Most systems could be trained for light touch duties like checking for degree education, living in a relevant location, having the right to work in that location etc