Now that the entire world is flooded with news and articles about ChatGPT, the issue of training AI models might become interesting!  A lot of people believe that adapting an algorithm is just a matter of adding additional data.  When results are not good, it seems to be a matter of adding additional data…to some extend this is correct if you have vast amounts of data….

But HR related is not that easy to find. It is not because a job is a bad match for someone, that it is for someone else or that one bad case is enough to learn to get better matches as of that moment…

Not all jobs are the same, and a bad match for one job is not a bad match for a similar job.

The amount of data needed to train an AI system for job and candidate matching depends on several factors, such as the complexity of the problem, the diversity of the data, and the performance requirements of the system.

Typically, more data leads to better performance in AI models, but there is no minimum amount of data required. With one job and feedback on 100 CV’s, it may be possible to train a basic model, but the performance may not be sufficient for real-world use. The quality of the data is also important, as data with high variability, diversity, and no biases is more useful for training.

In general, it is recommended to have at least several hundred to thousands of labeled examples to train an AI model with good performance. However, the specific amount of data needed can vary depending on the specific problem and the type of AI model used. It may also be necessary to continuously update and retrain the model with new data to ensure its performance remains accurate over time.  But learning from a single case will remain difficult – self learning for HRM is not as easy as in more general domains…

To address this, one possible approach is to use domain-specific knowledge and job-specific metrics to determine the similarity between jobs and candidates. That is what Actonomy is doing while using feedback from recruiters and candidates.  Our hybrid approach results in an easy way to manage feedback without the need to find hundred of thousands of qualified feedback on good and bad matches.  Contact us and find out more on how this is done and how the ‘bad match’ button works in our ATS/CRM plug-ins.