End of the Resume Parsing and Candidate Matching as we know it

Krishna Khandelwal
April 20, 2023

"Any sufficiently advanced technology is indistinguishable from magic." - Arthur C. Clarke

I felt very close to magic in real world last week when Hunar.AI's co-founder and CTO, Shantanu Bhattacharyya, demoed the contextual matching algorithms for matching candidates to jobs, which he and his team built over the last 6 months. The algorithms are built using Language Learning Models (LLMs) such as GPT3+ and Neural Net models. This new technology could revolutionize the way we approach recruitment and staffing by making the process more efficient and accurate. Please see for yourself here


Why did Shantanu Bhattacharyya even start building these algorithms ?

His team started on the journey to build the contextual matching algorithms because they discovered that the prevalent ATSes and commonly used resume parsing solutions failed to match candidates to jobs accurately more than 70% of the time. They investigated and found that today's resume parsing and matching rely heavily on text-based keyword matching between job descriptions and resumes, which does not work well in the frontline job market. In this market, job roles, descriptions, titles, and work experiences under commonly used job titles, as well as requirements such as past education and years of experience, are not standardized. Therefore, the evaluation of candidates for job roles is highly contextual and typically relies on an experienced recruiter's tribal knowledge. This lack of accuracy in resume parsing and matching may also be why many recruiters dislike using ATSes.

What exactly goes wrong with today's resume parsing & candidate matching ?

Resume parsing using text-based matching and basic NLP techniques should allow recruiters and hiring managers to quickly identify qualified candidates and focus their attention on the most relevant information. But this doesn't happen. Let's understand what goes wrong with the way resume parsing works currently -

  • The inconsistent format of CVs, especially for frontline candidates, makes it difficult to extract accurate information.
  • Additionally, unstructured data such as free text descriptions of job responsibilities or skills, along with abbreviations and acronyms, are difficult to interpret correctly.
  • The most significant challenge is the contextual matching needed between open job roles and candidates. For example, if you are looking for a Field Sales Representative, the candidate's work experience selling different products to various types of customers may be relevant, but if the candidate's CV does not contain the specific job title or similar permutations, the current resume parsers won't pick up the candidate's profile.
No alt text provided for this image

Because of above mentioned reasons, resume parsing generally leads to a drop in recruitment efficiency instead of increasing it, which is not ideal for recruiters and hiring managers. It's no wonder that some recruiters hate using ATSes, which have become tools for just tracking the recruitment lifecycle without providing any intelligence to make their lives easier.

Is there a solution? if yes, then what is it ?

The solution to the challenges faced by resume parsing is to use language learning models such as GPT models that can match context rather than just keywords or text. This will bridge the gap between the recruiter's tribal knowledge and the current resume parsing solutions, leading to highly relevant recruitment funnels for recruiters to work on and increasing their throughput. The inaccurate or irrelevant matching in current resume parsing solutions leads to downstream inefficiency, with recruiters spending a lot of time calling candidates only to discover that they are not a match.

For recruitment or staffing businesses, recruiter efficiency is paramount, and this solution can significantly improve it, reducing recruitment costs and increasing the overall efficiency of the business. This can also be a reason why recruitment technology has low penetration in frontline hiring compared to white-collar recruitment.

What Shantanu Bhattacharyya and his team have built at Hunar.AI could lead to a new wave of technology driving efficiency in the recruitment processes, not just in India but worldwide.

Please reach out to him in case you want to chat on the models in detail and want to contribute towards his build.