AI in HRtech Failed Us: How to Improve It?

The promise of AI in HRtech has fallen short of recruiter, industry and startup investor expectations. AI has the potential to revolutionize the recruitment process, making it more efficient, effective, and fair.

However, the current state of AI in HRtech has failed us, primarily due to the limitations of available data, biased decision-making, and a lack of focus on underlying human-level insights.

How can we improve AI in HRtech? By incorporating contextual, two-sided job matching and creating a user experience that mirrors the bi-directional nature of recruiting, plus using data that reflects each of us and our personal strengths, competencies, preferences and mindset. This goes way deeper than skills and biographical data.  

CVs are marketing docs full of noise

CVs, which are often the cornerstone of recruitment, are marketing documents designed to showcase a candidate’s skills and experience. However, from a data perspective, CVs are full of noise and fluff, which can make it challenging for AI algorithms to accurately assess a candidate’s suitability for a role. For example, candidates might exaggerate their skills, use buzzwords without context, or focus on irrelevant details.

To improve AI in HRtech, we need to focus on extracting relevant, structured data from CVs, such as quantifiable accomplishments and specific skill sets. This can help eliminate the noise and fluff, allowing AI algorithms to make more accurate assessments of candidate suitability.

Job posts are ads full of noise and fluff

Just as CVs are marketing documents, job posts are ads designed to attract potential candidates. As such, they too are filled with noise and fluff that can obscure the true requirements of a role. Companies may use buzzwords, vague descriptions, or list unrealistic expectations, making it difficult for AI algorithms to accurately match candidates with suitable positions.

To address this issue, job posts need to be more standardized, must include salaries (so job markets with prices can start functioning) and focused on essential skills and competencies.. However, that still won’t solve the problem…

Job Boards are ad exchanges

Job boards are essentially ad exchanges, where job posts serve as ads. They’re incentivized to sell you more job posts for longer periods of time. This creates a marketplace where the primary focus is on attracting attention, rather than accurately representing the needs of employers and the skills of candidates. This environment exacerbates the noise and fluff issue, making it even more challenging for AI to function effectively in HRtech.

The lack of salary data is fundamental to job markets (job boards) functioning poorly – the price signal is missing in the market.

To counter this problem, we need to reimagine the job board model, focusing on creating platforms that prioritize accurate representation and matching, rather than simply gaining attention.

Biased decision-making in recruiting

Human decision-making in recruiting is inherently biased, with factors such as gender, race, and age often influencing hiring decisions. AI has the potential to reduce these biases by analyzing data objectively. However, if the data fed into AI algorithms is biased itself, the output will be biased as well. For example, if an algorithm is trained on a dataset where men are predominantly employed in tech roles, it may perpetuate this bias in its recommendations.

To improve AI in HRtech, we need to ensure that the datasets used to train algorithms are representative and unbiased. This can help promote fair and equitable hiring practices.

Focus on public data is wrong!

Over 99% of AI HRtech startups have focused on public data, such as CVs, LinkedIn profiles, and job posts. While this data is undoubtedly valuable, it fails to capture the underlying human-level insights and user-generated proprietary data that enables accurate job matching. This data includes personal values, work preferences, and cultural fit, which are critical factors in successful job placements. Humans are much more complex than data off of CVs.. 

To enhance AI in HRtech, we need to shift the focus from public data to proprietary data that captures these human-level insights, enabling more effective job matching and improving the overall recruitment process.

Recruiting is a two-way street: user opt-in and UX are key

Recruiting is much like dating, in that it is a two-way street where both parties are trying to find the right match. However, many HRtech solutions fail to consider this dynamic, providing a user experience (UX) that is one-sided and overly complex.

To improve AI in HRtech, we need to develop bi-directional UX that simplifies the process for both candidates and employers. This could involve creating intuitive, easy-to-use interfaces that enable candidates to express their preferences and employers to communicate their requirements effectively. By fostering clear communication and engagement between both parties, AI can facilitate better matches and improve the overall recruitment process.

For user-generated data to be effective, job seekers must first opt-in and actively engage with the process. This can be done through various means incentivizing participation eg. by providing career guidance. By opting in, users are granting permission for their data to be collected, analyzed, and used for the purpose of job matching.

Proprietary, User Generated Data is key to fixing Job Market issues 

The modern job market is changing rapidly with remote work and upskilling, and with it comes the need for innovative solutions to match job seekers with their ideal roles. Proprietary, user-generated data is emerging as a key solution to addressing this problem. 

Job matching is a contextual and two-sided process, where candidates and employers seek the right fit. Traditional recommendation systems and models often fail to account for this, resulting in suboptimal matches. To improve AI in HRtech, we need to develop recommendation systems and models that incorporate both sides of the equation and are based on job fit assessments.

One approach is to use collaborative filtering techniques, which can take into account the preferences and experiences of similar users on both sides (candidates and employers). By incorporating this contextual information, AI algorithms can provide better recommendations that account for the specific needs and preferences of both parties.

User-Generated Data in Occupational Psychology 

In the field of occupational psychology, user-generated data refers to information collected from individuals about their skills, experiences, preferences, and goals. This data can be collected through various means such as surveys, assessments, and even social media activity. Psychometrics, on the other hand, involves the measurement of psychological traits, such as aptitude, personality, and intelligence, often through standardized tests.

By combining these two areas, data scientists can create rich, personalized profiles for job seekers, allowing them to better understand their strengths and weaknesses, and ultimately find roles that align with their unique capabilities. 

Recruiting is like Dating and Job Matching is a Dating Game

The process of matching job seekers with potential employers can be likened to a dating game, where compatibility is key. In both scenarios, participants are looking for a suitable match based on factors such as values, interests, and long-term goals.

By leveraging proprietary, user-generated data, the job matching process becomes more precise and efficient. Much like modern dating apps that use algorithms to suggest potential partners, advanced machine learning techniques can be employed to identify suitable job matches based on the unique profiles of job seekers and the requirements of employers.

Similar to the world of dating, recruiting is all about finding the right fit between two parties – the job seeker and the employer. By understanding the needs and preferences of both sides, proprietary user-generated data can help streamline the recruitment process and lead to better outcomes for everyone involved.

In Summary

Given the above, think about ChatGPT (ChatGPT plugins, automation, etc.) being used on public data eg. using data from CVs (marketing fluff), job posts (marketing fluff) and other ‘signaling’ information will not improve the ills of job market matching, because it’s a peer-to-peer problem that companies are responsible for the most (as we like to put it: 80% of problems in recruiting rest with organizations, not job-seekers).

ChatGPT and LLMs will help save time through automation, but we’re wondering whether what it will in effect do is create an even bigger screening problem – more job post clicks, more applicants per job post, more spam on LinkedIn messages, more spam in email, and even more disengaged candidates who have to go through recruitment marathons and stupid hoops inside every company. Hence, deepening the problems of candidate screening and person-to-job and person-to-team matching in recruitment. 

AI has the potential to transform HRtech, but it has fallen short due to a focus on noisy, biased public data and a lack of attention to human-level insights and contextual, two-sided job matching. By addressing these issues, we can unlock the true potential of AI in HRtech and create a more efficient, effective, and fair recruitment process that benefits both candidates and employers. This will enable the future of work!

Robert