Search is not matching

Most hiring systems begin with two documents: a résumé and a job description. They identify titles, skills, qualifications and keywords, then use those signals to filter or rank candidates. This approach can establish whether the word “Python” appears in both documents. It cannot reliably determine whether someone used Python to automate internal reports, develop scientific models or build production infrastructure serving millions of users. The same keyword may represent fundamentally different levels of depth, ownership and complexity. The reverse problem also occurs. A candidate may possess highly relevant capabilities expressed through an unfamiliar title, a different industry or a non-linear career path. A system optimized for exact correspondence may treat that difference as evidence of weak fit rather than evidence of transferable experience. Two candidates with similar résumés can perform very differently in the same role. Two jobs with the same title can require entirely different levels of autonomy, technical judgment, collaboration and operating speed. A résumé describes selected parts of a person’s professional history. A job description describes an organization’s current interpretation of a role. Matching them as documents cannot fully explain whether the person and the opportunity make sense together.

Building the Intelligent Matching Engine

Employza Research & Development is building an Intelligent Matching Engine designed to evaluate professional compatibility beyond titles, filters and keyword overlap. Our work starts from a different premise: A meaningful match is not a similarity score between two documents. It is a structured assessment of the relationship between a person and an opportunity. That relationship may involve multiple forms of alignment:

Capability Whether the candidate can perform the work at the required level.

Depth How extensively and independently the candidate has applied relevant knowledge.

Transferability Whether experience developed in another role or industry can succeed in this context.

Trajectory Whether the opportunity is consistent with the candidate’s direction and potential for growth.

Environment Whether the role’s pace, autonomy, structure and collaboration model support strong performance.

Constraints Whether practical requirements such as location, availability and compensation align.

Intent Whether the opportunity reflects the work the candidate actually wants to pursue.

These signals do not operate independently. Strong technical overlap may still produce a weak match when the role moves a candidate in the wrong direction. An unconventional background may initially appear less relevant while containing highly transferable experience. A candidate may satisfy most listed requirements but lack the degree of ownership the position actually demands. The central research challenge is therefore not collecting the largest possible number of signals. It is determining which signals are meaningful, how they interact and how confidently they support a match.

From ranking to understanding

Most recommendation systems ultimately produce a list. A candidate sees jobs in a particular order. A recruiter sees profiles ranked against a vacancy. But an ordered list does not necessarily create understanding. An intelligent matching system should be able to examine more than whether one result ranks above another. It should help establish:

where the match is strongest,

which capabilities appear transferable,

where meaningful uncertainty remains,

which requirements may be essential,

which differences materially reduce compatibility,

and what additional information would improve the assessment.

A score should not function as an unexplained verdict. It should represent an interpretable assessment built from relevant evidence. This matters because employment recommendations influence real decisions. They affect which roles receive a candidate’s attention, which people reach a recruiter and which possibilities are never examined at all. A system that ranks confidently without understanding the relationship underneath can reproduce the same failure at greater speed. Our objective is different. We are building toward matching that is more precise, more explainable and more capable of recognizing professional potential that conventional search systems miss.

Evaluating matching as a system

Improving matching requires more than adding a new model to an existing recommendation pipeline. It requires a way to evaluate whether the system is identifying meaningful compatibility rather than merely producing plausible results. Our Research & Development team is developing the evaluation methods, signal structures and scoring systems required to test this distinction. We examine where matching succeeds, where it fails and which assumptions produce unreliable conclusions. We study difficult cases:

candidates with transferable experience from adjacent industries,

roles whose titles do not accurately represent their responsibilities,

profiles with missing or incomplete information,

skills expressed at substantially different levels of depth,

conflicting signals between capability and intent,

and requirements whose importance is unclear from the job description alone.

These are not edge cases surrounding the matching problem. They are the matching problem. A reliable system must reason effectively when career paths are non-linear, professional language is inconsistent and neither side of the market provides a perfect description of reality. That requires continuous evaluation, structured testing and refinement against real employment contexts.

The cost of poor matching

Bad matching does more than waste time. It creates false signals that candidates, recruiters and companies may accept as reality. A qualified candidate repeatedly rejected by narrow filters may conclude that their experience has no market value. A company searching within rigid criteria may conclude that the required talent does not exist. A recruiter may spend more time processing applications than understanding the people behind them. At market scale, the result is a familiar contradiction: Millions of people are searching for work. Companies report difficulty finding qualified talent. Application volumes continue to rise. Yet meaningful connections remain comparatively rare. The market appears saturated on both sides while repeatedly failing to connect compatible people and opportunities. More applications will not solve this. More recommendations will not necessarily solve it either. The underlying matching infrastructure must improve.

What we are building toward

The first generation of employment platforms made job listings searchable. The next generation made them recommendable. The next challenge is to make professional compatibility understandable. That means moving beyond: Does this résumé resemble this job description? And toward: Is there a meaningful, evidence-based reason to believe this person and this opportunity can succeed together? Answering that question reliably requires a different class of system. It requires richer signals, multidimensional reasoning, interpretable scoring and evaluation methods designed specifically for the realities of employment. This is the direction of our work at Employza Research & Development. We are building the Intelligent Matching Engine to reduce low-value recommendations, surface overlooked compatibility and improve how people and opportunities discover one another. Job matching remains a difficult problem. We believe it is also a solvable one.

Research note

This article was developed by Employza Research & Development as part of our work on intelligent matching, professional discovery and the future of employment infrastructure. Our research focuses on developing and evaluating systems capable of interpreting multidimensional compatibility between people and professional opportunities. The underlying architecture, signal models and scoring methodology are proprietary. Employment figures referenced in this article are drawn from “Hidden Workers: Untapped Talent,” a joint research initiative by Harvard Business School’s Project on Managing the Future of Work and Accenture.