State of Resume Review

Introduction

Resume review is an industry unto itself.  This is the primary effort of most recruiters that use sourcing tools to identify candidates who have resumes.  Once they find a potential candidate, they reach out in the hopes of getting the resume from the “passive” candidate and then they put it in a pile for review.  The passive candidates are considered to be the best candidates, and an enormous effort is made just to get the resume.

At the same time employers post jobs and the “active” candidates apply and provide their resume and they too go into the big pile for review.

Some organizations focus on passive candidates but still must review the active candidate resumes while other organizations find the people they need in the active candidates that apply.

In all cases the resume must be reviewed.

Resume Review is a tedious task that is often done manually because either the organizations lack trusted technology, to do the work, or it represents part of the job and therefor livelihood of the sourcing recruiter.

And then there are the trusted technologies that are expected to reject 75% of what they review.  The dreaded ATS or Applicant Tracking System for you beginners.

“ATS platforms reject about 75% of resumes”

                                                            CEOMichaelHR

In all cases the review is too cursory for most people’s tastes.  Automated systems are using keyword searching techniques that are 10+ years old.  Manual review requires the ability to read document after document, make a complete inventory of the candidate’s experience and accurately compare that to each requirement in the job description.  And that is beyond the capacity of most humans.

The result is people are rejected with little to no reasoning or they are ghosted having never been properly considered.  Candidate experience is at an all-time low and organizations fail to prioritize the people that truly fit best, so productivity falls in the talent acquisition team and in the overall workforce as well.  This represents huge cost to corporations and to the economy.

What is the state of Resume Review?

The answer is…Not Good! 

Let’s look at manual efforts first. 

This is done by people reading resumes and people don’t like to read.  They hate to scroll and after little effort experience fatigue. 

When an organization posts a job for a professional position it often receives over 500 applicants.  Sometime only a couple of hundred and sometimes over one thousand.  Organizations often post the job and turn off the post after just a few days because of the huge number of inbound resumes.  This overloads their process and also prohibits them from considering good candidates that didn’t apply before the post was shut off.

There is simply too much to review.  Humans are not consistent machines.  They vary in skill level from reviewer to reviewer.  Some are simply better at doing review than others.  But even the best reviewers, and the worst, get tired and so the quality of their review diminishes from hour to hour.

“34% of employers say they receive too many resumes to read”

                                                            TestGorilla

In order to manually do a good review of a resume the reviewer needs to:

1)    Keep the details of the jobs requirements in mind while doing each review,

2)    Keep a detailed inventory of the candidate resume in mind,

3)    Do a detailed and accurate comparison of the candidate’s capabilities to the job’s requirements, and

4)    Calculate some type of determination of fit.

This type of effort would typically take an analyst about 30 minutes to do for each resume.  However, industry estimates for average time spent one manual resume review is 6-8 seconds.  That tells you how accurate it is.

Now let’s take a look at Automated Solutions. 

A wise person once said that to err is human but to really foul things up requires the use of a computer.  While it is true that a computer system or some kind of HR Tech designed to review resumes will move through them fast, it is also true that if it makes mistakes, it will make a whole lot of them in a short space of time.  This is what is happening.

The latest technology is about 7-10 years old and is largely based on keyword searching and uses older AI.  If it was built prior to 2020, and most are, it leverages older Machine Learning (ML) models that are characterized by the following issues:

·      They look for keywords found in Job Descriptions

·      They don’t understand the semantic context of the keywords

·      They take time to train so it is costly to get them running

·      They learn bias from either prior employment or recruiter behavior training data

·      They experience “drift” which means the models degrade in accuracy over time

·      They are an opaque box so they can’t explain or produce an audit of how they came to their conclusions

·      Humans are not in the loop and are not able to influence the decisions they make as they are making them

These systems are also imprecise when it comes to the scoring which directly effects the ability to rank results and properly match people to jobs.  The issues with scoring are as follows:

·      Most of the state-of-the-art technology assigns inaccurate or broad scores to candidates:

o   A, B, or C

o   Hot, Warm, or Cold

o   Harvey Balls or Percentage

·      Most only indicate a skill (keyword) is present and don’t score the skill itself:

o   How do you compare two people that have the same skill listed in the resume?

·      Most include non-skill characteristics in the score.  These should be filters, considered separately and not part of skill scores.

o   Prior employer,

o   College attended,

o   Location, etc.

·      You can’t produce a proper sequential rank with imprecise data like this.  And you need a ranking to go through the candidates in order of fit from best to worst.

So, the result is rapid inaccurate bias prone review and scoring of resumes and a large number of applicants are rejected instantly.

Also, automated resume writing technology can trick these systems and get a candidate resume into the acceptable list by adding keywords when the candidate really is not as good a fit as they should be to advance in the process.  The systems lack veracity checks to pick up on tricks and inconsistencies in the construction of the resume.

Many people don’t like to rely on automated resume review because of these issues.  When forced to resort to automated systems due to high inbound resume volume, organizations impact candidate experiences for a great many applicants.  It also reduces effectiveness of the Talent Acquisition team, protracts the Time-To-Hire and impacts employment costs and workforce productivity due to poor hires that leave quickly.

“30% of new hires leave within 90 days of starting”

                                                                        Forbes 2023

Please don’t get me wrong.  New hires are leaving due to poor manual resume review as well as poor automated resume review.  We are populating our new hire pipelines with poorly fitting candidates with both our manual and automated efforts.

The interviewers and hiring managers are doing the best they can and hiring the cream of the crap.

So, where are we with Automated Resume Review?

We are actually at a very interesting place.  Only a geek like me would find this interesting but I know there are many of you out there in HR Tech land.

I like to describe where we are as on the verge of a new generation of computer guided Resume Review technology.

Manual efforts are still the best!  The only problem is that markets have evolved to a point where the volume has reduced average review time to just a few seconds per resume and that is just not enough time to do comprehensive, accurate, and fair comparisons.  But the computer is catching up, becoming collaborative, and will now change the state of affairs.

So, what are the generations of Automated Resume Review?

The First Generation of Automated Resume Review technology was non-specialized primitive efforts that were built into the earliest HR Systems that were pointed at the hiring process.  You can think of these as the earliest ATSs.

Many of the newer versions of ATSs have modules for Resume Review but, as I am sure you know, ATSs are built for tracking candidates and not so much for reviewing, scoring, and ranking them to optimize all the downstream processes of interviewing, evaluating, and selection.

The Resume Review modules are afterthoughts or checkboxes for the most part.  This is why the ATS providers are acquiring Second Generation technologies to replace their Resume Review modules.  You will note some recent purchases by leading ATS vendors.

The Second Generation of Automated Resume Review is done by purpose-built Candidate Matching & Ranking systems that are designed to do just this job, and some related tasks but they are the first effort of purpose-built solutions.  They are pretty good, but they are 7-10 years old and have the limitations of Automated Systems built with older AI/ML models that lack precision and understanding of skills in context as described above.  (They also tend to have a fair amount of overlap with ATS features but that is for another article.)

In Silicon Valley there is a saying amongst computer technology developers that goes something like this…

·      Version 1 is what the software engineers wanted,

·      Version 2 is what the users needed, and

·      Version 3 is what the users needed, working!

We are at the cusp of Version 3, transitioning from Second Generation to Third Generation which will be what the Resume Review users need, working!

The Third Generation of Automated Resume Review is just now emerging.  The technology is very new and uses more modern AI than the Second Generation.  This is a key point in that it uses Generative AI instead of basic Machine Learning Models and if it is good at what it does it uses the AI differently as well.  The difference being it doesn’t ask the AI to make qualitative decisions about the candidates and their fit to the job.  Don’t be fooled by Second Generation systems that are adding GenAI on the periphery and not replacing the core engine with a completely different way of operating!  Also, don’t be fooled by systems that ask the GenAI to do too much including making judgements!

These brand-new solutions are designed to avoid the pitfalls of the current Second Generation offerings.

The issues with the Second Generation solutions have caused many organizations to back off using them and revert to manual processes.  Specifically, they have determined, with the help of their legal departments, that learned bias and a lack of an audit trail “expose the organization to an unacceptable level of risk”.  AI has become a bad thing and legislation like NY Law 144 has frightened people back to no computer systems.  But the Third Generation is not using AI the same way.

The Third Generation is designed to avoid the problems of the past and not make the decisions for the reviewer but rather do very detailed analysis and summarize it for the reviewer so the reviewer can make Rapid Better-Informed Decisions!  The solutions are collaborative and support the reviewer.

Third Generation resume review technology uses modern methods of AI and Analytics to do analysis that is more akin to the methods used in a Cyber Security Operations Center to help people make decisions on what is the most important thing to look at first, then second, and so on.  They make much more accurate unbiased estimates of skill sets and mastery levels by understanding skills in the context of their domain, the amount of time the candidate used the skills, the recency of using the skills, and by assigning scores to each skill that are summed to produce the overall fit score and ranking. 

Now Resume Reviewers will know which candidate to look at 1st, 2nd, 3rd, etc.  based on precise scoring and ranking.  When looking at the candidates in a precise ranked order the reviewer will be supported with a new level of optics or visibility into how the candidate fits the position and will fair in the role over time.  This will help to stabilize candidate careers and improved the productivity accrued by adding better fitting candidates with optimal skills sets to the organization.

The systems correlate a lot of information and guide the pilot instead of taking over and flying the plane.

Conclusion

This is an exciting time.  We are on the verge of a new generation of Candidate Review that will deliver better fitting happier hires, productivity, and profitability for the employees and the companies.  It will optimize the hiring process, ultimate productivity, and competitive posture of all participants.

In order to take advantage of this new generation and get the benefits you will need to look beyond the marketing speak of the Second Generation players, peel the onion, and do real evaluations of the new Third Generation solutions to make sure it will deliver for you.  I know you will do great.

I wish you all the very best as you navigate this exciting evolution in resume review!