About this document

This document gives an introduction to the search model of Eva, and how to get the most value from it.

The search practice today on Job Boards

Most of the recruiters encounter search for sites like Monster, CareerBuilder etc., We have tried to keep the search metaphor similar, yet supporting the ability to provide advanced search.

Simple keyword search

Here, the recruiter provides a set of keywords and the site throws up candidates who have the most proximity to the keywords. This is the most used search style among recruiters and the most flawed. 

There is no way to mention importance of the various skills in this type of search. All skills are treated as equal, which is never the case in the real world. Imagine looking for a java developer with java and html skills. The 2 skills are not the same. You expect the candidate to have worked extensively on Java with some experience in html. A candidate who has a lot of experience in html with some experience in Java is not a good fit.

The volume of results is high, but with a lot of candidates missing the primary skill set because of the lack of any mechanism to show which are the main skills to look for.

Must have keywords

This is usually supported in addition to the Simple keyword search capability and helps in eliminating the candidates who do not have the skills that the requirement needs mandatorily. This is a better constraint to provide in addition to the simple keyword search.

This search still has the constraint of not being able to call out the more and less important skills of the candidate. This will return candidates who may have all the skills required but not with the depth/expertise required. The consequence is for the recruiter has to open every resume, only to find the candidate not having the level of expertise for the specific skill and rather having a passing knowledge.

Boolean search

This is the most advanced version of the search that the sites provide. This form of search takes some time to get used to by the recruiter and is rarely used. This search gives much better quality results compared to the above but has its own limitations.

In addition, Boolean search also does not allow for indicating the level of expertise that the requirement needs for the various skills.

Eva Search

Eva search considers the following for skills:

• Or group - A group of skills, where the presence of one of the skills indicate that the requisite skills are available. E.g., A group (angular OR react OR jquery) indicates the need for a candidate who has any of the skills angular, react, or jquery.

• Flag if the skill or a group is mandatory or not.

• Skill level - The level of expertise expected from each skill. The details are documented in the next section.

The goal of the search model

• To provide a unified search model that works seamlessly across the various sites.

• To provide a search model that is simple enough to understand and yet returns the best-fit candidates for a job.

• To provide a search that can truly represent the requirement, and can even be a better reflection of the requirement than the JD itself.

Search within Eva

Eva has 3 skill levels

1. Expert

2. Proficient

3. Nice to have

You can also indicate if the skill is “Mandatory”.

Step - 1 (Defining the title to look for)

The first step is to decide if you need to specify the role at all. For this check for the below indicators:

1. Important that the candidate has played a specific role in the past

2. If the role is an important defining feature of the requirement beyond the skills. 

Example 1:  For a CA (Chartered accountant) it may be important to specify the role as part of the search and not depend on the skill “accounting” only. A candidate could have had accounting skills, but as a treasurer and not as a CA. 

Example 2: For a Java developer position it is important to specify the role as developer and not only specify the skill as Java. A candidate could have had Java skills, but as a tester instead of a developer. In this case, indicating the role of the developer will filter out the testing profiles from being returned.

Step - 2 (Define the expert level skill)

The next step is to indicate the expert-level skill for the requirement. 

An expert-level skill is a skill that the candidate has predominantly worked on. 

This is a skill that the candidate is not expected to merely possess or even have some good experience in, but rather a skill that the candidate must have primarily worked on or used. 

Example 1: A simple example is that of a “Java developer” where the candidate is expected to have predominantly worked on Java throughout their career. Hence “Java” would be their expert skill while “developer” would be their expected role.

Example 2: If a JD is specified asking for “MongoDB” (A nich database skill) do not set this as an “expert”. Almost always for a requirement like this, the expectation is for the candidate to have good experience in MongoDB, but not to have the primary skill to be MongoDB. You can make the skill mandatory, but not set as an expert. Note: Mandatory does not mean expert-level skill.

Tips to remember about expert skills:

1. You will almost never have more than 1 expert-level skill for a requirement.

2. Expert skill would usually be listed in the first line of the requirement. 

3. The skill would usually be a pervasive one within the skill domain (E.g., Java). The skill will usually not be a niche skill (E.g., Hyperion or ElasticSearch).

4. Almost always a skill specified within the role indicates an expert skill level. E.g., If your requirement comes in as a Java developer it means "Developer" is the role expected and "Java" is the expert level skill.

Step - 3 (Define the proficient level skill)

A proficient level skill is one that the candidate is supposed to have good knowledge in. 

Tips to remember about proficient level skills

1. All the skills that come with "X+ years of experience" are usually proficient skills and mandatory.

2. All the important skills beyond the expert-level skills are proficient skills. Make it mandatory only if specifically called out as mandatory by the hiring team.

3. Do not add more than 3-4 proficient-level skill groups as mandatory. That brings down the search volume considerably.

Eva searches for candidates who have worked on all the proficient-level and ranks them at the top. Within that group it looks for candidates who have worked the longest in the proficient-level skills or have highlighted it the most to be ranked at the top. 

Eva also considers the candidates with missing proficient-level skills as long as they are not marked as mandatory, albeit with lower ranking.

Step - 4 (Define the Nice to have skills)

These are skills that improve the probability of the candidate getting selected but are not a criterion that the client would reject the candidate for, during the shortlisting phase.

Quality of candidate ranking in search

The quality of ranking of candidates depends on multiple factors:

• How the resume was written. If the candidate does not represent the skills they have worked on well in the resume the ranking could be low.

• Skill packing - candidates attempt to pack up the resume with skill keyword. Crintell algorithm tries to recognize and de-prioritize such skills. But this could impact the ranking of the candidate.

• Format of the resume - In rare cases resume are un-parsable. This can severely impact the rating of the profile.

Quality metric

Eva search is continuously improved to provide a “shortlist” % of 30%+. This means for every 10 profiles shown to the recruiter at least 3 candidates must be good for consideration for a job. We continue to work and improve on this number.

Rating buckets to consider

• Any candidates above 80% can be considered an excellent fit.

• Any candidates between 60-80% would be considered a good fit

• Candidates below 60% till 40%+ can be considered if there are not many excellent and good fit candidates to pick from. This can happen if the skills expected are not found in the market or in the internal database.

• Profiles with less than 40% are the ones where most skills are missing or the parsing of the resume has failed for some reason.