Purpose

This article explains how auto-calibration works within the Eightfold platform, ensuring the availability of a talent pool for every position.

Introduction

Auto-calibration happens by parsing the information available in the position. It continuously updates the calibration for open positions without requiring any intervention from users (recruiters, hiring managers and others.)

Auto-Calibration using positions

• Fetching of similar positions: auto-calibration pulls in positions based on their similarity. It checks the position titles to calculate these similarities.
• Enhancing existing positions calibration : Once it gets a set of similar positions, it identifies the best data for the following fields - titles, skills, ideal candidates, and minimum years of experience.
• Auto-calibration of positions happens periodically. It is refreshed once in 30 days (if a position is auto-calibrated at Day one, then it will be considered for auto-calibration again after 30 days).
• If a position is manually edited (any changes in the position calibration page) by the user, then it is not auto-calibrated again (to ensure auto-calibration doesn't overwrite any of the data entered by the user).
• While auto-calibrating a position, only one role from JIE is used (most similar to the position).
• For any position, the auto-calibration response includes
  • Calibration Titles: maximum = 8
  • Calibration Skills: maximum = 20
  • Ideal Candidates: maximum = 10
  • Minimum years of experience

While auto-calibrating a position, many fields are used to help with a final calibration. A few of them are:

• Similar positions (based on title match).
• Job description text without boiler-plate (length of text should be greater than 100).
• JIE Role - both implicit & explicit (may or may not be available for the position).
  • If available, then the role calibration data (skills, titles or more) is appended to the rest of the data (generated by JD similar positions) and a reordering is done on the basis of the similarity with position title and JD.

Following are the fields considered when recommending the role for a position.

• Ideal Candidates Match
• Job Code Match
• Seniority Match
• Title Match
• Skill Match

Role based Auto-Calibration of positions

• Auto-calibration pulls in one relevant role and few similar positions based on similarity
• From this set, it identifies the most relevant fields enhances the position calibration

JIE role for a position

• For a position, the top role from the Calibration Assistant role recommendations is used as a probable candidate for auto-calibration.
• A similarity score is calculated between the position and the top recommended role (based on a combination of calibration titles match, skills match, ideal_candidates match, and seniority match). If that score is above a threshold value, then only the role is used in auto-calibration. The default threshold is set to 0.6 which means that the position and role should be at least 60 percent similar for the role to be used in that position’s auto-calibration.

Calibration Assistant role recommendations logic

While showing the role recommendations on a position’s calibration page, several fields are considered and are extracted from the current position in context. These fields are then used to match against the role space (all JIE roles) of the group_id to rank the roles in order of similarity against the position.

The following position fields are considered for role recommendations:

• Position title
• Job description
• Job code of the position

The first two are used to generate a set of skills and titles.

Data points used in the query are:

1. Titles (position_title and similar titles extracted using the models)
2. Skills (skills_based_on_JD and skills_based_on_position_title)
3. Associated Role_ID of the position (if available)

After all the data points are fetched from a position, a search is performed on all the roles present in the group_id. Associated Role_ID matching is given the highest boost, followed by title matching, followed by skills matching. Boosting the matching parameters takes care of the ordering of search results.

The top 20 search results are then populated. These results are then filtered to remove any thin roles (if the role is not well defined, missing skills, titles ). The remaining roles are populated as role recommendations in calibration assistant.

How the associated role is fetched for a position

A role can have an association with a position (or an employee) based on certain conditions.

There are two types of role associations supported in JIE:

1. Implicit association: Also known as tag-based association with the role (position’s/employee’s tag values match with the role’s tag values). This association information is calculated at runtime.
   1. Tags that are to be considered for the association purpose are defined in the role_library_config at ‘integration_config.configuration_tags’ (mostly JobCode).
   2. If there are multiple configuration tags present (Such as title, job family, business unit ), then the corresponding tag values of all the tags, roles and the positions/employees should match in order to be implicitly associated.
2. Explicit association: When the user explicitly associates a position/employee with a role manually on the position/employee side. Explicit association works as an override of implicit association.

While fetching the associated role for a position, the explicitly associated role is returned. If not available, then the implicitly associated role is returned (may or may not be available).

JIE Role in Position auto-calibration

While auto-calibrating a position,

• If the position has an associated role, do a full copy of calibration from the associated role to the position. Do not execute the other parts of the auto-calibration algorithm i.e. using similar positions, job description and more.
• If the position does not have an associated role, continue to run other parts of the auto-calibration algorithm i.e. using similar positions, job description etc.

A config is introduced to enable this full copy flow:

• Set copy_calibration_from_associated_roles field to true in *calibration_config *for the target group_id.

Positions auto-calibrated after enabling the config are considered for the full copy.

Language-specific position auto-calibration

• Auto calibration works with various languages. While our language models are trained on multiple languages, some have limited data, which may affect the quality in those languages.
• LLM-based auto-calibration supports English only as of now. A role or position calibrated in a non-English language is not considered.
• If you use the ‘full copy of JIE Role in Position auto-calibration’ feature, then you will be able to copy the language-specific calibrations from roles to positions, assuming that the roles are calibrated in the desired language.