Introduction

The AI-enabled skills proficiency inference feature is a core component of Eightfold's talent intelligence platform. This article outlines the feature's scope, technical implementation, and application for key stakeholders.

Note: This feature is available only to limited customers.

Feature Summary

This feature utilizes a predictive model to automatically assign a proficiency score to any skill on an Employee's Career Hub profile. The score is used to categorize the user's expertise into the organization's proficiency scale (e.g. 5 point scale, 3 point scale).

How Skill Proficiency Is Calculated?

The system computes skill proficiency using a machine learning model optimized for low-latency, on-demand prediction.

Model and Output

• Model: An XGBoost model (XGB model) utilizes multiple employee profile elements to predict a skill's proficiency.
  
• Raw Output: The model's direct output is a probability score between 0 and 1.
  
• Proficiency Levels: The system linearly maps this score into the organisation's proficiency scale levels.
  
  • Note: Because the model provides a 0-1 probability, the user experience (UX) can be configured to any point scale, such as a 4-point scale.
    
• Training Data: The model has been trained on profile–skill pair data across various roles and industries, using both manual and LLM-based annotations.

Model Features

The key features used in the model establish the relationship between the skill (the skill being measured) and the employee’s professional profile. The modeling outcomes have shown that these features provide much value to proficiency prediction.

The key features used in the model establish the relationship between the skill (the skill being measured) and the employee’s professional profile. The modeling outcomes have shown that these features provide much value to proficiency prediction.

How are the Inferred Skill Proficiencies displayed in Career Hub?

The inferred skill proficiencies are displayed on the Skills Assessments page in Career Hub. In the absence of self or manager assessments, the Inferred proficiency will be displayed up front on the skills slider bar. If a self or manager assessment is present for the skill, then that takes precedence and is displayed on the slider bar. However, on hovering on the slider, both the inferred and self / manager assessed proficiencies can be viewed.

While assessing the skills, the inferred proficiencies can be referenced simultaneously using the ‘Show inferred assessment’ toggle.

How do the Inferred Skill Proficiencies power downstream recommendations?

If self or manager-assessed proficiencies are unavailable, inferred proficiencies are used to personalize course recommendations.

For example, if the skill ‘Strategy’ lacks self or manager assessment but has an inferred proficiency of 2 out of 5, the system prioritizes courses with a difficulty level above 2 to support skill development.

How to report on the Inferred Skill Proficiencies?

Go to Insights -> Data Hub, and select the profile_skill_assessment table, and get the data.

The inferred_proficiency field in the table displays the inferred proficiency value for each skill across all employees.

FAQs

• Does the inference work if the proficiency scale is customized to a 4-point scale?
  The model gives a probability form 0-1. The UX can be modeled to any point scale.

• Do we support explainability on the inference (list the sources at the least)?
  No, we don't.

• Is employee ‘Certification’ considered for inferring proficiencies?
  No, Certification data has been sparse and not considered

• What level of accuracy is expected in ‌proficiency inference?
  As the model’s outcome depends on factors such as the richness and recency of an employee’s profile, and the nature of the skill (generic or customer-specific), the inferred proficiencies should be treated as indicative - serving as a guideline to encourage employees to review and update their skill information.

• What are the benefits of inferred skill proficiencies without Roles design or JIE?
  Inferred proficiencies operate independently of roles designed. Without role design, the “Required by Role” skill set won't be available, and only the skills listed in the employee’s profile will be considered. The model will still infer proficiency levels for these profile skills, enabling personalized course recommendations based on the inferred proficiencies.

• What are the benefits of inferred skill proficiencies without role-skills benchmarking?
  Currently, inferred proficiencies aren't directly used in the skill gaps calculation on the Skills Assessments page, which determines whether a skill is categorized as a strength, skilled, or opportunity. However, inferred proficiencies are directly factored into course recommendations - courses with difficulty levels above the inferred proficiency are recommended. As a result, the absence of role-skills benchmarks doesn't affect course recommendations driven by inferred proficiencies.

• Can we decide that some factors should be omitted when inferred skills are in use, e.g. education? No. The model is trained to automatically identify the profile features that contribute most to accurate proficiency inference. Consequently, it isn't possible to manually modify or customize which factors the algorithm considers.

• Can the model take into consideration Self, Manager, and Peer assessments to infer proficiency? The model is intentionally designed to be independent of manual assessments. This ensures that inferred proficiencies are unbiased and are based solely on profile data such as experience, education, and skills. Self, manager, or peer assessments can be used separately to validate or complement the inferred proficiencies, but they aren't inputs to the model itself.