Skip to main content
AI Ethics

Human-First AI, Built Responsibly.

Every algorithm on BaobabPact works for the candidate — not against them. We build AI that is transparent, explainable, bias-audited, and always subordinate to human judgement.

Quarterly Bias Audits
Explainable Scores
No Protected Characteristic Inference
Human Override

Our AI Ethics Principles

Six commitments that govern every AI system we build and deploy.

Transparency by Design

Every AI decision made on our platform is explainable. Candidates can always request a plain-language explanation of their match scores, skill gap analysis, and growth projections.

Bias Auditing

Our matching models are audited quarterly by independent reviewers for demographic bias across gender, ethnicity, age, and disability. Audit results are summarised in our annual Ethics Report.

Human Oversight

AI recommendations are advisory, not deterministic. Employers using our ATS pipeline see AI scores as one signal among many — not as hiring decisions. Final decisions always rest with humans.

Candidate Dignity

AI systems on BaobabPact are designed around the candidate's long-term wellbeing and career growth — not employer conversion metrics. We do not optimise for engagement at the expense of user autonomy.

Protected Characteristics

Our models are explicitly prohibited from using or inferring gender, ethnicity, age, disability, religion, or sexual orientation in any matching, scoring, or recommendation logic.

Responsible Innovation

Before deploying new AI features, we conduct internal ethics reviews, red-team exercises, and limited pilots. We do not deploy AI capabilities that we cannot explain and monitor.

How Our Matching Works

A transparent walkthrough of the signals and logic behind every match score.

01

Role Requirements Analysis

We extract structured skill requirements, experience levels, and role context from job listings using NLP. Salary ranges and growth signals are also parsed.

02

Candidate Profile Vectorisation

Candidate experience, skills, and stated preferences are transformed into a structured representation. We normalise across different ways of expressing the same skill or experience.

03

Multi-Signal Matching

Match scores are calculated across skills alignment (40%), experience depth (30%), growth trajectory (20%), and stated preferences (10%). No demographic signals are used.

04

Growth Score Modelling

Growth scores assess learning velocity, career progression rate, skill acquisition patterns, and role-specific development opportunities — not just current seniority.

05

Wellbeing Score Inference

Wellbeing scores are inferred from employer culture data, stated flexible working norms, historical employee reviews, and benefits transparency — not productivity metrics.

06

Human Review Gates

Automated AI shortlisting recommendations flag uncertainty levels. Cases with confidence below threshold are escalated for human review before candidate notifications.

Match Score Composition

Skills Alignment40%

Direct match between listed skills and role requirements

Experience Depth30%

Seniority match and domain experience alignment

Growth Trajectory20%

Career progression velocity and learning signals

Stated Preferences10%

Location, work style, salary, and explicit goals

No demographic characteristics are used or inferred in any scoring factor.

AI Limitations & Caveats

We believe radical honesty about AI limitations is more trustworthy than overclaiming capabilities.

AI is Advisory, Not Definitive

Match scores reflect signal alignment, not a guarantee of job success. Hiring involves complex human factors that AI cannot fully capture.

Data Reflects Historical Patterns

Training data reflects past hiring patterns, which may embed historical bias. We actively de-bias models, but imperfect correction is possible.

Models Are Not Static

AI models are retrained regularly. A score today may differ from a score in six months as models are updated. We version our models and maintain audit trails.

You Can Contest Any Score

If you believe an AI score is inaccurate or unfair, you have the right to request review. Contact us at ethics@baobabpact.com.

Our Public Commitments

  • We will never use AI to automate hiring rejections without human review
  • We will never sell AI-generated candidate assessments to third parties
  • We will publish an annual AI Ethics Report with bias audit results
  • We will provide plain-language explanations of all AI scores on request
  • We will not deploy generative AI in hiring-critical flows without oversight
  • We will consult with candidates and advocacy groups when updating AI models

AI Ethics FAQ

Contest a score or request a review?

Our AI Ethics team responds within 5 business days to all review requests.