Reduce Hiring Bias with a Custom AI Screening System
Yes, AI can reduce hiring bias by screening candidates based on job-relevant skills, not demographic data. It standardizes evaluations, removing subjective judgments that introduce bias in initial resume reviews.
Syntora helps small companies reduce unconscious bias in their hiring process by developing custom AI screening systems. These systems identify job-relevant skills from resumes and standardize evaluations, enabling more objective candidate assessments.
A custom system is not an off-the-shelf product. Implementing a bias-aware screening system requires access to your Applicant Tracking System (ATS) data and a minimum of 12 months of hiring history for specific roles. This allows a tailored model to learn what skills predict success from your own past decisions, rather than relying on a generic industry template.
What Problem Does This Solve?
Most small companies rely on their ATS for initial screening, but tools like Greenhouse and Lever offer primitive keyword matching, not true AI. This approach is easily biased. A filter for 'MIT' or 'Stanford' might screen out qualified candidates from state universities. Relying on years of experience can discriminate against younger applicants with equivalent skills.
A 30-person software firm we worked with was hiring a senior engineer. Their recruiter set up a filter in their ATS to find candidates with 10+ years of experience who worked at FAANG companies. This simple rule-based filter immediately disqualified 80% of their applicant pool, including several high-potential candidates from successful startups and open-source contributors who were ultimately better fits. Manual review is no better; studies show reviewers spend just 7 seconds on a resume, relying on pattern-matching that reflects their own biases.
These built-in ATS features create a false sense of objectivity. They automate a flawed manual process, encoding existing biases into the workflow. Because the logic is simple keywords and boolean flags, it cannot understand context, skill equivalency, or a candidate's trajectory. This leads to a homogenous pipeline and missed opportunities.
How Would Syntora Approach This?
Syntora's approach to reducing bias in hiring would begin with a discovery phase. We would audit your existing hiring process and data within your Applicant Tracking System (ATS) to identify key data points and potential areas for bias. We would then work with your team to define specific, objective criteria for candidate evaluation and establish a project scope.
The technical implementation would involve connecting to your ATS API to pull historical applicant data, including resumes and hiring outcomes. We have built document processing pipelines using Claude API for financial documents, and the same pattern applies to parsing unstructured text from resumes and cover letters. The Claude API would parse this text into a standardized JSON object, extracting features such as technical skills, certifications, and years of experience with specific tools. This creates a clean dataset for model training.
Using Python and scikit-learn, we would train a gradient boosting model on this structured data. The model would learn the statistical relationship between skills on a resume and your historical hiring decisions (e.g., advanced to interview, hired). Crucially, we would explicitly exclude demographic proxies like names, locations, and university prestige from the feature set, focusing the model purely on demonstrated abilities and experience.
The trained model would be packaged as a lightweight API using FastAPI and deployed to a serverless function on AWS Lambda. When a new candidate applies through your ATS, a webhook would trigger the function. The API would ingest the resume, generate a score, and write that score plus a short summary of the candidate's top skills back into a custom field within your ATS.
The delivered system would provide a ranked shortlist, not an automated decision, ensuring a human recruiter still makes the final choice. We would also build a simple dashboard using Streamlit that tracks key fairness metrics, such as the pass-through rate for different demographic groups, to monitor model performance. Typical build timelines for this complexity range from 6 to 10 weeks, depending on data availability and client integration needs. Your team would need to provide access to ATS data, clearly defined hiring criteria, and ongoing feedback during the training and validation phases.
What Are the Key Benefits?
A Shortlist in Minutes, Not Days
The system processes 200 new applications and produces a ranked, scored list in under 10 minutes. Your recruiters start their day with a qualified pipeline, not a cluttered inbox.
No Per-Seat Subscription Fees
This is a one-time build engagement. After launch, you only pay for minimal AWS Lambda hosting costs, typically under $50 per month, regardless of how many recruiters use the system.
You Own the Code and the Model
We deliver the full Python source code and trained model files in your private GitHub repository. You have complete control and ownership of the system we build for you.
Proactive Fairness Monitoring
We configure CloudWatch alerts that notify you in Slack if fairness metrics, like demographic parity between scored groups, deviate by more than a 15% threshold.
Integrates Into Your Current ATS
The system writes scores directly into custom fields in Greenhouse, Lever, or Ashby. Your team's workflow doesn't change; they just get better data where they already work.
What Does the Process Look Like?
Week 1: ATS Data Audit
You grant read-only API access to your ATS. We analyze historical applicant data for at least one target role and deliver a Data Quality Report outlining model feasibility.
Week 2: Model Development
We build the resume parsing pipeline, train the scoring model on your historical data, and validate its performance. You receive a Model Validation Report with key accuracy metrics.
Week 3: Deployment & Integration
We deploy the scoring API on AWS Lambda, configure ATS webhooks, and run end-to-end tests. You get a live system scoring new applicants as they arrive.
Week 4+: Monitoring & Handoff
We monitor the first 200 live candidates and provide a detailed System Runbook. After a 30-day monitoring period, we transfer full ownership and control of the infrastructure.
Frequently Asked Questions
- How much does a custom AI screening system cost?
- Pricing is based on scope. Key factors include the number of distinct roles to model, the quality of your historical ATS data, and the specific ATS platform. A single-role model with clean data from Greenhouse is a straightforward 3-week build. We provide a fixed-price quote after the initial discovery call and data audit.
- What happens if the AI system fails or makes an error?
- The API is wrapped in error-handling logic. If the resume parsing or scoring fails, the candidate is automatically flagged in the ATS for immediate manual review. This ensures no applicant is ever lost due to a technical glitch. The system is designed for decision support, complementing human recruiters, not replacing them entirely.
- How is this different from Textio or other bias reduction software?
- Tools like Textio help you write more inclusive job descriptions to attract a diverse applicant pool. Our system works at the next stage: evaluating the candidates who apply. We screen and rank applicants based on skills. They are complementary; Textio widens the top of the funnel, and we help you fairly assess the talent within it.
- Where is our candidate data stored and processed?
- The entire system is built and deployed within your own AWS cloud account. Candidate data is pulled from your ATS, processed by the Lambda function, and written back to the ATS. No personally identifiable information ever passes through Syntora's servers or is stored by any third-party SaaS. You maintain full data residency and control.
- How do you prevent the AI from learning our existing biases?
- We train the model on your past hiring decisions but intentionally exclude features that are known proxies for bias. By removing data like names, schools, and zip codes, the model is forced to find patterns based only on skills, tools, and quantifiable experience. We also measure and report on fairness metrics during and after the build.
- What if we don't have enough historical hiring data?
- A model needs a clear success signal to learn from. We require a minimum of 50 historical applicants for a single role, with at least 10 who were hired or advanced to a final round. If you don't have this data yet, we would advise against a build and suggest focusing on structured data collection in your ATS first.
Ready to Automate Your Professional Services Operations?
Book a call to discuss how we can implement ai automation for your professional services business.
Book a Call