Identify Temp Agency Red Flags Before You Sign

Most firms use their Applicant Tracking System, like Bullhorn or JobDiva, for basic tracking. But the 'notes' field is a black hole of unstructured text. You cannot run a report to find which agency consistently submits candidates who pass the second interview, because that data is buried in free-text recruiter comments.

A 15-person tech recruiting firm wanted to partner with a new agency specializing in contract developers. The agency sent over 200 anonymized resumes as PDFs and a spreadsheet of 50 placements from the last quarter. The spreadsheet had inconsistent job titles ('Sr. Dev', 'Software Eng. IV') and was missing interview-to-placement ratios. It took two senior recruiters a full day to manually standardize the data and guess at the agency's real performance.

This approach is fundamentally broken. Manual review is subjective and misses patterns. One recruiter might favor candidates with specific certifications, while another prioritizes years of experience. Without a systematic way to parse resumes and track performance metrics across all partners, you are choosing partners based on gut feel, not data.

Syntora would begin with a comprehensive discovery phase to audit your current vetting process, identify critical red flags, and map out available data sources, including ATS APIs like Bullhorn REST API or SFTP servers for spreadsheet exports. This phase defines the scope of data ingestion, normalization, and the key performance indicators (KPIs) relevant for agency evaluation.

For data ingestion and normalization, Syntora would design a Python pipeline leveraging Pandas. This pipeline would connect to source systems, ingest data, and apply standardization logic. This includes unifying job titles using a predefined taxonomy and mapping inconsistent placement statuses (e.g., 'Filled', 'Closed-Won') to a single 'Placed' state, creating a clean dataset for benchmarking.

To process unstructured data such as candidate resumes and recruiter notes, the Claude API would be employed. Syntora would configure the API to extract specific features like years of experience with particular skills (e.g., Python, AWS), education, and past job tenures. This structured data would then be stored in a Supabase Postgres database, ready for analysis.

The core logic for processing new agency data and generating reports would be deployed as a FastAPI service on AWS Lambda. When new data dumps are uploaded by your team, the Lambda function would be triggered, processing the files, extracting features, and generating a detailed PDF report. This report would benchmark the new agency's performance against your historical data on agreed-upon metrics such as submittal-to-interview ratio and average time-to-fill.

A simple front-end application built on Vercel would provide your team with an interface to upload files and view generated reports. All processing logs would be written to AWS CloudWatch with structured logging via `structlog`. For robustness, the system would include error handling and alerting, notifying a dedicated Slack channel if a file fails to parse or an API call times out after retries, allowing for immediate investigation.

A typical engagement for developing and deploying a system of this complexity, including discovery, development, and production readiness, spans approximately 10-14 weeks. The client's team would need to provide access to relevant internal systems and collaborate closely on defining the specific metrics and red flags to be monitored. Deliverables would include the deployed data pipeline, a user interface, comprehensive documentation, and knowledge transfer to client personnel.