AI Integration with Lever

AI integrates with Lever through its REST API and webhook system, acting as an intelligent orchestration layer that reads from and writes to key objects: candidates, opportunities, interviews, feedback, and notes. The primary integration surfaces are:

• Candidate Pipeline: Inject AI scoring into opportunity custom fields, trigger stage transitions via POST /opportunities/{id}/stage, and automate outreach using POST /messages.
• Interview Coordination: Use webhooks for interview.created and interview.updated to trigger AI agents that schedule panels, generate briefing docs, and synthesize feedback submissions.
• Data Enrichment: On candidate.created, call AI services to parse resumes, extract skills into Lever tags, and suggest relevant requisitions for matching.

A production implementation typically uses a queue-based architecture to handle webhook bursts and ensure idempotency. For example, a candidate.application webhook triggers an AI screening job that:

1. Fetches the resume via GET /candidates/{id}/resumes.
2. Runs parsing and scoring against the job's requisition requirements.
3. Updates the opportunity with a match score in a custom field and, if a threshold is met, automatically advances the stage. All actions are logged with POST /audit_logs for governance. This turns manual resume review from a hours-long task into a minutes-long automated triage, allowing recruiters to focus on high-potential candidates.

Rollout should be phased, starting with a single requisition type (e.g., Software Engineer) to validate scoring logic and recruiter adoption. Governance is critical: implement a human-in-the-loop approval for any stage transition that results in a rejection, and use Lever's permission sets (POST /permissions) to control which users can view or modify AI-generated data. For teams managing compliance, all AI prompts and model decisions should be stored in an external audit trail, linked back to the Lever opportunity ID. This controlled approach de-risks the integration while delivering immediate productivity gains in screening and scheduling.

A production integration with Lever is built on its webhooks and REST API, treating the ATS as the system of record. The core pattern is event-driven: Lever fires a webhook for key candidate stage changes (e.g., candidate.stage_change), note.created, or opportunity.created. This event payload is queued, triggering an AI agent to perform a specific task—like screening a new application, generating interview questions, or summarizing panel feedback. The agent acts on the data, then writes results back to Lever via the API, typically updating custom fields on the candidate or opportunity, adding a note, or triggering a stage transition. This keeps all data and audit trails inside Lever.

For example, an AI screening agent listens for the application.received webhook. It fetches the full candidate profile and resume via GET /v1/candidates/{id} and GET /v1/candidates/{id}/resumes. Using a configured skills model, it parses the resume, scores match against the job requisition, and writes a structured assessment to a custom field like ai_screening_score. It can also post a note summarizing key qualifications. For interview coordination, an agent listening for interview.scheduled can fetch panelist details, generate role-specific questions from the job description, and post them as a private note for the hiring manager, all within seconds.

Rollout is phased, starting with a single, high-volume workflow like automated resume screening for a specific department. Governance is critical: all AI-generated notes are tagged with a source (e.g., [AI Agent]), and scores are presented as recommendations, not decisions. A human-in-the-loop approval can be configured for stage promotions based on AI scores. Implementation includes robust error handling for API rate limits, idempotent processing to avoid duplicate actions, and comprehensive logging of all AI inputs and outputs to an external audit trail for model performance review and compliance.

Secure API Integration & Data Handling

• All AI interactions with Lever occur via its secure REST API and webhooks, using OAuth 2.0 for authentication and role-based access controls (RBAC) to respect existing user permissions.
• Candidate PII (resumes, notes, contact info) is never stored in external vector databases without explicit encryption and data residency controls. We recommend processing data in-memory or within your VPC, using Lever's candidate, application, and opportunity objects as the source of truth.
• Audit trails are maintained by logging all AI-generated actions (e.g., score updates, note additions, stage transitions) back to Lever's activity feed or a custom audit object, linking to the triggering user and model version for full traceability.

Phased Rollout for Risk Mitigation Start with low-risk, high-ROI workflows before expanding AI's decision influence:

1. Phase 1: Augmentation – Deploy AI for summary and drafting tasks. Examples: auto-summarizing interview feedback into Lever notes, generating first-draft outreach messages, or extracting skills from resumes into custom fields. All outputs require human review before commit.
2. Phase 2: Recommendation – Introduce AI suggestions that recruiters can accept/reject. Examples: candidate ranking scores on a 1-10 scale added to a score_ai field, suggested interview questions based on job requisition, or talent pool rediscovery alerts.
3. Phase 3: Controlled Automation – Automate specific, rule-based tasks with human-in-the-loop gates. Examples: auto-advancing candidates who pass a score threshold, triggering background checks upon offer acceptance, or sending scheduled status updates—all with configurable approval steps and exception queues.

Model Governance & Compliance

• Use separate, fine-tuned models for different tasks (screening vs. communication) to contain drift and allow targeted updates.
• Implement regular bias testing on screening models using Lever's demographic data (where legally collected) to detect and correct for unintended skew.
• For regulated industries, maintain a clear separation between AI-generated insights and final hiring decisions, ensuring human accountability is preserved in Lever's approval workflows. Rollback plans should include the ability to disable specific AI agents without disrupting core Lever operations.