What makes Lever easy or hard to automate
Lever exposes two APIs. The Postings API is a public read of your open jobs plus an apply endpoint, which careers sites use. The Data API is the one that matters for automation: opportunities, candidates, stages, interviews, feedback, offers, notes, users and postings, all as JSON over REST. Authentication for in-house work is an API key sent with HTTP Basic auth, so any platform that can make an HTTP call can connect.
The easy part is reading and reacting. Lever’s webhooks cover the moments recruiting teams care about, such as stage changes and hires, so a workflow can start the second something happens instead of polling.
The hard part is that every Lever account is configured differently. Stages, archive reasons, tags and feedback forms are custom, and Lever’s “opportunity” model (one person can have several applications) trips up anyone who assumes one candidate equals one record. Map your account’s objects before writing workflow logic.
The AI layer
Lever is the system of record for your pipeline. The AI work sits next to it:
- Reading resumes and writing summaries that point to evidence (“five years running AP for a multi-entity company”) rather than producing a score.
- Writing for people, such as job description drafts, candidate update emails and the weekly hiring note for managers.
- Turning interview notes into structured feedback that the interviewer reviews and submits themselves.
Hiring is a regulated decision. Keep the model in a support role, log what it produced, and review outcomes by stage for unexplained gaps. EEOC guidance applies to AI used in hiring, and NYC Local Law 144 requires bias audits and candidate notices for automated employment decision tools used for New York City roles. This is not legal advice; involve counsel before you change how decisions are made.
Limits to plan around
- Super Admin dependency. Keys and integrations can only be created by Super Admins. If that person leaves, find out which keys they own before access is removed.
- Rate limits. Lever’s docs describe a default per-key limit of 10 requests per second with bursts up to 20, and note these defaults may change. Backfills of thousands of opportunities need pagination and backoff.
- Custom configuration. Stage IDs, tags and archive reasons differ per account. Store them as configuration, not code.
- Confidential roles. Access to confidential data is a separate permission that can only be granted at key creation. Decide deliberately whether an AI workflow should ever see it.
How Lumeor typically connects it
We usually run n8n with a Basic auth credential holding a Lever API key, a webhook receiver that verifies Lever’s signature, an LLM step for summaries or drafts, and a Slack or email step for the human review. Summaries are written back as Lever notes so recruiters see them where they already work. The responsible resume screening guide covers the review pattern, and the interview scheduling guide covers calendar handling. If you also run Greenhouse in another business unit, the same pattern applies with different object names.