What this automation does
The coordinator stops being a human switchboard between candidates and interviewer calendars.
- Trigger on stage change. When a recruiter moves a candidate to an interview stage in the ATS, a webhook starts the workflow with the candidate, the job and the interview plan.
- Resolve the panel. The interview plan says which interview types are needed, how long each runs, and which pool of trained interviewers can cover each one.
- Find valid slots. The workflow queries free/busy for every eligible interviewer, applies working hours in each interviewer’s time zone, buffers, a daily interview cap per person, and the required order of sessions.
- Offer options. The candidate gets an email with three to five complete itineraries in their own time zone, each one bookable in a click.
- Book everything. On confirmation, the workflow re-checks availability, creates one event per interviewer with the video link and scorecard link, and sends the candidate a single confirmation with the agenda.
- Handle replies. If the candidate replies in free text instead of clicking, a model classifies the reply as confirm, reschedule, cancel or question, and extracts any new availability.
- Remind and close out. Reminders go to candidate and interviewers the day before, and the ATS activity feed is updated with what was booked.
When it is worth building
It pays off when one or more coordinators spend most of their day scheduling, when panels have three or more interviewers, or when time-to-schedule is losing you candidates to faster employers. A rough test: if the median gap between “move to onsite” and “onsite booked” is more than two business days, scheduling is a bottleneck.
It does not pay off for one-to-one screens with a single recruiter. A booking link handles that for a small subscription. It also does not pay off if your ATS scheduling add-on already covers your panel rules. Ask your vendor before you spend on a build.
Architecture
| Layer | What we use | Why |
|---|---|---|
| Trigger | ATS webhook from Greenhouse or Lever, through their REST APIs | Starts the moment a recruiter acts, with the interview plan attached |
| Orchestration | n8n | Built-in Google Calendar, Outlook, Gmail and Zoom nodes, plus HTTP Request for the ATS |
| Availability | Google Calendar free/busy query or Microsoft Graph getSchedule | Returns busy blocks without exposing the contents of interviewers’ calendars |
| Slot engine | A code step with explicit rules | Time zones, buffers, ordering and caps are exact math, not a job for a model |
| Reply parsing | A small, fast model such as Claude Haiku 5.5 with a strict JSON schema | Cheap and quick for short emails, and only extracts intent and constraints |
| Candidate page | A lightweight hosted page or signed email links | Lets candidates pick an itinerary without logging into anything |
| Video | Zoom API, Microsoft Teams online meetings, or Google Meet | Unique link per interview, created with the event |
How to build it
- Write down the interview plans. For each role, list sessions, lengths, order constraints and the interviewer pool for each session. This is usually the hardest step, because it lives in coordinators’ heads.
- Set interviewer rules. Capture working hours, time zone, maximum interviews per day and per week, and blackout periods. Store them in a table interviewers can update.
- Connect calendars with least privilege. In Google Workspace, use free/busy scope where possible. In Microsoft 365, scope the app’s calendar access to the recruiting group’s mailboxes with RBAC for Applications in Exchange Online, which replaces the older application access policies.
- Build the slot engine. Generate candidate itineraries, score them by fewest days, least interviewer load and candidate time-of-day preference, and keep the top few.
- Send options and hold slots. Email the itineraries, place short tentative holds, and release them after a set time if the candidate has not chosen.
- Book atomically. On confirmation, re-check free/busy, then create all events. If any event fails, roll back the others and offer new options.
- Add reply parsing. Route candidate replies through the model. Confident reschedule requests regenerate options; anything else goes to the coordinator with a summary.
The downloadable starter template below handles the stage-change trigger, Google Calendar free/busy lookup, option email and booking for single-session interviews, which you can extend to panels.
What breaks in production
- Time zone shorthand. A candidate writes “3pm EST” in July and means Eastern Daylight Time. Normalize abbreviations to a region, such as America/New_York, and show times back to the candidate for confirmation.
- Calendars that lie. Interviewers leave events as “free” when they are not, or block whole days with “focus time” they are happy to move. Agree on a convention and let interviewers mark protected blocks explicitly.
- Race conditions. Two candidates pick the same slot within a minute. The re-check at booking time and tentative holds prevent most of it.
- Interviewer declines after booking. Watch event responses. When an interviewer declines, find a substitute from the same pool before bothering the candidate.
- Replies landing in the wrong inbox. Candidates reply to a recruiter’s personal address. Use a shared scheduling address as the sender and reply-to.
- Room resources. For onsite interviews, rooms are calendars too, and they have their own booking rules and approvals.
Timeline and cost
Most teams are live in two to three weeks. Lumeor builds this in the range of $4,000 to $10,000, with the low end covering single-session scheduling on one calendar system and the high end covering multi-session panels, interviewer load balancing and reschedule handling. Running costs are mostly workflow hosting; the reply-parsing model costs very little at typical interview volumes. If you are weighing platforms for this, see n8n vs Zapier. The multi-step, multi-calendar logic tends to fit n8n better.
How to measure success
- Business hours from stage change to confirmed interview
- Coordinator hours per week spent on scheduling
- Reschedule rate, and the share of reschedules handled with no human involvement
- Candidate no-show rate
- Interviews per interviewer per week, to confirm load is spread fairly
- Share of candidate replies the model routes to a human, which should fall as you refine the parser
Scheduling sits between AI-assisted resume screening and the offer. The same calendar plumbing is useful later for onboarding meetings once the hire is made.