Handling applicants
Applicant handling splits cleanly. The correspondence — acknowledging, answering questions about the role and the process, chasing what somebody said they would send — is ordinary inbox work Connect does well. The assessment is not: no capability recorded here scores, ranks or shortlists a person, and this page does not pretend one exists.
The part that is ordinary inbox work#
- Acknowledging an application so nobody waits in silence, grounded in what Knowledge says about the role and the process.
- Answering the questions that repeat — where the job is based, what the stages are, when people hear back — where those are written down as facts.
- Reading what arrives. CVs come as PDF and DOCX, spreadsheets as XLSX or CSV, and all are read with the standard library; DOCTYPE and ENTITY declarations and zip bombs are refused.
- Chasing. A commitment — references, a portfolio, availability — becomes a follow-up with a reason and a due time, on the
taskchannel where only a person can do it. - Keeping one record per human. A candidate who writes from two addresses is two identities until somebody merges them, and merging is a human decision.
The part that must stay with a person#
| Activity | Position | Why |
|---|---|---|
| Acknowledge and inform | Connect, under the channel autonomy rule | Correspondence, grounded in stated facts |
| Summarise what a document says | Connect, into the record | Reading is not judging |
| Rank or shortlist applicants | Not available, and not to be improvised | No such capability is recorded, and an unrecorded one is an invented one |
| Decide an outcome | A person, always | The decision is the job |
| Communicate a rejection | A person decides; the message can be drafted | ask_before_send exists precisely for messages of this weight |
| Explain why somebody was rejected | A person | An explanation Connect composed would describe a decision it did not make |
Fairness, in the parts Connect does touch#
Three structural behaviours help, and they are worth naming because they are not settings anybody has to remember.
- Grounded answers only. An answer Knowledge cannot support is refused rather than improvised, so two candidates asking the same question get answers from the same source.
- Everything is recorded, including refusals. The audit trail carries what was decided, under which rule, and what happened. A refusal is a decision and is logged as one.
- Contact rules are absolute. Suppression, unsubscribe, complaints and do-not-contact are checked in one place before any outreach, and the Assistant cannot clear a do-not-contact entry — including for somebody who asked not to be approached again.
What none of that gives you is a fairness guarantee about your own process. Connect records; it does not audit your hiring.
What is not here#
- An applicant tracking system
- Nothing in these sources documents an ATS integration. Candidates live as people with lifecycle stages, follow-ups and cases — workable at modest volume, and not a substitute for a hiring system at scale.
- Interview scheduling against a calendar
- No calendar integration is documented. The demo lifecycle and the dated follow-up queue are what exist.
- Text-message reminders
- Outbound SMS is provider-dependent and the live carrier carries none.
- A recorded screening call
- Recording is foundation and is not enabled on the live carrier. A phone screen leaves a transcript and a summary, and the realtime path replies at a 3.3 s median on the best measured call — a pace candidates notice.
- A retention policy for applicant data
- Not documented in these sources. Decide it deliberately rather than inheriting it by accident, and read the security section before assuming a figure.
What it costs#
Hiring is spiky: nothing for months, then several hundred applications in a fortnight. That pattern meets the daily allowance, which holds work rather than dropping it and leaves the read cursor where it was, so a flood defers rather than disappears. Reading attachments is part of the same work; phone screening is the line item that would change the shape of the bill, because audio is metered at several times the rate of text.
Questions#
Can Connect shortlist candidates for us?
No. No scoring or ranking capability for people is recorded, and inventing one out of a tool built to qualify commercial prospects would be wrong twice over — technically and ethically. Connect handles the correspondence and keeps the record; the shortlist is yours.
Can it send rejection emails?
It can draft them, and with the channel on ask_before_send a person approves each one, with the approval recorded against them. Running rejections autonomously is a decision to be taken deliberately, not one to arrive at because a channel-wide setting happened to be on.
Does it read CVs?
It reads the documents: PDF through its text layer or by the model when there is none, DOCX, XLSX, CSV, images and plain text, with malformed documents and zip bombs refused. It summarises what a document says into the record. It does not evaluate the person the document describes.