Post a decent role and a few hundred applications turn up. A recruiter working that inbox by hand does the same four things three hundred times: open, skim, decide, maybe reply.
The “maybe” is where it falls apart. Strong applicants wait nine days for a first response and take another offer. Borderline ones hear nothing ever. And the recruiter spends the week on scheduling email instead of talking to people, which is the one part of the job that can’t be delegated.
Greenhouse’s 2022 Candidate Experience Report found more than 75% of job seekers had been ghosted after an interview. Not after applying. After turning up and talking to someone.
The jobs@ blueprint
Acknowledge everything, immediately, from a template. We have your application, here’s the process, here’s roughly when you’ll hear back. No AI on this step — it should be byte-identical for every candidate, and the only interesting decision is whether the timeline you promise is one you’ll keep.
Turn the application into fields. Current role, relevant years, location, the specific skills the ad asked for, portfolio links. Screening has to run on structured data. It cannot run on somebody skim-reading PDFs at seven in the evening on day four of a search, because that’s where the bad decisions come from.
Screen against what you published. The requirements in the job ad, and nothing else — not what the hiring manager also secretly wants, not a pattern the model noticed. What comes out is an assessment with reasons attached rather than a number with no working shown.
Branch, and give a person two of the three branches.
- Clear match: draft a reply offering two or three concrete times, queued for the recruiter to approve and send.
- Borderline: into a review queue with the assessment attached, so a human decides with context rather than starting cold.
- Clear miss against the published criteria: a kind decline, within days rather than months — and a recruiter still approves the batch before anything goes.
Nothing in there asks software to make a judgement call about a person on its own. That’s not an accident and it’s not caution for its own sake; it’s the design.
Scheduling
Scheduling is the other half of the time sink — three emails over two days to agree an hour. An agent can offer the times you’ve made available and confirm which one they picked, which collapses most of that. It won’t reach into your calendar and book it; there’s no calendar integration, so the last step is a booking link or a person.
Getting the screening right
If you automate one thing carefully in your whole company, make it this one. Screening is where automation can do real harm to real people, and where the harm is hardest to see from the inside.
Screen only against written, published criteria. If a requirement isn’t written where the hiring team can read it, it isn’t a requirement. This sounds bureaucratic and it’s actually the single most useful discipline in the list — most unfair screening comes from criteria nobody ever wrote down.
A person owns every rejection. Software never declines a strong or borderline candidate by itself. The workflow gathers the evidence and drafts the message; a recruiter reads the assessment, makes the call, and puts their name on it.
Keep out the inputs that stand in for protected characteristics. Graduation year signals age. Address and commute distance signal race and class. Employment gaps signal caregiving, illness and disability. None of these need to name a protected trait to act as one. The test that catches most of it: could you comfortably explain this requirement to the candidate it excluded?
Log the decisions. Which criteria each candidate met, what the assessment said, who approved the outcome, when. If a candidate asks why, you can answer. If a regulator asks, you can show your process was consistent rather than asserting it.
That last one isn’t hypothetical any more. New York City’s Local Law 144 requires bias audits and candidate notice for automated employment decision tools, and the EU’s AI Act classifies hiring systems as high-risk with human oversight attached. “The software decided” is not something you want to say to a regulator, and it’s not something you want to say to a candidate either.
The part that pays off later
The obvious return here is recruiter hours. The one that compounds is that everybody gets an answer.
The strong candidate hears back the same day with times. The borderline one gets a human decision within the week. The person who was never a fit gets a straight no instead of six weeks of silence and a slowly forming opinion about your company.
Candidates talk — to each other, in reviews, and inside whichever company they join instead. This year’s decline is next year’s strong hire, or a referral, or a customer. Silence gets you none of those, and it’s the cheapest thing in this entire post to fix.
The condensed version, with the workflow laid out, is on the recruiting email automation page.