Pile size as the metric
“Processed 500 resumes” is meaningless.
What matters is how many resumes the recruiter had to read by hand. A pipeline that processes 500 resumes but still requires reading 500 of them hasn’t solved anything.
AI screening at volume
AI screening that surfaces the right 80 candidates.
At 500 resumes per role, the bottleneck is human review, not parsing. ShortlistTable runs AI screening at batch scale with per-column confidence thresholds — auto-categorise the obvious cases, surface only the ambiguous ones to a focused review queue. Read the 80 that matter, not all 500. Zero auto-rejections, full audit trail.
Confidence thresholds · Focused review queue · Parallel batching
shortlisttable.app / volume-pipeline
Active roles · Volume Pipeline
- Needs reviewCustomer Support · 500 inbound112 shortlisted · 84 review · 304 hold
- Shortlist readyWarehouse Associate · 380 inbound210 shortlisted · 31 review
- ScreeningSDR · 240 inboundScreening in progress
Workspace summary
1,120
Resumes processed
322
Auto-shortlisted
115
In review queue
683
Hold queue
500+
Resumes per role per batch
Per-column
Confidence threshold
Focused
Review queue, not full pile
0
Auto-rejections
Where the real cost lives
Speed isn’t the bottleneck. Human attention is.
High-volume hiring tooling usually optimises for the wrong axis. Parsing is already fast; the actual cost is the recruiter’s attention. If a recruiter has to look at every one of 500 inbound resumes individually, you have a throughput problem the database speed cannot fix.
We close the loop by doing the per-criterion screening pass once per resume, then surfacing only the candidates with ambiguous verdicts. The recruiter reviews 80 — not 500. The remaining 420 sit in a hold queue with their evidence intact, ready for the next round or for compliance review.
Auto-reject creep
Hidden screen-outs scale badly.
At high volume, auto-rejection thresholds become a major bias amplifier — small model errors get applied to thousands of candidates silently. We never auto-reject.
Audit cost
High volume means high audit cost.
When a regulator asks why 300 candidates were screened out, you need a per-candidate paper trail. The trail has to be cheap to query, not buried in a notes field.
Volume-grade screening
How we cut review load without cutting candidates.
01 · Feature
Confidence threshold per column
Each screening column has its own confidence threshold. Auto-categorise the obvious cases, surface only the ambiguous ones to the review queue.
0.75
default per-column
threshold
02 · Feature
Focused review queue
The queue is ordered by ambiguity, not by score. Reviewers spend their time on the candidates the engine wasn’t sure about — not the obvious yeses or noes.
84
in review queue
(out of 500)
03 · Feature
Hold queue with evidence
Low-fit candidates land in a hold queue, not the bin. Evidence stays intact, the queue is searchable, and re-runs against new criteria are supported.
Hold queueSearchableRe-runnableAudit trail
04 · Feature
Parallel batch processing
500 resumes finish in minutes, not overnight. Per-row processing runs in parallel with backpressure so the worker doesn’t fall over.
Upload 500→Parallel run→Done in minutes
05 · Feature
Duplicate detection
Same candidate applying through two channels gets merged with a configurable matching rule. The merge is logged, never silent.
alice@email + alice.chen@email
Merged · 1 record
06 · Feature
Per-candidate audit trail
Every verdict — original, override, re-run — is logged with timestamp, actor, and reason. Queryable in bulk for compliance review.
AI verdict · 2026-05-25 09:14
Recruiter override · 2026-05-25 11:02
Re-run after column edit · 2026-05-26 08:30
How it compares
High-volume screening vs the alternatives.
| Property | ATS bulk import | Generic AI ranker | Spreadsheet at scale | ShortlistTable |
|---|---|---|---|---|
| Resumes per batch | ✓Thousands | ✓Thousands | ✕Drifts ≥ 80 | ✓Thousands |
| Per-column confidence threshold | ✕No | –Single global | ✕No | ✓Per column |
| No auto-reject | ✕Common | ✕Common | ✓Manual | ✓Hold queue only |
| Per-candidate audit trail | –ATS-shaped | ✕Limited | ✕None | ✓Timestamped per-cell |
| Duplicate detection | –Limited | ✕Rare | ✕Manual | ✓Configurable merge rule |
✓ Supported– Partial / manual✕ Not supported
FAQ
Common questions on high-volume screening.
Can the system handle thousands of resumes per role?+
Yes. The pipeline is built around batching — 5,000 resumes in a single table is supported. The constraint at that scale is usually the speed of the recruiter reviewing the surfaced 100–200, not the engine.
What about duplicate detection?+
Yes — duplicate resumes (same candidate applying via different channels) are merged with a configurable matching rule. The merge is shown in the audit trail, never silent.
Does the hold queue ever expire?+
Only by your retention policy. You can keep it open for the role’s lifetime, or set a 90-day auto-archive. We never delete candidate data on our own schedule.
What happens if I add a column halfway through a 500-resume run?+
Only the affected cells re-run, not the whole table. Previous cells stay valid and the audit trail records the column edit.
How do confidence thresholds get set?+
Per workspace default (0.75 by default), with per-column overrides. Most volume customers tune them up over time as they learn which columns the engine is reliably confident on.
Can I export only the review-queue subset?+
Yes — every export preset has a filter for any column. Export shortlist only, review-queue only, or hold queue only.
Volume without compromise
Read the right 80 candidates, not all 500.
Try ShortlistTable on a 25-resume pile to see the queue + threshold workflow. Free, no credit card.