Hires from a pool of 100000
Here’s an uncomfortable detail from the 2026 Hiring Benchmarks Report from Starred. Nearly 90% of candidate survey invitations in 2025 went to rejected candidates, and about 70% of those rejections happened at the application stage. For most applicants, the whole candidate experience is a form, a wait, and an email. Often with no human anywhere in it.
That’s the norm now. Criteria’s 2026-2027 benchmark report found 56% of hiring professionals use AI somewhere in their process, and screening is the most common use. Meanwhile global candidate NPS (Net Promoter Score) has barely moved from +17, while hiring manager NPS sits at +73. If your metrics were built for a world where a recruiter called everyone back, they’re measuring the wrong thing. Below: the classics worth keeping, the one that’s quietly misleading you, and five new metrics to add when software makes the first cut.
Candidate experience metrics are the numbers that show how applicants experience your hiring process, from first click to final decision. The core set is candidate NPS, application completion rate, time to first response, time to hire, withdrawal rate, and offer acceptance rate. All of them still matter. None is enough alone.
| Metric | How to calculate it | What it tells you |
| Candidate NPS | % promoters (9-10) minus % detractors (0-6) | Overall sentiment from one question |
| Application completion rate | Completed applications ÷ started applications × 100 | How much friction your apply flow adds |
| Time to first response | Application submitted to first meaningful reply | Whether candidates feel ignored |
| Time to hire | Application date to offer acceptance, averaged | Where momentum dies |
| Withdrawal rate | Candidates who withdrew ÷ candidates who entered the stage × 100 | Where people pick someone else |
| Offer acceptance rate | Offers accepted ÷ offers extended × 100 | Whether the experience held up to the end |
If completion rate is where you leak, our guide to finding and fixing candidate drop-off shows how to diagnose it stage by stage. For the survey basics, see our earlier piece on how to evaluate candidate experience. This one covers what changes when software does the first cut.
Because most of the people answering it were rejected. Starred’s data shows nearly 90% of survey invitations go to rejected candidates, whose average cNPS is -7, while hired candidates score overwhelmingly positive. Your blended score reflects who you asked more than how you did.
Two companies can run identical processes and report completely different cNPS just because one rejects earlier or surveys later. So stop reporting one number. Split it three ways: by outcome (hired, rejected, withdrawn), by stage, and by source. The source split alone is revealing. Starred found candidates approached by a recruiter scored +43, while those from job posting sites scored +20.
Here’s how that plays out, using made-up numbers. Say 200 candidates answer your survey: 20 hired (cNPS +60), 160 rejected (-5), 20 withdrawn (0). Your blended score is +2, which looks fine. Next quarter you tighten the application-stage filter. Now 10 hired, 180 rejected, 10 withdrawn, and every segment score is exactly the same. The blended number drops to -1.5. Nothing got worse for any candidate. The mix changed, and the dashboard called it a crisis.
Also track response rate by stage. Candidates are answering surveys less at nearly every stage, and a score built on a thin sample can’t be a metric. If you’re still designing the survey itself, our piece on why candidate experience surveys matter covers the basics. Send at least two versions: one after the application, one after the final decision.
Add metrics that measure clarity, not just speed: an expectation-setting score, rejection clarity rate, time to first human touch, application-stage rejection share, and automation sentiment. Starred found expectation setting drove more dissatisfaction than anything else, appearing in 37.8% of candidate comments, 74.5% of them negative. Vagueness hurts more than rejection.
| New metric | How to calculate it | Why it earns a slot |
| Expectation-setting score | Average rating on one survey item: “I knew what to expect at each stage” | Biggest driver of dissatisfaction |
| Rejection clarity rate | Rejected candidates who got a reason or the criteria used ÷ all rejected × 100 | Shows whether early rejections come with context |
| Time to first human touch | Application to first message from a named person | An auto-reply is not a always response candidates prefer |
| Application-stage rejection share | Rejected at application ÷ all rejected × 100 | Shows how much of your experience software decides. |
| Automation sentiment | Open-text comments that mention AI or automation, split positive and negative | 56% of AI-related comments were negative |
These five are our suggestions, not industry standards, so define them once and keep the definition stable. Trust is the backdrop. Gartner’s survey of 2,918 job candidates found only 26% trust AI to evaluate them fairly, even though 52% believe it is screening their application. Starred saw the same split from the other side: candidates who mention AI after being hired are mostly positive, those who mention it after a rejection mostly negative. Same software, different story, and the difference is usually whether anyone explained what it was doing.
Pair every sentiment metric with a conversion metric. cNPS tells you how candidates feel. Stage drop-off, time in stage, and offer acceptance tell you what they did.
If hiring managers rate the process +73 and candidates rate it +17, you’ve optimized the inside of the process. Part of closing it is consistency, because interviews are inconsistent across hiring managers and candidates feel that as unclear expectations. Part is speed, because slow processes lose happy candidates; our deep dive on time-to-hire and candidate ghosting covers that one. And part is judging outcomes. In 82% of interviews with top-performing teams, quality of hire came up as the most important result, so close the loop with a quality of hire questionnaire at 90 days.
Start with three survey touchpoints and one spreadsheet. You don’t need a platform to get the first quarter of useful data, just consistent definitions and a habit of tagging every response with the outcome, stage, and source. Here is a setup most teams can finish in a week:
1. Send a two-question survey at three moments: after the application, after screening or interviews, and after the final decision. Ask the 0-10 recommend question plus “I knew what to expect at each stage.”
2. Tag every response with outcome (hired, rejected, withdrawn), stage, and candidate source before it goes anywhere near a dashboard.
3. Pull completion rate, withdrawal rate, and time in stage from your ATS, stage by stage, instead of as one funnel average.
4. Add a rejection-reason field and a sent-or-not flag, so rejection clarity rate becomes a count instead of a guess.
5. Build one monthly page: cNPS by outcome, completion rate, time to first human touch, rejection clarity rate, and response rate. If it doesn’t fit on one page, nobody will read it.
Savos isn’t a survey tool and won’t replace your survey platform, if you have one. What Savos changes is what sits behind the numbers. It gives candidates a guided, structured application that feeds one shared candidate profile, so nobody is asked to repeat themselves at the next stage. ScaleScreen asks context-aware questions instead of one template for everyone. TalentLens keeps evaluation criteria visible and every signal traceable, so when a candidate asks why, there’s an actual answer on record. That record is the raw material for rejection clarity and expectation setting, the two metrics that matter most once software makes the first cut. Our buyer’s framework for candidate screening software and our look at what’s actually new in AI recruiting software cover how to evaluate tools on this.
What are the most important candidate experience metrics?
Start with candidate NPS split by outcome, application completion rate, time to first response, withdrawal rate, and offer acceptance rate. If software handles early screening, add an expectation-setting score and rejection clarity rate.
What is a good candidate NPS?
The global average is at +17 in its 2026 benchmark, with rejected candidates averaging -7 and recruiter-approached candidates at +43. Compare yourself against your own mix of outcomes and sources, not one headline number.
How often should you measure candidate experience?
Continuously, triggered by stage: after the application, after screening or interviews, and after the final decision. Review trends monthly and change the process quarterly, so one bad week doesn’t drive decisions.
Does automated screening hurt candidate experience?
Not by itself. Missing transparency about how decisions are made is the source of the friction. Candidates who understand what’s being assessed and when they’ll hear back tend to read the same software far more kindly.
