The 'Sunday Scaries' index
Resignation isn't a moment. It's a slow Sunday-night dread that shows up in your data months before the email. Here's the signal — and how to read it without becoming creepy.
Most attrition prediction models are built on lagging indicators: engagement scores, exit interviews, promotion velocity. The most useful signal HR analysts have seen in five years is none of those. It's the slope of the Sunday-evening dread — what HR analysts call the Sunday Scaries Index — and it shows up cleanly 60 to 90 days before the resignation email.
- Track team-level aggregates only — never individual surveillance dashboards.
- Use voluntary signals (reactions, opt-in pulse questions) before involuntary ones (after-hours activity).
- Disclose the metric publicly in your people analytics policy. If you can't say it out loud, don't measure it.
- Pair every quantitative signal with a manager conversation — never with an automated nudge to the employee.
- Team-level aggregates.
- Voluntary, opt-in signals.
- Published methodology.
- Manager-led conversations.
- Individual dashboards available to anyone above the employee.
- Passive monitoring of keystrokes, mouse movement, camera presence.
- Hidden scoring nobody can audit.
- Automated 'wellness' nudges based on the score.
T.D. Borkovec's anticipatory anxiety research shows that humans rehearse stressful events for weeks before they happen — and the behavioral signature of that rehearsal is observable. The 'Sunday scaries' aren't a personality trait; they're a measurable spike in cortisol-related behaviors (poor sleep, late-night work, Sunday-night Slack pings, weekend calendar avoidance) starting roughly 72-90 days before a resignation. By the time the resignation is verbal, the system has been broadcasting it for two months.
Add prospect theory's reference-point dynamic (Kahneman & Tversky, 1979): an employee deciding to leave compares the future at the current job against an imagined alternative. The decision crystallizes when the reference shifts from 'staying' to 'leaving' as default. That shift is what the leading indicators capture — long before HR analytics catches the resignation itself.
A 1,500-person SaaS company, as one HR leader recounted, in 2024 had 18% regrettable attrition with no leading indicators. We instrumented 4 signals (anonymized, aggregated): Sunday-night message volume, optional-meeting acceptance rate, PTO clustering, 1:1 cancellation patterns. Managers got a private weekly 'team energy' read — no individual surveillance, just team-level slope. When the slope dipped, managers ran 1:1s with a retention frame. Twelve months later, regrettable attrition was 9%. They didn't catch the resignations. They prevented them.
- Sunday-night message volume per team. Trend it weekly.
- Optional-meeting acceptance rate per individual (anonymized to manager).
- PTO clustering — concentrated PTO in odd weeks is a sign of interview season.
- 1:1 cancellation patterns — repeated last-minute cancellations is a high-confidence signal.
- Voluntary feedback frequency — engaged employees give more, disengaged give less.
- Train managers to act on slope, not points. The first sign-of-life is week 70, not week 10.
- Aggregate only — never use these for individual surveillance. Trust dies faster than attrition.