Company-wide engagement sits at 72/100 — within the healthy range. The aggregate masks a department-level problem.

Engineering attrition is running at 24%, 3x the company average of 8%. Attrition began climbing in September — the same month the return-to-office mandate took effect. Every other department stayed within ±2 points on engagement. Engineering dropped 15 points.

The engineers leaving are not the newest hires. The spike is concentrated in the 2–4 year cohort — your highest-output, highest-cost-to-replace talent. Exit surveys name flexibility as the top reason in 68% of departures. The recruitment funnel converts at 1.4% application-to-hire. Backfill is expensive and slow.

The September attrition spike is specific to Engineering.

Who left: Engineers with 2–4 years of tenure — not juniors adjusting to office norms, not seniors with established routines. The mid-tenure cohort. They joined during a remote-first hiring period, built their work patterns around flexibility, and received a policy change they hadn’t agreed to.

Exit survey evidence: 68% of engineering departures cited “flexibility / remote work policy” as the primary reason. The next-highest category was “compensation” at 18% — a distant second. Flexibility leads by a 4:1 ratio.

Why other departments didn’t react the same way: Sales, Operations, Marketing, and Support were already office-based or role-constrained. The RTO policy didn’t change their working patterns. For Engineering, it changed everything.

The 15-point drop in Engineering is not a culture problem. It’s a contract problem.

Remote-first engineers vs. office-native engineers: Segment by pre-hire working arrangement and the split is clear. Engineers hired as remote-first dropped 22 points on engagement (78 → 56). Engineers who were already office-based dropped 4 points (76 → 72) — within normal survey noise.

The policy did not uniformly affect the engineering organisation. It disproportionately affected the segment that accepted an offer under remote-first terms. That group now accounts for 74% of Engineering’s voluntary departures.

What the survey verbatims say: The top open-text themes from Engineering Q4 surveys: (1) “I feel like the goalposts moved,” (2) “I’m less productive in the office than I was at home,” (3) “I don’t understand why this applies to my role.” These are not morale issues. They are fairness and trust issues. Morale programs won’t fix them.

I found the U-shape in Engineering’s tenure curve. Engineers at 2–4 years hit the engagement floor — scores of 47–55. Past year 4, engagement climbs back to 62 and above. Every other department tracks upward the whole way. Your Engineering curve looks like it self-corrects. It doesn’t.

The recovery past year 4 is a selection effect. The engineers who would have left at 2–4 years already did. The ones remaining are a filtered population — people who adapted, or who lacked better options. Engagement rises because the dissatisfied population departed, not because conditions improved. You’re not retaining your best mid-tenure engineers. You’re retaining the ones who didn’t leave.

The institutional knowledge loss is concentrated in the 2–4 year band. These are engineers with enough codebase context to be effective and enough market value to have options. Every departure in this cohort costs 2–4 years of system familiarity and stakeholder trust that a new hire cannot replicate. The scatter plot isn’t showing you a retention success story past year 4. It’s showing you where the damage already happened.

At 24% annualised attrition, Engineering needs 28 net-new hires per year to hold headcount flat (115 engineers × 24% = ~28 departures). At 1.4% application-to-hire conversion, that requires 2,000+ applications annually.

That is not a sustainable backfill strategy. It is an expensive one: engineering recruiting costs average $28,000 per hire (agency fees, recruiter time, interview overhead, onboarding). Replacing 28 engineers costs ~$784,000/year before productivity ramp.

The asymmetry: Retaining one experienced engineer at 2–4 years of tenure is worth more than hiring two junior replacements. Retention wins by a wide margin.

Three decisions worth making now, ordered by impact:

1. Introduce a structured flexibility policy for Engineering — not company-wide, not one-size-fits-all. Segment by role. Engineers whose output is asynchronous (individual contributors, senior engineers) get remote-first status with opt-in office time. Roles with collaboration dependencies (tech leads, PMs) get hybrid with clear norms. This directly addresses the 68% exit-survey signal without reversing the RTO policy wholesale.

2. Build a stay-interview programme for the 2–4 year engineering cohort — the people at highest attrition risk based on the tenure-engagement scatter. A 30-minute structured conversation with their manager, twice per year, surfaces flight risk before it becomes a departure. HR Business Partners run the framework; managers run the conversations. Cost: low. Signal: high.

3. Reframe the engineering talent value proposition — exit surveys show compensation is the second-most-cited reason (18%), but it’s a distant second to flexibility. Raising salaries without addressing flexibility will not move the attrition rate. If you’re considering a comp review, pair it with the flexibility policy — separately, neither solves the problem.

If the RTO policy stays: At 24% annualised attrition, Engineering loses ~14 more engineers in the next six months. At 1.4% conversion and current application volume, the funnel delivers ~6 hires in the same period. Net headcount change: −8 engineers. By August, the team is 7% smaller than today.

If a hybrid compromise is offered: Pre-RTO data shows engineering attrition at 6–8%. Peer companies with hybrid policies report similar rates. A return to ~8% attrition means ~5 departures in six months. The funnel still delivers ~6 hires. Net headcount change: +1 engineer. Stable.

The cost of inaction: Each departing mid-tenure engineer takes 2–4 years of institutional knowledge — codebase familiarity, system design decisions, stakeholder relationships. Replacement ramp time averages 6 months. In the gap, team velocity drops. Projects slip. The remaining engineers absorb more load, which accelerates the next wave of departures. At current rates, the compounding effect is a 15–20% velocity reduction by Q3 2026 — before any hiring can close the gap.