We finally put a real number on what a humane on-call rotation costs versus a bare-minimum one. The gap was smaller than finance expected, and the return was bigger.

Every humane on-call change we'd made — stipends, recovery windows, bigger rotation pools, a dedicated runbook-maintenance hour — had been approved piecemeal, each time as its own small ask. Nobody, including us, had ever added up the total cost against the total return. So we did.

The full cost, added up honestly

Stipends, the extra headcount needed to hit our six-person minimum rotation size, and the paid weekly maintenance hour came to roughly 2.3% of total reliability-team payroll cost annually. That's the real number — not hidden, not rounded down to sound better.

sustainable_on_call_cost_model (annualized, reliability org)
  on_call_stipends: 0.9% of team payroll
  headcount_to_hit_min_rotation_size(6): 1.1% of team payroll
  paid_runbook_maintenance_hour (weekly, all rotations): 0.3% of team payroll
  total_direct_cost: ~2.3% of team payroll

  offsetting_return (measured, same period):
    voluntary_attrition_on_reliability_team: down from 22% to 9% annually
    avg_incident_MTTR: down 18% (attributed partly to lower fatigue,
      partly to fresher runbooks)
    recruiting_cost_avoided (fewer backfills needed): ~1.4% of team
      payroll, roughly offsetting the direct cost on its own
"Finance asked what humane on-call costs. The more useful question turned out to be what un-humane on-call costs, and we'd just never been forced to add that number up before."

Who actually pays for it

The direct financial cost is the company's, and it's modest. But there's a cost sustainable on-call doesn't erase, and we're honest about that too: the recovery windows and maintenance hours mean slightly less raw engineering-hours available for feature work, a tradeoff product leadership accepted explicitly rather than having it happen invisibly through burnout and attrition later.

Why we're publishing the number

Every time we've made a humane on-call change piecemeal, it competed against feature work for budget attention and looked expensive in isolation. Presented as a single 2.3%-of-payroll line item against a measured drop in attrition and MTTR, it was an easy approval. We'd recommend any team trying to fund this kind of change do the same full accounting, rather than asking for each piece separately.

  • Add up the full cost of sustainable on-call as one number — piecemeal asks look expensive; a total against measured returns usually doesn't.
  • Track attrition and MTTR as the offsetting return, not just morale survey scores, when making the financial case.
  • Be explicit that the tradeoff is real feature-work capacity, not free — hiding that cost just delays the real conversation.