We killed our weekly reliability meeting and replaced it with one that has a fixed agenda, a hard 30-minute limit, and produces decisions, not updates.

Our old reliability review was 45 minutes of teams reading their dashboards out loud. Nobody left with a decision made. We rebuilt it around three questions instead.

Three questions, every time

Which services burned more than 25% of their error budget this week? What toil showed up more than twice? What runbook or alert got used and turned out to be wrong? Every other topic gets redirected to a different meeting or a written doc. This constraint is what makes the meeting short and the same length every week.

from slo_burn
| where week_start == startofweek(now())
| summarize budget_consumed_pct = max(cumulative_burn_pct) by service
| where budget_consumed_pct > 25
| order by budget_consumed_pct desc

Every item leaves with an owner and a date

A finding without an owner is a complaint. We stopped accepting "someone should look at this" as an outcome — every item raised gets a named owner and a date before the meeting ends, tracked in the same doc we review the following week. Items that roll over three weeks unaddressed get escalated automatically to the engineering managers, not discussed again in the same meeting.

Rotate who presents

We rotate which team presents their burn-rate and toil numbers each week rather than having one reliability lead present everyone's data. This forced every team to actually look at their own numbers before the meeting instead of outsourcing that thinking, and it surfaced problems teams hadn't noticed in their own services.

Kill the meeting if there's nothing to discuss

Some weeks all three questions come back empty — no budget breach, no repeat toil, no wrong runbook. We cancel the meeting outright in that case instead of filling 30 minutes anyway, and send a one-line async note confirming the quiet week. Treating a cancelled meeting as a legitimate, positive outcome — not a missed obligation — was what finally convinced skeptical teams the review wasn't just theater.

  • Fix the agenda to three questions: budget burn, repeat toil, wrong runbooks/alerts.
  • Cap the meeting at 30 minutes and enforce it.
  • Every finding leaves with a named owner and a date, not a vague action item.
  • Rotate presenting teams so ownership of the numbers doesn't concentrate in one role.
  • Cancel the meeting outright on genuinely quiet weeks instead of padding the agenda.