customer-support-ops

Customer Support Operations (triage, SLAs, quality, staffing, deflection)

Triage: severity × impact, decided by rule not by mood

Every inbound ticket gets a priority from a written matrix — severity (broken for the customer?) × impact (how many customers / how much revenue?) — plus routing to the queue that can actually resolve it. Triage done by whoever reads it first, on vibes, means priority tracks how angry the customer sounded.

P1  service down / data loss / security — page someone, all-hands until mitigated
P2  core function broken, no workaround — same-business-day engagement
P3  degraded or has a workaround — normal queue
P4  question / feature request — normal queue, feeds the KB and product backlog

Escalation paths are named in advance: which tier owns what, what evidence must accompany a hand-off (repro steps, account, logs), and who is paged for P1. An escalation that restarts diagnosis from zero doubles resolution time.

SLA design and the metrics that matter

first response time (FRT)  = ticket created → first human (not auto-ack) response
resolution time            = created → resolved, clock paused while pending-customer (say so)
SLA attainment             = tickets meeting target / tickets in period — per priority tier
first contact resolution   = resolved with no reopen and no additional agent contact / resolved
reopen rate                = reopened within N days / resolved  (the lie detector for "resolved")
backlog age                = distribution of open ticket ages, not just the count

Set targets per priority, from measured capability plus headroom — an SLA copied from a competitor's marketing page is a promise chosen at random. Report percentiles (e.g. 90th) rather than averages: an average FRT of 2h coexists comfortably with a tail of customers waiting three days.

Quality: CSAT, CES, NPS — and QA scorecards

• CSAT (post-resolution, per ticket) measures the interaction. Report response rate alongside — 20% response CSAT is a self-selected sample. CES (effort) predicts loyalty better for support interactions. NPS measures relationship, moves slowly, and is mostly not a support metric; don't hold agents to it.
• QA scorecard: sample tickets per agent per week, scored against written criteria (accuracy, completeness, tone, process adherence) with calibration sessions so two reviewers give the same ticket the same score. Uncalibrated QA is a popularity contest.
• Never target reopen-able metrics in isolation: pushing FRT alone produces fast useless replies; pushing handle time produces premature "resolved". Pair every speed metric with FCR/reopen and CSAT.