SLA metrics
Median first-response, median resolution, percent within SLA, breaches, and warnings. All measured against your office-hours window.
SLA metrics
Ochre's response and resolution KPIs live on /analytics. They are intentionally simple: two numbers, both medians, both honest about outliers.
There is no separate SLA-policy engine, no per-policy target, no "percent within" calculation. Ochre exposes the raw response and resolution times. Treat them as the operational heartbeat — set your own internal targets against them.
The two numbers
Median first response
The middle value of every conversation's time-to-first-reply in the period. Median, not mean, because one bad weekend should not move your headline number.
First response is measured from the customer's first message to the first non-internal reply that goes back out. AI auto-sends count. Human replies count. Internal notes do not.
The tile shows the count of conversations that have replied (n=X replied). If n is small, the median is volatile — read the bigger volume before drawing conclusions.
Median resolution
The middle value of every conversation's time from creation to status='closed', across conversations that closed in the period.
Snoozed time is included. If a conversation snoozed for 4 hours and then closed, those 4 hours are part of the resolution time.
The tile shows the count of resolved conversations in the period (n=X closed).
Why medians, not means
A single 3-week-old conversation that finally closes today doesn't drag the median. It would drag a mean. Medians give you the "what does a typical conversation look like" answer, which is what most teams actually want.
If you do care about long-tail conversations (the ones languishing in the queue), look at the inbox directly with a stale filter — that's a better tool than a percentile chart for finding the specific tickets that need attention.
How to read them together
The two numbers move differently:
- Median first response goes up but resolution stays stable. Fewer agents on shift, or a load spike. First-touch is suffering but the team is closing what they pick up.
- Median resolution goes up but first response stays stable. Sticky tickets — engineering escalations, refund cases, multi-touch back-and-forth. Look at top tags.
- Both go up. Volume spike or staffing problem. Cross-check the volume chart.
- Both go down. Either you got faster, or a routing change is closing things sooner. Cross-check CSAT — if scores stayed flat, you got faster; if scores dropped, you got more aggressive.
Period selector
Same 7d / 14d / 30d / 90d picker as the rest of /analytics. Median resolution especially benefits from a longer window — over 7 days, a small number of long tickets dominates. Over 90 days, the picture stabilizes.
What we don't do
A few things teams sometimes ask about that Ochre doesn't ship today:
- No first-class SLA-policy engine. No per-segment "1 business hour, 99% within."
- No business-hours pause built into the response timer. The medians are wall-clock.
- No breach / warning escalations.
If you need policy-level SLA tracking, export raw conversation timestamps from /analytics and compute against your own targets in a spreadsheet or warehouse.
Common questions
Why is median first response so low?
If the AI is set up to auto-send on first touch, most conversations get a reply within seconds. That's not a measurement glitch — that's the AI doing what you asked it to do. The deflection KPI is the right place to read how often this happens.
Why is a conversation in the median when the customer was happy?
Median first response measures time, not satisfaction. A 12-hour first response can earn a 5-star CSAT. Pair the response medians with the CSAT card to weigh outcomes.
Can the AI hit the response number?
Yes. Replies from autopilot count as first response. Many teams see medians drop dramatically once autopilot is on for first-touch.
Related
Was this article helpful?