> ## Documentation Index > Fetch the complete documentation index at: https://docs.keloa.ai/llms.txt > Use this file to discover all available pages before exploring further. # Response time > First-response and resolution medians, percentiles, and where the slow queue actually lives. How fast you reply is most customers' single biggest signal. Keloa tracks two clocks per conversation — **first response** and **resolution** — and surfaces both as percentile distributions in **Analytics → Response time**. ## The two clocks | Clock | Starts | Stops | | ------------------ | ------------------------------ | --------------------------------- | | **First response** | Customer's first message lands | First reply (AI or human) is sent | | **Resolution** | Customer's first message lands | Conversation is resolved | Both clocks pause when the SLA policy says to (out-of-hours, waiting on customer). See [SLA timers](/inbox/sla) for the pause rules. ## Percentile cards Top of the page: * **First response — median (P50)** with a delta vs the previous window. * **First response — 90th percentile (P90)** — the slowest 10% of replies. * **Resolution — median (P50)**. Median tells you what most customers experience. P90 tells you the long-tail story — when P50 is good but P90 is awful, your bad days are catastrophic. Both numbers matter. ## Distributions A histogram per clock with bucketed durations. The shape is informative: * **Tight peak around the median** — consistent service. * **Long right tail** — a small set of very slow tickets are dragging the average. Drill into them. * **Bimodal (two peaks)** — typically AI vs human; the AI replies in seconds, human replies after the queue. Good signal that AI deflection is working. ## Breakdowns Slice each chart by: * **Channel** — webchat is usually fastest; email is slowest. If your email median is over 4 hours during business hours, a queue ritual is missing. * **Assignee** — workload distribution. Wide variance is a sign of poor routing or one teammate carrying a backlog. * **Tag** — *"refunds always take 3 days"* is a process problem, not a queue problem. * **AI vs human** — quantifies how much the AI buys you. Most teams see AI medians in seconds and human medians in tens of minutes. * **Hour of day / day of week** — when peaks land. A heatmap chart underneath shows volume by hour-of-week so you can spot the shift you're under-staffed on. ## Hour-of-week heatmap A 7-by-24 grid (Mon-Sun × 0-23h) coloured by conversation volume. Use it to plan staffing — if Tuesdays at 14:00 are your hottest hour, that's the shift to cover first. Tap a cell to filter the page to that hour. ## SLA breach overlay When you have an [SLA policy](/inbox/sla) configured (Business+), each percentile chart shows your target as a vertical line. P90 to the right of the SLA line means your top 10% are breaching — a clear signal to either staff up or relax the policy. ## Common interpretations * **First response P50 \< 5 min, P90 > 1 hour** — most customers see fast replies; a handful are stuck. Look at unassigned + low-priority queues. * **P50 jumps sharply between weeks** — usually a staffing change. Cross-reference the agent breakdown. * **Resolution P50 high but first response P50 low** — fast acknowledgements but slow follow-through. Often a knowledge or process gap, not a speed gap. ## Plan availability Headline percentile cards and channel breakdown are available on every plan. The hour-of-week heatmap and SLA overlay are **Business+** (paywall key `advanced_analytics.use`). ## Related Set the targets the lines on these charts compare against. Get conversations to the right person on the first try.