If you’ve ever opened a support tool’s analytics page and felt instantly overwhelmed, you’re not alone. Modern support platforms ship with 15-20 metrics out of the box. Almost none of them matter for a team of 1–20 people. Some are actively misleading at small scale.
Here’s the short list of metrics worth tracking weekly, the ones that look smart but mislead, and the dashboards small teams should ignore until they grow into them.
Five metrics worth tracking weekly
1. Median first-reply time
How long the typical visitor waits between sending their first message and getting any response. Track the median, not the average — averages get destroyed by one outlier where someone forgot to reply for 4 hours.
Why it matters: the strongest correlate of CSAT in our data is whether the first reply lands inside 60 seconds. Past 5 minutes, the conversation might as well be email.
Good target: under 60 seconds, ideally under 30 with AI on the front line. (How to get there →)
Tracking cadence: weekly. Look at the trend, not the absolute number.
2. Resolution rate at first contact
What percentage of conversations close without needing more than one round of back-and-forth. A customer asks; you (or the AI) answer; they’re satisfied; conversation closes.
Why it matters: high first-contact resolution is the cleanest signal that your team is solving problems efficiently. Low first-contact resolution either means questions are too complex for one reply (sometimes legit) or your team is being unclear and triggering follow-ups (usually the actual cause).
Good target: 50–70% for a normal SaaS / DTC mix. Higher for AI-heavy support. Lower for B2B with longer issues.
Tracking cadence: weekly.
3. Customer Satisfaction (CSAT)
Single thumbs-up / thumbs-down after a conversation closes. Sample every conversation, not a random 5%.
Why it matters: the metric closest to “is the customer actually happy?” Every other metric is a proxy.
Good target: 4.5+ out of 5 on a 5-point scale, or 90%+ thumbs-up on a binary. Below 80% is a real problem; above 95% means you might be sampling too lightly.
Tracking cadence: weekly trend + spot-check the low-scoring ones individually. The qualitative review of bad CSAT scores is where you actually learn.
4. Conversation volume per channel
Just count: how many conversations came in on chat vs. email vs. WhatsApp vs. Instagram, day over day, week over week. Don’t normalize.
Why it matters: trend changes here are early signals of every interesting thing happening in your business. A 30% chat volume spike means either growth, a bug, or a viral moment. You want to notice it on day one, not from analytics two weeks later.
Good target: depends entirely on your business. Look at the change, not the absolute number.
Tracking cadence: daily (literally just a quick eyeball; no need to obsess).
5. Abandoned chat rate
What percentage of chats end with the visitor not replying to your last message. (More on dropouts →)
Why it matters: this is the metric that captures invisible quality problems. AI loops, over-long replies, wrong answers — all show up here before they show up in CSAT (because customers who dropped out don’t bother to give a thumbs-down).
Good target: 15–25%. Above 35% means structural problem (probably AI is running visitors in circles); below 10% is suspicious (probably your team is over-engaging visitors who were already done).
Tracking cadence: weekly.
Three metrics that look smart but mislead
1. Average response time
Not median — average. The reason this is misleading is that one forgotten Friday-evening chat with a 14-hour response time will absolutely destroy your average. The team panics, “optimizes,” and finds it’s just the one outlier. Use the median; or if you must look at the average, also look at p90.
2. Average handle time
How long it takes to “finish” a conversation. The premise is “shorter is better.” For small teams it’s the opposite — pushing for shorter handle time trains agents to under-answer, which raises follow-up rate, which makes total customer time longer.
Average handle time becomes useful when you’re at 30+ agents with strict capacity planning. Below that scale, optimizing it actively hurts.
3. Tickets per agent per day
A vanity metric that incentivizes the wrong behavior. The agent who handles 80 conversations a day by giving cursory replies is “better” than the agent who handles 30 with thorough ones. CSAT and resolution rate matter much more than volume per head.
If you’re a 5-person team using this metric to evaluate performance, stop. Use CSAT and first-contact resolution. Different agents will naturally specialize; that’s fine.
The dashboards to ignore until you’re past 20 agents
Most support tools default to showing you:
- Hourly volume heatmaps
- Agent-by-agent performance leaderboards
- Conversation-tag distribution pie charts
- Channel-by-channel CSAT breakdown
- Average time-in-status across the funnel
These all look like they’re telling you something. They mostly aren’t, at small scale, because the sample sizes are too small to be meaningful and the moves you’d make from the data don’t pay back.
Resist the urge to make these dashboards. They are mostly there to make the support tool look powerful. The 5 metrics above + a weekly scan are enough.
What to actually do with the metrics
The weekly review
15 minutes, every Monday. Look at:
- First-reply time trend over the past 4 weeks. Improving, stable, or degrading?
- CSAT trend. Same question.
- Volume by channel — anything spike?
- Abandoned chat rate — any change?
- Spot-check 3 low-CSAT conversations from the prior week. What went wrong?
That’s it. The decisions that come out:
- If first-reply degrades → tighten the rotation; consider AI tweaks.
- If CSAT drops → look at the low-score conversations qualitatively.
- If abandoned rate climbs → check whether the AI’s running in circles.
- If volume spikes → check whether something broke or growth is real.
Most weeks: nothing interesting happens. The metrics are stable; the team is fine. The 15 minutes is insurance against the week where something does move.
What to do when a metric moves
The single rule: the cause is almost never random. If first-reply time doubled this week, something concrete happened. Maybe two people were on holiday, maybe the AI’s confidence threshold got bumped, maybe a feature shipped that broke a flow.
Find the cause before you “fix” the metric. Most teams overcorrect — they see a number move and immediately staff up or tighten rules. Half the time, the cause is something specific and small; identifying it is cheaper than restructuring.
When to add more metrics
The list above scales to about 20 agents. After that, you genuinely benefit from:
- Tickets-per-channel CSAT split — to identify which channel’s struggling.
- First-touch resolution by tag — to identify which question types need better articles.
- Average response time by agent — once you have enough volume per agent for the average to mean something.
These are real signals at scale. They’re noise at small scale.
The 1-line summary
Track median first-reply time, first-contact resolution rate, CSAT, channel volume, and abandoned chat rate — weekly. Skim everything else. Spend the saved time writing better help articles.
The teams that drown in support metrics aren’t being more rigorous; they’re being more anxious. The teams that drive support outcomes pick a handful of metrics, look at them often enough to notice trends, and spend the rest of their analytics time on the qualitative work — reading the actual bad conversations and figuring out what to change.
