For chat support, first-response time (FRT) — how long the visitor waits before any human or bot says anything — predicts satisfaction better than any other metric. Visitors will tolerate a slow resolution if the first reply was fast, and they'll churn during a long initial silence even if you eventually solve the problem. The goal: get FRT down without making the team miserable.
Step 1: instrument honestly
Most teams measure FRT wrong. They count from "visitor sent first message" to "agent sent first message" — but they forget that a pattern auto-reply doesn't count for the visitor. They feel acknowledged but they're still waiting for the human.
Step 2: deflect the obvious
Twelve patterns answer 60-80% of questions instantly. We have a separate playbook on which patterns to ship — the gist: pricing, refunds, hours, integrations, GDPR, login. Every one of those has a definitive answer that doesn't need a human.
Step 3: route by site, not by who's online
If you have multiple sites in one workspace, route conversations by site. The marketing site's questions go to the growth team; the developer docs site goes to the engineer-on-call; the careers site goes to recruiting. Don't route by "whoever has fewest open conversations" — it leads to the senior engineer answering "how do I reset my password" because they happened to be free.
Step 4: shift-coverage, not 24/7 expectation
Pick a coverage window and post it on the widget. "Mon-Fri, 9am-7pm CET". Outside that window, the auto-reply is honest: "We're offline until 9am tomorrow — we'll email you the answer first thing." Most visitors are fine with that. The team is fine with that. Trying to look 24/7 when you're not is what burns people out.
Step 5: snooze ruthlessly
A conversation that's already had its first response can be snoozed. The visitor isn't actively waiting; they'll come back when they have time. Snoozing keeps the inbox visually small, which keeps morale up.
Step 6: review weekly, not nightly
Once a week, look at: which patterns matched the most messages, which conversations breached the FRT-Human goal, which sites are noisy. Don't check FRT every hour — that pushes the team to game it (premature replies, unclosed quick-takes). Check it weekly; it's a strategic metric, not an operational one.
What we don't recommend
- →Pager-style alerts on every incoming message. They train the team to dread the notification sound.
- →A leaderboard ranking agents by reply time. They'll send empty "Hi! One sec…" messages to game it.
- →AI-only first responses without a human path. The 20% of messages that are nuanced will get the wrong answer and you'll lose the customer.
- →Replying to anything before reading the full message. It's the chat equivalent of cold-emailing.
FRT is a culture metric more than a tooling metric. The team that ships patterns + sets honest hours + snoozes ruthlessly will beat the team with the fastest software every single time.