Most live chat advice for SaaS is generic. This isn't.
These are five concrete, opinionated practices we've watched work for SaaS teams in the 1–20 person range. Not best-practice-by-committee — actual decisions you have to make on day one.
Where to put the widget
Default: bottom-right, fixed position, on every page including the dashboard. The exceptions are worth knowing:
- Pricing page: keep the widget visible. Pricing pages are high-intent — visitors are within 60 seconds of buying or bouncing. Chat catches the ones who would have bounced.
- Checkout / upgrade flow: keep it visible, possibly proactive. A visitor staring at the upgrade screen with a confused expression is the one conversation you most want.
- Docs: keep it visible but maybe small/collapsed. Docs visitors are usually self-serve; the widget is an escape hatch when the docs don't cover their case.
- App dashboard: debatable. Visible if your product is complex and users still need help post-onboarding. Hidden if you'd rather route post-onboarding support through email/ticket flow.
- Internal admin pages, public marketing landing pages targeting different personas: disable per-page via your tool's page-match rules.
What not to do: put the widget on a separate /support page only. The whole point of a widget is in-context help; making people navigate away kills the value.
What to make it say first
The default "How can I help?" greeting is wasted.
Better patterns:
- "Have a question about pricing? I can help." on
/pricing— catches the visitor right where they have the question. - "Stuck on setup? Ask me anything." on
/docs/install— frames the AI as the install helper. - "What are you trying to build?" on a product / features page — qualifies intent without sounding salesy.
If you're using AI, the greeting often doesn't matter much — the visitor types their question regardless. Where the greeting earns its keep is on pages where the visitor wasn't planning to chat but might if prompted: pricing, checkout, docs.
Anti-pattern: the proactive auto-open after 30 seconds. It's annoying. The pop-up-chat-bot-in-your-face pattern works for e-commerce cart-abandonment maybe; for SaaS it just trains visitors to dismiss the widget reflexively.
Hours vs always-on
Two questions to separate:
- Is a human available right now?
- Is the visitor going to get a useful response right now?
The old answer was: question 1 = question 2. If a human wasn't around, the visitor got nothing useful. So you set business hours.
With AI as the first responder, question 2 is always yes (assuming the AI is decent). Which makes the right answer:
- AI: always-on. No "we're offline, leave a message" experience. Visitor gets an answer regardless of timezone.
- Human availability badge: reflects real human presence (or your business hours). Sets the right expectation for when a human handoff will land.
Don't fake "online" when no human is around. The AI being on is honest; pretending a human is on isn't. The badge should tell the truth.
Hand-off rules
When the AI hands off to a human, three rules:
1. The AI says it's handing off
Visitor should see "I'm getting someone to help with this — they'll be with you shortly." Not silence after the AI gives up. The visitor needs to know the conversation didn't die.
2. The AI stays available during the wait
After flagging needs_human, the AI should keep responding to follow-ups in the same conversation. Visitor asks a different question while waiting? AI answers. Visitor says "thanks, never mind"? AI marks resolved. The handoff shouldn't freeze the conversation.
3. Notification has enough context to triage
The Slack ping / web push / email to the human shouldn't just say "new chat". It should include: who the visitor is, what they're asking, what the AI tried, the page they're on. The human shouldn't have to click through to figure out whether the ping is urgent.
Measuring it without obsessing
Three metrics that matter, in priority order:
1. Auto-resolution rate
Percentage of conversations the AI resolves without a human picking up. This is the headline number. If it's under 50%, your knowledge base needs work (the AI can't answer because the docs don't cover the topics being asked). If it's over 80%, the AI is doing the bulk of the work — you're free.
2. Time-to-human-response on flagged conversations
When the AI does hand off, how long until a human replies? Goal: under 15 minutes during stated availability hours, under 2 hours outside them. The visitor should never feel ghosted.
3. Conversation-to-conversion rate (for sales-pattern sites)
If your site is selling something — SaaS trial signups, demo bookings, paid plans — what percentage of chat conversations convert? Hard to measure perfectly (attribution is messy) but worth tracking directionally. If it's near zero, chat is a cost; if it's meaningfully above your overall page conversion, chat is a revenue lever.
Don't measure: chat volume, response time per message, "customer satisfaction score" from a 3-question survey at the end of chat. These are vanity metrics that you can game without making the experience better.
Putting it together
The high-leverage SaaS live chat setup for a small team in 2026:
- Widget on pricing, docs, marketing, dashboard. Hidden on internal admin.
- AI as first responder, knowledge base loaded with your docs + pricing + FAQ.
- Honest availability badge — AI always-on, human badge reflects real presence.
- Hand-off pings to Slack with full context.
- Auto-resolution rate as the headline metric you watch weekly.
That's it. Five decisions, each made once, that determine whether live chat is a force multiplier or a tax on your week.
For the tools side of it: we built Agentbot around exactly this pattern. Free during open beta. Or compare to the usual suspects before deciding.