It's 9:40 on a Tuesday night. Someone is on your pricing page, and she opens the chat bubble to ask what your product costs for a team of eight. Your website chatbot answers in seconds, asks five questions, scores her against your ICP, and offers a booking link. By morning, the meeting is on the calendar and the transcript is attached to a new CRM record.
That's the version worth building. Done right, it identifies visitors, qualifies them, and writes enriched records into your CRM. Done wrong, it's a contact form with a speech bubble.
The difference is the questions you ask and where the answers go. Most chatbot advice stops at "capture the email." That's half the job. This post covers the other half: five questions that route and score, plus the wiring that turns chats into pipeline. If you're picking a website chatbot for your site, use this as your build checklist.
What a website chatbot should do beyond collecting emails
Does a website chatbot generate leads? Yes, when it's built to. The evidence is decent, with a couple of caveats worth knowing.
One 2026 roundup reports a useful comparison. AI chatbot conversations captured leads at an 11.4% rate, while forms on the same pages converted at 3.8% (Conferbot A/B tests across 340 sites, via Zappiq). Other aggregated numbers point the same way: 64% of businesses using AI chatbots report more qualified leads, and real-time interaction has lifted B2B conversion rates by up to 20% (Martal).
Two caveats, though. "Engaged visitor" means someone who opened the chat and sent a message, not everyone who lands. Engagement itself typically runs 15 to 30% of visitors, so greeting copy and widget placement do real work (Zappiq). And these numbers are benchmarks to test against, not promises.
Speed is the second lever. The odds of qualifying a lead are 21× higher when you respond within five minutes instead of thirty, per the MIT Lead Response Management Study (Featurebase). Sales reps, meanwhile, spend under 30% of their day selling. A bot answers in seconds, around the clock, and never gets busy.
So the real job of a website chatbot for lead generation is four things: start the conversation, qualify, route, and record. Email capture is a byproduct, not the goal.
The missing half: identification and enrichment after the chat ends
Here's where most setups fall short. The chat ends. The email lands in a spreadsheet or a separate tool's inbox, and the visitor's company, role, and page history never connect to anything. A week later, outreach runs with no memory of the conversation.
The fix is identification and enrichment. A work email or domain lookup resolves the visitor to a company: size, industry, likely role. The transcript, pages viewed, and score attach to that record, so follow-up references the real conversation instead of a generic template.
This changes what your website chatbot asks. Anything you can enrich afterward, like company size from a domain or role from a profile, shouldn't be a question. Every question you cut is a 5 to 10% bump in completion rate, per Featurebase. Their figure; test your own.
In Zrevick, this is the Identity Graph. Chats, emails, CRM records, meetings, and documents resolve into one living profile per person and company. The chat isn't a dead-end channel. It's an intake point.
Designing qualification questions that route and score in one step
What questions should a website chatbot ask to qualify a lead? Enough to score and route, and no more. Keep it to three to five questions; past five, drop-off climbs fast (Featurebase). Three or four covering intent, fit, and action is the sweet spot (Featurebase).
Each question below does double duty: it earns points toward a score and picks a branch. The example assumes a seed-stage B2B SaaS seller whose ICP is technical founders and small teams. Treat the numbers as a starting point, not a rule.
1. Intent and timeline. "Are you evaluating options right now, or doing early research?"
- Evaluating now: +25
- Next quarter or two: +10
- Just researching: 0, and a hard blocker (see below)
2. Need. "What's the main thing you're trying to solve?"
Free text. The agent matches the answer against your top use cases. Match: +20. No match: 0, with a route to content. Support question? Branch to help content instead of the sales flow.
3. Authority. "Are you looking for this for yourself, or on behalf of a team?"
- Choosing for self or team: +15
- Researching for someone else: +5
4. Fit. "Roughly how many people are on your team?"
- 1 to 50: +10
- 51 to 200: +5
- Outside your segment: hard blocker to a resource path
5. Budget. "Is budget already scoped, or still being figured out?"
- Allocated: +20
- Still scoping: +5
- Rather not say: 0
Order matters. Ask budget last, once the bot has answered something useful. Chatbots that ask budget early see 40 to 50% abandonment. Asked at the end, completion runs 75 to 85% (Zappiq's roundup).
How the website chatbot applies the score
It sums the points and applies two overrides:
- Hard blockers beat totals. "Just researching" or an out-of-segment answer routes to nurture or content. A high total never overrides a blocker.
- Missing answers score zero, not negative. A skipped budget or authority question is missing information, not evidence of poor fit. Don't subtract; just don't award.
Routing thresholds for this example: 50+ offers a meeting. 25 to 49 enters a nurture sequence. Under 25, or any blocker, gets a resource and a clean exit. The maximum is 90 points, so 50 means most signals fired, not perfection.
Sanity-check the branches before launch. "Evaluating now" (25) plus need match (20) plus authority (15) equals 60, a meeting, even with budget unscoped. That's the right call for founder-led sales. "Just researching" with a 65 total goes to nurture. The blocker wins.
Keep an escape hatch in every flow: a "talk to a human" button. People who feel trapped close the tab (Featurebase).
From chat to pipeline: connecting conversations to CRM records and sequences
Where do chat leads go after capture? This step decides whether the website chatbot was worth installing.
The minimum wiring, per Featurebase's setup guide:
- A CRM record created the moment the conversation ends, with score, source, and full transcript
- A booking link for hot leads, offered inside the chat
- A notification to whoever works the lead
- A nurture sequence for warm ones
The Zrevick version collapses this into one system. The site, the chat, the CRM, and the outreach sequencer share a workspace, so integration is configuration. A qualified chat creates a record. The AI SDR runs outreach from that same record, and the transcript rides along. No exports, and no context lost between tools.
For a founder doing founder-led sales, the handoff matters too. Live chat software for startups usually means bot first, founder second. The bot handles first touch and qualification; you jump in when a lead runs hot or gets frustrated. Bots handle volume. Humans handle nuance.
One more leak to plug: drop-offs. Someone closes the tab mid-conversation, or leaves an email but skips the timeline question. A good bot follows up: "You were checking pricing earlier. Want a quick comparison?" Recovering half-finished conversations is where most setups leave pipeline on the table (Featurebase).
Setup walkthrough on a Zrevick site in an afternoon
Here's a realistic afternoon plan for standing up a website chatbot. Expect to iterate afterward; the first version of any qualification flow is wrong somewhere.
Hour 1: decide, don't build. Write down your ICP and what "qualified" means: segment, team size, timeline. Draft the five questions with options and point values. Set thresholds and hard blockers. This hour determines whether the bot qualifies anyone.
Hour 2: configure the chat. In Zrevick, the site and chat live in the same workspace as your CRM and sequences. Enable chat on the site. Write a greeting that offers help, not a pitch. Set the trigger; pricing-page visits work well. Give the agent your ICP definition and routing rules.
Hour 3: wire the routing. Hot leads get a booking link plus a notification. Warm ones enter a sequence. Cold ones get the newsletter or a piece of content. Verify each branch with a test chat. Check the record, the attached transcript, the score and source, and the sequence enrollment.
Last 30 minutes: break it on purpose. Test the edge cases. Try "just browsing," a mid-chat drop-off, a support question in a sales flow, and a "rather not say" on budget. Every branch should exit cleanly.
Cost note: Zrevick's Starter plan runs about $34 per month billed annually, with a lead allowance around 10,000 (pricing).
Honest timeline: Featurebase pegs a basic setup at about a day, and a tuned version takes two to three weeks of iteration on real conversations (Featurebase). A single afternoon is aggressive. It's realistic when site, chat, CRM, and sequences are one system, because there's no integration work. Ship the afternoon version. Refine weekly against chat logs.
Metrics: chat-to-meeting rate and what good looks like for early-stage traffic
Track one primary metric for your website chatbot: chat-to-meeting rate, meaning meetings booked divided by conversations started. It's the number that ties chat to pipeline. Supporting numbers: engagement rate (chats opened ÷ visitors), completion rate, and qualified rate (hot scores ÷ completions).
Here's an illustrative calculation, not a benchmark. Say 1,000 monthly visitors. Twenty percent open chat: 200 conversations. Half complete the questions: 100. One in five scores hot: 20. Half of those book: 10 meetings. That's a 5% chat-to-meeting rate, 10 of 200, and 1% of total visitors. Run your own numbers; every step is a visible lever.
Early-stage caveat: with low traffic, these numbers are noisy. Ten conversations a week means one good week can double your rate. Judge trends over 20 to 30 conversations, not days. Don't rewrite the flow every Monday.
What does good look like? Published benchmarks mostly cover lead capture, not meetings. Chatbots convert 10 to 15% of engaged visitors into leads, versus 1 to 3% for forms, per one roundup. Your chat-to-meeting baseline is yours to set in month one. Then tune the biggest leak. If engagement is low, rewrite the greeting. If completion drops, reorder the questions. If "hot" leads ghost you, move the threshold.
Review scoring rules monthly for the first quarter, then quarterly (Featurebase). The first model is always wrong. The second is usually close.
The short version: a website chatbot earns its place when the loop closes. The visitor chats, gets identified, scores against your ICP, and lands in the CRM your outreach runs from. Five questions, two overrides, an afternoon of wiring. Then iterate.
Back to the 9:40 p.m. visitor. The bot gets some of the credit, but the meeting happened because the chat, the CRM, and the outreach lived in one system, so nothing fell through the cracks between tools. If you want that loop on your own site, Zrevick's Starter plan runs about $34 a month, and the afternoon plan above is your first version. Build it this week, then tune it against real conversations as they come in.