Capturing a name and email is only the first step. Here is how lead capture, human handoff, and the analytics pages fit together in a working website chatbot.
Let AI handle the repetitive questions. Free your team to focus on real conversations.
Get Started FreeSetup in minutes · No credit card required
A chatbot that answers questions well is useful. A chatbot that also captures the right contacts and escalates at the right moment is a channel. The difference is mostly configuration.
Lead capture runs as a step in the chatbot flow rather than as a popup competing for attention. When the step runs, the visitor gets a form in the chat, and a successful submission saves both a contact and a lead record scoped to that workspace and chatbot.
Beyond name and email, you can define custom lead fields for your workspace so the form asks what your team actually needs, such as company size, service interest, or location. Field availability follows your plan limits.
Some conversations should not end with an answer. A handoff step moves the conversation into a waiting state and notifies workspace users.
Notification routing depends on who is around:
Users who are present in the app get an in app notification
When nobody is present, an email fallback goes out instead
Browser push is delivered when a user has subscribed and push is configured
Conversations that sit pending are closed by a timeout worker, so a missed handoff does not stay open forever.
The conversation archive lists sessions with status, a preview, metrics, and visitor context. Opening one shows the full transcript with source references, captured leads, session details, and any images the visitor uploaded.
Visitor context stays deliberately coarse. Country, region, and city come from hosting headers where available, along with device type, browser, operating system, language, timezone, screen, viewport, referrer, and page URL. Raw IP addresses are never stored for analytics.
Each analytics page answers a different question:
Analytics for the recent window overview, currently the last 30 days
Funnels for where conversations stop progressing
Costs for what your usage is actually consuming
Cohorts for whether returning visitors behave differently
Insights gaps for questions the chatbot could not answer
Sentiment, confidence, summaries, handoff status, channel, and tags are stored on the records themselves, so what you see is what was measured at the time, not a guess made from the message text afterwards.
If you only check one page each week, make it knowledge gaps. It is the shortest path from a real question to a better answer.
The pattern that works is small and repeatable. Read the gaps, add or correct a training source, retrain, then confirm the fix in Playground. Watch the funnel to see whether lead capture is firing at the right moment, and adjust the step if visitors are dropping before it.
Do that for a few weeks and the chatbot stops being a widget you installed and starts being a part of how the site works.
Everyone assumes you need a developer to build an AI agent. You don't. If you have a website, a no-code AI platform can turn it into a working conversational AI agent in minutes, one that handles support, captures leads, and never clocks out. Here's exactly how to train, test, and launch it right.
