Flocci Chat Start free
Plain answer

AI chatbot with human handoff, done properly

Short answer

Good handoff has four parts: a deterministic trigger (the visitor asks for a human, the topic is sensitive, or the answer failed), a ticket that carries the full transcript, a way to reach the visitor afterwards, and a reply path that lands both in the chat and in their inbox. Flocci Chat ships all four — escalations create FC-XXXXXX tickets in a built-in dashboard inbox, and operator replies appear in the widget and by email.

The failure mode everyone has experienced

A bot loops. You type "agent", it offers three articles. You type "human", it apologises and offers the same articles. That is not an AI problem — it is a missing escalation contract. Handoff has to be a deterministic rule the model cannot talk its way out of, not a suggestion inside a prompt.

The four triggers worth having

What has to travel with the escalation

A ticket that says "customer needs help" wastes everyone's time. Flocci Chat attaches the full conversation transcript, the page the visitor was on, the trigger that fired, and the visitor's contact details when they are known. If the visitor is anonymous, the widget renders an inline email-capture form once per session so the reply has somewhere to go. If they are a verified logged-in user — proved with an HMAC-SHA256 signature from your own backend — their identity and traits ride along and no form is needed.

Where escalations land

Two paths, both real. The built-in provider is Flocci Desk: tickets get FC-XXXXXX references, live in a dashboard inbox with a status lifecycle and a reply composer, and are the system of record. Operator replies insert into the conversation so the visitor sees a "Support team" bubble in the widget, and the same reply is emailed to them. The alternative is a webhook connector into an existing helpdesk, with credentials encrypted at rest (AES-256-GCM).

The email detail that decides whether handoff works

Escalation notifications carry a one-click entry link that signs the workspace owner into the right account and deep-links straight to the ticket — even if a different account is already active in that browser. Support that requires three logins before you can read the question is support that happens tomorrow.

Measuring it

The dashboard tracks how many conversations escalated and why, alongside first-token and completion latency, fallback rate and active alerts. A rising escalation rate on one topic is a content gap with a name; the unanswered-intent list turns it into a page you should write, and one click converts it into a canonical Q&A the assistant can answer next time.

Frequently asked questions

How does an AI chatbot know when to hand off to a human?

Through deterministic policy rather than model judgement: an explicit request for a human, a sensitive-topic keyword rule, or a genuinely failed turn (no usable retrieval, provider error, empty answer). Flocci Chat evaluates all three on every message and emits a handoff event the widget renders immediately.

Do I need a separate helpdesk for escalations?

No. Flocci Chat includes Flocci Desk — FC-XXXXXX ticket references, a dashboard inbox with status lifecycle, full transcript and a reply composer. If you already run a helpdesk, a webhook connector forwards escalations there instead, with credentials encrypted at rest.

How does the customer get the human's reply?

Both ways: the operator's reply appears in the chat widget as a Support team message if the visitor is still on the page, and it is emailed to them if they have left. Anonymous visitors are asked for an email once per session at the moment of escalation.

Does an escalation cost extra credits?

No. Only answered visitor messages are metered, at one credit each. Handoffs, ticket creation, operator replies and email notifications are free.

Can the chatbot escalate before the customer gets frustrated?

It escalates on sensitive topics and genuine failures without waiting to be asked, which covers most of the frustration curve. Sentiment-triggered escalation ahead of an explicit request is on the roadmap rather than shipped — we would rather say so than imply it.

Try it on your own website

One script tag. 500 free credits — roughly 500 answered visitor messages. Every feature on, no card. The widget in the corner of this page is the product, running on its own site.

Create your workspace — free