When to use it
-
You want a quick overview of how ClaudIA works — from the help desk to auditing
-
You need to explain the tool to someone new on the team
-
You want a summary of the available settings and operational best practices
Prerequisites
-
Access to the ClaudIA project Hub
-
An integrated help desk and the operation up and running
About this article
This article brings an explainer video and the structured FAQ transcript with everything you need to understand about how ClaudIA works: integration, per-message decisions, content, auditing, and settings.
ClaudIA's Settings have been redesigned into a Hub (one card per section, opening a focused editor). The video below still shows the old screen in some parts, but the section names and paths mentioned in this text already follow the new Hub. For the current navigation, see ClaudIA Settings — the new Hub.
Integration and operation
What is ClaudIA and how does it integrate with my help desk?
ClaudIA connects to your help desk (WhatsApp, chat, etc.). Every customer message enters the help desk, is routed by rules/automation and, when assigned to ClaudIA, the conversation flows through it with full mirroring — you see everything both in the help desk and on the Cloud Humans platform.
When is the integration suspended?
The integration is paused when:
-
The conversation is resolved at N1
-
ClaudIA closes the ticket directly
-
There is prolonged inactivity with no new message from the customer
-
The ticket is transferred to N2 / manually "taken over" in the Hub
What should I not do in the help desk?
-
Do not remove the ticket from the inbox in the middle of the flow
-
Do not reply "on the side" without taking over the ticket
-
Do not "hand back" to ClaudIA after a human has taken over
Each help desk has its own particularities — check specific recommendations in the Help desk integrations category.
How do I "take over" a ticket manually?
In the Hub, view the open tickets and click Take all or Take a specific one. From that point on, the human takes over the flow of that conversation.
ClaudIA's per-message decisions
How does ClaudIA decide what to do with each message?
When it receives a new message, it evaluates:
-
Attachments (audio/image/document)
-
The possibility of resolving directly
-
Signs of frustration
-
Language
Then it searches for content, chooses the necessary agents, and only then replies, transfers, or triggers an interactive flow.
How does language detection work?
Identification happens at the beginning of the conversation (not on every message). From there, the conversation continues in the detected language, with automatic translation when needed.
How does ClaudIA handle attachments?
Attachments are processed and can be used to enrich the reply. Details in the specific FAQ.
Content and agents
What are N1, N2, and Interactive content?
-
N1 — the basis for ClaudIA's direct replies
-
N2 — guide transfers to a human
-
Interactive — trigger "another agent" (Controlled Flow) that drives the conversation end to end
When is the Clarifier triggered?
In two scenarios:
-
When there is not enough content to reply
-
When there is ambiguity (e.g., two pieces of content that could both apply)
When does ClaudIA transfer to N2?
When it identifies that the relevant content is N2.
An excess of low-information N2 content can bias the system toward transferring too often.
Can I control how often Interactive flows repeat?
Yes — you can limit how many times an interactive flow can repeat and define its start triggers.
Auditing and continuous improvement
How do I see what ClaudIA used to reply?
In Auditing, click View details: you can see which pieces of content ClaudIA considered and which one it used.
Can I edit content directly from Auditing?
Yes. When you edit there, the update already applies to the next tickets, speeding up the improvement cycle.
Best practices
-
Keep an eye on the dashboard and metrics
-
Audit 5–10% of conversations weekly
-
Optimize content with question variations for the same answer — this improves coverage without inflating N2
-
Use the Playground to test changes before publishing
What is the Playground for?
To simulate replies and compare versions of settings/content. The tests are recorded as test conversations and can also be audited.
Behavior settings
What tone and instruction adjustments can I make?
In Behavior → Base prompt, you can complement the prompt with company and ClaudIA context, define tone/empathy, and adjust style and tone of voice (line breaks, emoji use, jargon, slang) as well as the project-specific rules.
Can I break up long messages?
Yes. In Behavior → Message delivery → Split large messages, segmentation splits long replies into smaller parts.
How does the first-message greeting work?
ClaudIA recognizes the first message and can use the Initial greeting (dynamic), configured in Lifecycle → Greetings. If you enable the Static greeting, it takes precedence and the dynamic one no longer applies.
And the fallback / business hours?
You configure the Default business hours in Transfer → Human availability. Outside business hours, when ClaudIA needs to transfer, it sends the Out-of-hours message and avoids routing to someone who is unavailable (optionally being able to resolve as N1 outside business hours).
What is the Problem Resolved Checker?
It lives in Lifecycle → Automatic Resolution and checks with the customer whether the question was resolved before closing. It has two independent modes — you can enable one, the other, or both at the same time:
-
Scheduled — after a few minutes with no reply (defined in Minutes until the question), it sends a check-in ("Can I help you with anything else?")
-
On message (real time) — it evaluates each new customer message and reacts on the spot
Based on the reply: if the customer confirms everything is fine → it moves to closing; if they ask a new question → it exits the cycle and keeps helping; if they don't reply → the ticket is closed by the neighboring Automatic closing block, using its own Minutes with no reply.
Details in Problem Resolved Checker and in Automatic ticket closing at N1 (Ticket Resolver).
What are "transfer detection" and "repeated sections limit"?
They are mechanisms that monitor when to transfer (to avoid improper routing) and enforce limits when there are repeated attempts in the same session. The Repeated sections limit lives in Transfer → N2 Transfer.
Content modeling tips
How do I avoid transfer bias?
-
Create N2 content only when it truly requires a human
-
Give informational density to N2 content (the "when to transfer" should be clear)
-
Expand N1 with question variations to increase coverage without pushing cases to N2
How do I expand coverage without losing quality?
For each N1 answer, add synonyms and alternative ways of asking. This reduces ambiguity, triggers the Clarifier less often, and improves the N1 resolution rate.
Operational summary
-
Full mirroring between the help desk and the platform
-
Per-message decisions: attachments, language (at the start), frustration, resolve/transfer/clarify/interactive flow
-
Transparent auditing with quick content editing
-
Flexible settings: style/tone, message segmentation, greeting, human availability, and fallback
-
Problem Resolved Checker (two independent modes) + Automatic closing on inactivity, with configurable times
-
Recommended routine: monitor metrics, audit 5–10%/week, iterate on content with variations
Notes
-
To start with the initial settings: How to do ClaudIA's initial and essential settings
-
To understand the auditing process: How ClaudIA works and the auditing process
-
To understand the Hub screen: What are the main screens of ClaudIA's Hub