Home Conversations and Dashboard | Monitoring ClaudIA
👀

Conversations and Dashboard | Monitoring ClaudIA

Conduct audits to optimize ClaudIA's content and monitor all your conversations and generated data
FabrĂ­cio Rissetto
Ariana Carvalho
By FabrĂ­cio Rissetto and 6 others
‱ 16 articles

How to get a table with ClaudIA's raw data

When to use - You need to analyze raw data from ClaudIA conversations outside the dashboard - You want to cross ClaudIA metrics with data from your own systems (CRM, BI, data lake) - You want to query a conversation programmatically, from your own environment Prerequisites - Access to your project's Cloud Chat as a supervisor, CX Engineer or administrator — for the Report Builder - For the technical paths (Data Export and API): a technical or data team on your side About this article There are three ways to get ClaudIA's data as a table today. Which one to choose depends on who will consume it: someone in the operation, your company's BI, or a system. How it works 1 — Report Builder (no technical team needed) Build your own tables, filters and cross-analyses inside Cloud Chat using the ClaudIA Topic: conversations, messages, generated responses, audit feedback, CSAT and knowledge base content. Exports to CSV, XLSX or JSON. It is the fastest path, and the one we recommend when your operation has no dedicated data team. 2 — Data Export: raw data in your own data lake (S3) We provide an S3 bucket dedicated to your company, with Parquet files partitioned by date and a daily D-1 load (today you have everything up to yesterday). This is the path for BI and data lake work, including joining support data with your other internal sources. The ClaudIA tables delivered to the bucket: | Table | One row is | | --- | --- | | claudia_conversation | a conversation ClaudIA handled | | claudia_messages | a message exchanged in the conversation | | claudia_prompt_revision | one call to the model, with the content it retrieved and the action that followed | | claudia_contents | a knowledge base content section | | claudia_conversation_feedback | a quality review made by a human agent | | claudia_csat | a CSAT response | | eddie_result_outcomes | one controlled flow execution and its outcome | The Cloud Chat tables come along with them (conversations, contacts, tags, CSAT, daily agent and team metrics). The Cloud Chat conversation id is the key that ties both sides together. Access is read-only, and control stays with you: your team sends the identifier of a role in your own AWS account, and we authorize it on the bucket. We do not issue any credential of ours to your team. 3 — Public API (one conversation at a time) We also have a documented public API. It is meant for querying and integrating, not for bulk extraction: - ClaudIA API — a conversation summary (status, resolution reason, tag, whether it went to N2, and the messages) and the execution trace behind a processed message - Cloud Chat API — list conversations, find a conversation by any identifier, list messages, contacts, labels, inboxes and teams Authentication uses the sign-in of a Cloud Chat user from your own account. Responses are paginated and filters follow a fixed allowlist of parameters — which is why, for a historical BI load, Data Export is the right path, not the API. How to request it 1. Report Builder: already available in Cloud Chat, under Reports. 2. Data Export: ask your Cloud Humans point of contact (CSM) or our support team. The service is not enabled by default. Our team provisions the bucket, your team sends the identifier of the role in your AWS account, and delivery starts the next day. 3. API: the documentation is public, and the Find your credentials page in the technical documentation issues the token for your user. Notes - Technical documentation for your data team: docs.cloudhumans.com — the Data Export table and column catalog, the full API reference (ClaudIA and Cloud Chat), bucket layout and best practices. - Data Export step by step: How to access your operation's raw data (Data Export) - ClaudIA Report Builder: How to create custom ClaudIA reports

Last updated on Sep 10, 2026

How to Use the ClaudIA Metrics Dashboard

When to use - You want to understand how to use the ClaudIA Metrics Dashboard end to end - You need to know what each tab shows and how to filter by project and period - You want to understand why conversations were transferred to N2 and how that connects to your settings - You are getting started with the dashboard and want a visual onboarding (video in Portuguese, English, or Spanish) Prerequisites - Access to the ClaudIA project ClaudIA panel with the Dashboard enabled About this article The Metrics Dashboard is the consolidated view of your project's ClaudIA performance. It updates automatically and brings together, in one place, retention, quality (CSAT and GenCX Score), transfer reasons, content usage, and flow performance. This article explains how to navigate the dashboard and includes the explainer videos in three languages. How to access and filter Open the dashboard from the project ClaudIA Panel (with the Dashboard enabled). At the top of the screen, use the global filters — they apply to every chart: - Project — selects the operation being analyzed - Start date / End date — the time range applied to all cards - Aggregation (day / week / month) — granularity of the time series How the Dashboard is organized The dashboard is split into tabs, each with an analysis focus: | Tab | What it is for | |---|---| | General Information | Overview: ticket volume, retention vs. goal, response time, and reasons for transfer to N2 | | CSAT | Customer-reported satisfaction, response rate, and breakdown by tag | | GenCX Score | AI-evaluated quality (% of bad cases), including by tag | | Tags | Volume, retention, and CSAT by topic/tag | | Content | Usage of N1/N2/interactive content, retention, and content that most often transfers | | Controlled Flows | Eddie flow performance (executions, drop-off, retention, escalations) | | Classifier and Clarifier | Clarification activations and cases with a high chance of missing content | | Bot Handoff (Asked for Human) | Topics that most drive human requests and content used before the request | | Audit | Audited errors by content, tag, and criticality | | Agentic Usage | Metrics for conversations where Agentic was triggered | | Data to export | Ticket and content listings for external analysis | The structure repeats by dimension (parallelism) One thing that makes the dashboard much easier to read: the same business metrics are repeated for every analysis dimension. In other words, the same "kit" of indicators — Usage, Tickets, Retention, CSAT, and GenCX Score — is available by Content, by Tag, and by Eddie/Flow. Learning to read one dimension means you can read them all. | Dimension | Where to find it | Available metrics | |---|---|---| | Content (section) | Content tab | Usage, tickets, retention, CSAT, GenCX + how much each content transfers to N2 | | Tag (topic) | Tags tab | Volume, tickets, retention, CSAT, GenCX over time | | Eddie / Flow | Controlled Flows tab | Executions, drop-off, tickets, retention, CSAT, escalations | Because the lens is the same, you read any dimension in the same order: Volume → Retention → Quality (CSAT/GenCX) → dig deeper into whatever is off. What you can do with this data (business applications) The dashboard is not just a scoreboard — it points to where to act. Some practical uses: - Prioritize content: find content with low retention or heavy N2 usage and prioritize creating/adjusting material — the most direct path to raising retention. - Turn N2 into N1: the card "Top content transferring to N2 - Potential to increase Retention" lists content currently resolved in N2 that could become N1. Each one is a concrete retention opportunity. - Fix section selection: "N2 Section Selection Errors — Used vs Correct" shows where ClaudIA used the wrong section — a direct content-review backlog. - Assess quality by topic: cross CSAT and GenCX Score by Tag to find topics with the worst experience and act on the prompt or the content for that topic. - Measure each Eddie/Flow's ROI: compare retention, drop-off, and escalations per flow to decide which to keep, adjust, or retire. - Close gaps that drive "Asked for Human": the Bot Handoff (Asked for Human) tab shows the topics that most request a human and the content used right before the request — great to discover what is missing. - Diagnose transfers: use the Reason for Transfer to N2 (below) to tie each handoff to a setting and prioritize the right fix. - Export for external analysis: the Data to export tab provides ticket and content listings for your own cross-analysis or A/B test tracking. :::info Drill-down example — transfers by N2 content. If you notice many conversations going to N2 because of N2 content, you can investigate in layers: 1. Open the "Top content transferring to N2" card and identify the content that transfers the most. 2. Use the ClaudIA Panel links (by content / by reason) to open the real tickets for that content. 3. Decide the action: turn the content into N1 (if ClaudIA can resolve it on its own) or fix/complete the material when the selection was right but the answer was incomplete. ::: Overview: "Resolution Reason N2" metrics (Reason for Transfer to N2) In the General Information tab, the card "Percentage of Tickets by Reason for Transfer to N2" shows why each conversation ended in N2 (transferred to a human). The key point: each reason maps directly to a behavior or setting in your operation. The friendly label shown in the chart corresponds to a rule you defined for ClaudIA — so this card works as a diagnosis of your own configuration. If a reason grows, you know exactly which setting to review. | Reason shown in the Dashboard | What it means / related setting | |---|---| | Customer asked for a human | The customer explicitly requested to talk to an agent | | Frustration detected | Frustration detection identified dissatisfaction in the conversation | | Negative feedback in the satisfaction check | The in-conversation satisfaction check indicated the problem was not solved | | ClaudIA was not confident in the answer | The confidence/attempt limit was reached | | Number of interactions above the message limit | The conversation went past the configured message limit | | ClaudIA went more than 10 minutes without responding | Handover by timeout (no response within the time window) | | Clarification attempt limit reached | The clarifier ran out of attempts to clarify the question | | ClaudIA detected a transfer action | A transfer message/action defined in content was triggered | | ClaudIA used N2 content / N2 content was first in the TOPK / Section reuse limit reached | Rules tied to content marked as N2 in the knowledge base | | No relevant content found | There was no content in the base to answer the request | | Eddie transferred / Error calling Eddie / Repeated Eddie flow / Incorrect use of Eddie | Behavior of controlled flows (Eddie) | | Escalated outside human service hours | The transfer happened outside the configured service hours | | Escalated due to an attachment | The conversation received an attachment and was transferred | | Forced N2 handover | A forced-handover rule was applied | | Multiprompt | The N2 answer was returned via multiprompt | :::info Next to it, in the same card group, "Links to access the ClaudIA Panel - Resolution Reason" takes you straight to the tickets for each reason in the ClaudIA Panel — useful to inspect real examples and understand what led to the transfer. ::: Videos by language Prefer a step-by-step video? Watch the full Metrics Dashboard tutorial in three languages: đŸ‡§đŸ‡· Portuguese (Brazil) https://youtu.be/PxY58YqvGUQ đŸ‡ș🇾 English https://youtu.be/yrsvO3rgzqQ đŸ‡Ș🇾 Spanish https://youtu.be/D524FmoNdX8 Notes - To understand the settings that trigger each transfer reason: ClaudIA Settings — Transfer and ClaudIA Settings — Behavior - To track retention: How to Improve Claudia's Retention? - To use the GenCX Score: How to best use the Gen CX Score - To understand CSAT: How to improve ClaudIA's quality (CSAT)

Last updated on Aug 26, 2026

📊 How to Improve Claudia's Retention?

Explanation of each tab, chart, and table, along with important notes on how to use and interpret the data. 🧭 General Structure of the Retention Section 📌 Important: - The start and end date filters do not apply; they only apply to the main metrics tab. Data is organized in fixed and complete weeks to ensure performance and standardization of analyses. - All percentages shown in the tables are relative to the total tickets for that week. For example, if an escalation reason appears with 2%, it means that 2% of all tickets in that week had that cause. đŸ§© Tab 1 – Retention Metrics - Overall This is the managerial tab, used to monitor progress, goals, and offenders in a consolidated manner. đŸ”¶ Chart: “Total Tickets × % Retained per Week” - Orange bars: total tickets received per week. - Orange line: percentage of tickets retained by Claudia (without escalation). - Helps understand volume trends and AI effectiveness. đŸ”¶ Side indicators: - Retention goal (set in the current operation's target) - Difference from last week vs. goal - Last week's retention - Maximum retention in the last 8 weeks - Average retention in the last 8 weeks đŸ”¶ Table: “Percentage of Tickets by N2 Reason” - Shows the main reasons for escalation to N2, week by week. - Highlights the main offender from the past 4 weeks, with an average value. - Helps identify persistent patterns or reasons, such as use of N2 content or transfer by Controlled Flow. 🛠 Tactical / Operational Tabs These tabs aim to support fine-tuning, content review, and case analysis. The structure is similar across all: 📁 Tab: Used N2 Content đŸ”¶ Table: “Retention Gain Potential by Content – Last 8 Weeks” - Displays the most recurring N2 articles in escalated tickets. - Weekly columns show the percentage contribution of each content item relative to total tickets handled by AI. - Use to prioritize adjustments and migration of content to N1 or interactive (Controlled Flow). đŸ”¶ Table: “Retention Gain Potential by Tags – Last 8 Weeks” - Shows the most recurring tags in escalated tickets. - Weekly columns indicate the contribution percentage of each tag relative to total tickets handled by AI. - Use to prioritize adjustments related to themes/tags. đŸ”¶ Chart: “Top Contents and % Retention Potential – Last 14 Days” - Shows the accumulated retention percentage that could be achieved by optimizing the Top N contents. - Example: optimizing the top 10 contents could yield an approximate 7.7% gain in retention. - Clearly demonstrates the impact of prioritizing adjustments on the right articles. đŸ”¶ Table: “Section Selection Errors – Used vs. Correct” - Shows when Claudia used incorrect content, based on support feedback. - Displays which section was used and which should have been (N1 or INTERACTIVE). - Helps identify content usage issues. 📁 Tab: Transferred by Controlled Flow (by design) đŸ”¶ Table: “Retention Potential (%) by Controlled Flow” - Indicates which Controlled Flows appear most frequently in escalations. - Values per week, allowing trend analysis. - Used to prioritize content or flow structure adjustments. đŸ”¶ Table: “Tickets Escalated by Flow” - Lists real tickets, transfer reasons, and flow links. - Serves for validation and context review. 📁 Tab: Customer Requested Human đŸ”¶ Chart: “Agent Interactions Before Handover” - Shows how many messages the customer exchanged with Claudia before requesting human support. - If most cases fall in the 3 to 5 interactions range, it may indicate lack of initial engagement or ineffective content. đŸ”¶ Table: “Tickets by Interaction Range” - Lists tickets corresponding to the interaction range depicted in the chart. đŸ”¶ Chart: “Top 5 Topics Customers Ask for Human Support” - Shows the most common topics where customers request human support. - Column: absolute volume of tickets | Line: % of total requests. - Helps identify topics with perceived low performance. đŸ”¶ Table: “Content Used Before Customer Requests Human” - Indicates which contents were shown by Claudia before the customer requested support. - Useful to understand if content contributed to frustration or confusion. 📁 Tab: Transferred Due to Controlled Flow Call Failure đŸ”¶ Table: “Controlled Flow Call Error % by Flow” - Shows flows that failed in technical calling, leading to escalation. - Used to identify infrastructure issues or bugs. đŸ”¶ Table: “Tickets with Call Error by Flow” - Lists tickets and flows with errors. - Ideal for validation with the technical team and specific fixes. 📁 Tab: Transferred Due to Controlled Flow Repetition đŸ”¶ Table: “% of Controlled Flow Repetitions by Period” - Indicates which flows reached the attempt limit without resolution. - Shows problems with content effectiveness or the need to adjust the number of attempts. 🧠 Best Practices for Analysis - Use the main tab to monitor overall progress, goals, and major offenders. - Use the tactical tabs to explore specific causes and prioritize content, flow, or response model adjustments. - Always consult ticket links when validating with real examples. If you have questions on how to interpret any data or want to suggest dashboard improvements, contact our team! We’re here to help 🧡

Last updated on Jun 24, 2026

Filters and Search System in Conversation

Filters 1. What is the Session ID used filter? This filter allows you to search for conversations based on the session ID (i.e., what content or section Claudia used in the response). To use this filter, you need to enable it in the panel settings: go to Settings → Other Settings → Show Session ID filter. 2. How does the Helpdesk ID filter work? This filter allows searching for conversations by the external identifier assigned to the ticket by your support system (such as Zendesk, Freshchat, etc.). In the case of Freshchat, there is a peculiarity: the ID that appears in the URL (the "internal ID") may be different from the ID used in Claudia's filter. To perform the correct search, it may be necessary to convert the internal ID to the ID used by the filter using the Freshchat API or request this conversion from support. 3. What are the Start Date / End Date filters for? These filters limit the time period of the analysis, showing only conversations that were closed within the specified range. They are essential for time-based searches, monthly comparisons, trend analysis, and specific audits. 4. What does the Tags filter mean? Allows filtering only those conversations that received certain tags during their flow or at closing. For example: login_error, billing, complaint. If no tag filter is specified (or "All" is selected), all conversations will be displayed regardless of whether they have tags or not. 5. What is the purpose of the Audit filter? This filter can be used to segment conversations that have already been audited from those that have not. Useful for quality teams, reviews, or controls, to focus only on conversations that have been analyzed or those that need to be analyzed. 6. What can the Identifier (Customer / Agent / AI) filter mean? This field probably defines who initiated the conversation: - Customer — conversation started by the end user - Playground — test conversations created in the Playground screen - Optimizer — batch-created conversations based on system history for evaluation. Usually done before launching new operations This filter helps differentiate automated flows from those initiated by humans. 7. What is the Resolution (N1 / N2) filter for? Although not explicitly documented, it is reasonable to assume this filter classifies the type of resolution of the conversation: - N1 — conversations resolved by AI (ClaudIA) - N2 — conversations that required human intervention (handover) With this, you can measure Claudia's autonomy versus cases that demanded escalation. 8. What does Reason for Resolution mean? This filter lists the recorded reasons when closing a conversation. It indicates which functionality was used to finalize in N1 or to transfer to N2 in each conversation. For a clearer parameter, you can access our FAQ category "Conversation Closure" or view the "Resolution Reason" in the internal note of each conversation. This helps you understand why conversations were closed, in addition to their status. 9. What is the purpose of the A/B Test filter? This filter allows selecting conversations belonging to one of the experiment groups (A or B), useful when you or your team are testing different versions of flow, prompts, or AI models. It is a helpful tool for performance comparison between variations. 10. What does Intent / Topics mean? This filter likely allows selecting conversations based on the LAST topic or intent identified by Claudia. The main use case is filtering cases where: Claudia identified the end and needed to close OR "Bot Sale (Forward to Human Check and Forward to Human Multi-prompt)" 11. What could Gen CX Score be? This field represents a score of experience or sentiment generated by the AI (similar to a customer rating or an internal metric). If available, it allows filtering conversations based on the AI's assessment (positive, neutral, negative). It helps prioritize audits at risk of being detrimental or cases where the AI did not behave as most appropriate. 12. What is the purpose of the Controlled Flow Only filter? It is also a reasonable assumption that this filter allows viewing only conversations responded to by the Controlled Flow model — an advanced or specific version of Claudia. This would be useful to compare Controlled Flow's performance versus the standard AI responses.

Last updated on Aug 26, 2026

How to Make the Best Use of the Gen CX Score

What is the Gen CX Score The Gen CX Score is a "Sentiment" metric assessing customer satisfaction by analyzing the entire conversation context. We use Large Language Models (LLMs) for this evaluation, which can range from: - Good (Score 5) - Neutral (Scores 2, 3, or 4) - Bad (Score 1) We categorize the metric this way instead of the traditional CSAT classification to identify extremes. Want a ticket with excellent service, clearly indicating that the issue was resolved? Look for "Good" Want to see tickets with a high risk of negative experience? Look for "Bad" Fluctuations within Neutral mean it's unclear whether there was a negative experience involving service, process, or product. Why Use the Gen CX Score? The goal of the Gen CX Score is to serve as an auxiliary metric for auditing, quality assessment, and CSAT prevention. It doesn't replace any of the above but can generate valuable auxiliary insights. It helps us gain a clear understanding of quality even when the volume of CSAT responses or audits is low. Additionally, when we start analyzing detailed metrics, the number of evaluations and audits disperses, making conclusions difficult (e.g., Controlled Flow performance or Content) Where and How to Use the Gen CX Score? Identify conversations with a high risk of negative experience: Access Conversations > Filter > Gen CX Score ![Image] (https://cloudchat.cloudhumans.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBBMktaRmc9PSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--e6cdd11c91e1e8eb06fd5c852a2c9223afc5f2f6/image.png) Filtering for "Bad" will help identify cases with high friction potential for auditing, content adjustments, or configuration changes. Monitor operation with an auxiliary indicator: Access Dashboard > CSAT and GenCXScore ![Image] (https://cloudchat.cloudhumans.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBBM0NaRmc9PSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--42252a30ffd73cafeb57fa3c0891dc0f574f2287/image.png) Keep track of operational performance with an additional indicator. Granular insights into content or Controlled Flows: Access Dashboard > Content or Controlled Flows ![Image] (https://cloudchat.cloudhumans.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBBNEdaRmc9PSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--2406bab85ff86f6b4ee4823688f0d7b8ba24cbdf/image.png) Compare specific content or Controlled Flows results to ensure you have a high number of evaluations per interaction.

Last updated on Jun 24, 2026

How to create custom ClaudIA reports (Report Builder)

When to use - You want to create custom reports with the data from conversations handled by ClaudIA (conversations, messages, generated responses, audit feedback, CSAT and content) - You need to go beyond the Dashboard and build your own tables, filters and cross-analyses - You want to export ClaudIA's data in CSV, XLSX or JSON Prerequisites - Access to the Cloud Chat of your project - Being logged in as a supervisor, CX Engineer or administrator About this article The Report Builder lives in Cloud Chat, but it also covers ClaudIA's data. When building a report, you'll find the ClaudIA Topic, with the tables of the conversations handled by the AI. This article shows where to access it — the full step-by-step is in the Cloud Chat guide (links at the end). How it works Where to access it The Report Builder is opened from Cloud Chat, in the Reports area. It's the same tool used for Cloud Chat reports — the difference is the topic you select when building the query. The ClaudIA Topic When selecting the fields, choose the ClaudIA Topic to work with the data from the conversations handled by the AI: - Conversations and messages - Responses generated by ClaudIA - Audit feedback - CSAT - Knowledge base content What you can do - Select dimensions (groupings) and measures (aggregations) to build the table - Apply advanced filters (dates, multiple conditions, across queries) - Use pivot tables, joins and period-over-period (PoP) analysis - Export in CSV, XLSX or JSON :::info Prefer to track ready-made indicators? Use the ClaudIA Metrics Dashboard. The Report Builder is for when you need to customize the analysis. ::: Notes - Full step-by-step: How to use the Report Builder — complete guide - Tables and columns of the ClaudIA Topic: Explanation of the Report Builder tables and columns - To bring the raw data into your own data lake or BI (Data Export and API): How to get a table with ClaudIA's raw data

Last updated on Sep 10, 2026

How to reduce N2 Content to increase retention

When to use it - You identified in the dashboard that N2 Content is one of the main reasons for transfers - You need a practical roadmap to reduce N2 without losing quality - You want to turn N2 into N1, automate it through a "Fluxo Controlado" (Controlled Flow) or remove duplicates Prerequisites - Access to ClaudIA's retention dashboard - Permission to edit content About this article N2 Content is one of the main reasons ClaudIA transfers conversations to humans. This guide shows you how to act immediately on the N2 content that generates the most transfers. :::info How to use it: read everything first, then open the dashboard to put it into practice. ::: Before you start :::warning The goal of the analysis is not just to say whether the transfer was right or wrong. Each piece of content needs to be evaluated carefully. When you open each ticket, analyze: - If incorrect → adjust it immediately to fix triggering failures - If correct → think about how to evolve it: - Turn it into N1 - Automate it through a "Fluxo Controlado" (Controlled Flow) - Adjust part of the content - Remove duplicates This careful look ensures a real improvement in retention, always focused on reducing unnecessary transfers. ::: đŸŽ„ Introductory video: https://youtu.be/3UWhUU4OmmA Where to start the analysis In the dashboard, go to the "ConteĂșdo N2 usados" (N2 Content used) tab. The tables show which N2 content is transferring the most tickets to humans. Aba ConteĂșdo N2 usados Time to act Practical suggestions :::success - Audit every ticket of the N2 content with the highest volume - Reclassify it as N1 when ClaudIA can solve it on its own - Create a "Fluxo Controlado" (Controlled Flow) for cases that depend on checking an external system - Remove duplicates that confuse the model - Refine the title of the N2 content so that it only triggers cases that really need a human ::: Notes - For the overall retention picture: How to track and increase ClaudIA's retention - To reduce Customer Asked for a Human: How to reduce Customer Asked for a Human to increase retention - To reduce Transfers through a "Fluxo Controlado": How to reduce Transfers through a Fluxo Controlado to increase retention - For N1/N2 best practices: Best practices for creating and maintaining content - To create Fluxos Controlados: How to request a Fluxo Controlado from the Cloud Humans team

Last updated on Aug 29, 2026

How to reduce Customer Requested a Human to increase retention

When to use - You identified in the dashboard that Cliente Pediu Humano (Customer Requested a Human) is impacting retention - You need a practical playbook to reduce this reason - You want to understand whether customers are requesting a human right at the start or after interacting Prerequisites - Access to ClaudIA's retention dashboard - The Venda da IA (AI Pitch) feature — Forward to Human Check — active in the project — see How to set up the retention attempt before transferring About this article Cliente Pediu Humano (Customer Requested a Human) is one of the reasons ClaudIA transfers a conversation to human agents. To validate whether this reason should be your focus, first check How to track and increase ClaudIA's retention. :::warning Important: viewing customers requesting a human is only possible if the Venda da IA (AI Pitch) — Forward to Human Check — feature is active. ::: :::info How to use this guide: read everything first, then open the dashboard to put it into practice. ::: The goal is not only to audit tickets, but also to drive improvements that increase retention. Where to start the analysis Go to the dashboard, on the Cliente Pediu O Humano (Customer Requested the Human) tab. Initial analysis - Is the customer interacting with ClaudIA before requesting a human, or do they ask right at the start? - What are the main topics that make customers request a human? Resolution reason filter = N2 - Customer requested a human (Cloud Chat) The first pieces of information on the tab refer to the number of interactions between the customer and ClaudIA before the request. Interactions before the request Time to act Practical suggestions :::success - Adjust the persuasive message of the Forward to Human Check - Identify frequent topics and improve the corresponding content - Reduce the time to the first useful answer to lower impatience ::: Notes - To understand the overall retention picture: How to track and increase ClaudIA's retention - To set up the retention attempt: How to set up the retention attempt before transferring (Forward to Human Check) - To reduce transfers via Fluxo Controlado (Controlled Flow): How to reduce Transfers via Fluxo Controlado to increase retention - To reduce N2 content: How to reduce N2 Content to increase retention

Last updated on Sep 16, 2026

How to reduce Controlled Flow transfers to increase retention

When to use it - You identified in the dashboard that Transfers via Controlled Flow are impacting retention - You need a practical roadmap to audit the Controlled Flows that transfer the most - You want to tell correct transfers apart from incorrect ones and act on both fronts Prerequisites - Access to ClaudIA's retention dashboard - Permission to edit the project's Controlled Flows About this article Transfers via Controlled Flow are one of the main reasons ClaudIA transfers to humans. This guide shows you how to act immediately on the flows that generate the most transfers, reducing the volume sent to a human. :::info How to use it: read everything first, then open the dashboard to put it into practice. ::: Before you start :::warning The goal of the analysis is not just to say whether the transfer was right or wrong. Each piece of content needs to be evaluated carefully: - If incorrect → adjust it immediately to fix triggering failures - If correct → identify improvement points even when the transfer was done well This careful look ensures that any audit results in a real improvement in retention, always focused on reducing unnecessary transfers. ::: Where to start the analysis In the dashboard, go to the "Transferidos por Fluxo Controlado (by design)" (Transferred by Controlled Flow — by design) tab. There you will find tables that help you quickly identify the biggest impacts on retention. Transferidos por Fluxo Controlado tab 1. Identify the Controlled Flows that transfer the most âžĄïž Table: "Potencial de Retenção (%) por Fluxo Controlado" (Retention Potential (%) by Controlled Flow) — shows which flows are driving the most transfers. :::success Focus first on the flows with the highest transfer volume. ::: Time to act Practical suggestions - Audit tickets from the flows with the highest transfer % - Map which flow conditions lead to the transfer - Optimize the flow to retain more customers (e.g., a more persuasive message, more options within the flow) - Eliminate unnecessary transfers when the case could have been solved by ClaudIA Notes - For the overall retention picture: How to track and increase ClaudIA's retention - To reduce "Cliente Pediu Humano" (Customer Asked for a Human): How to reduce Customer Asked for a Human to increase retention - To reduce N2 content: How to reduce N2 Content to increase retention - For Controlled Flow call errors (a related transfer reason): How to reduce Controlled Flow call errors

Last updated on Aug 29, 2026

How to audit ClaudIA's messages (sessions used and Tagger)

When to use it - You need to understand which sessions ClaudIA used to answer - You want to give feedback (positive or negative) about the use of a session - You want to audit the conversation's final tag based on the sessions used Prerequisites - Access to the Painel da ClaudIA (ClaudIA Panel) with auditing permission About this article This FAQ explains in detail the logic behind ClaudIA's selection of the sections it uses and the tagging feature (Tagger) when a ticket is closed. It also shows how to audit this data directly in the Painel da ClaudIA to get insights through the dashboard and improve tagging. đŸŽ„ Demo video Sessions used by ClaudIA What are "sessĂ”es utilizadas" (sessions used)? Every time ClaudIA answers a customer, it consults one or more sections of the base content (N1 or N2) to build the answer. These sections are called sessions used — the knowledge blocks the AI used to formulate the answer. Why does this matter? - Understanding how ClaudIA is reasoning in each answer - Auditing whether it is using the right content - Training and improving the AI's behavior by identifying incorrect usage - Defining the conversation's tag — the session with the highest weight influences the automatic tagging algorithm - Detecting a transfer to N2 — and whether it was made based on the correct content What was released You can now: - See directly which sessions were used in each ClaudIA answer - Give positive or negative feedback about the use of these sessions - Audit the conversation's final tag based on the sessions used How to view the sessions used 1. Open any ticket in the panel 2. Find ClaudIA's answer 3. Click the â„č icon to open the answer details 4. You will see: - ClaudIA's reasoning (searched term + intent) - The sessions used to compose the answer - An icon identifying which sessions were actually used Notes - For auditing best practices: What are the best practices for auditing ClaudIA - For error classification: How to classify errors when finishing an Audit - To understand the Tagger: How ClaudIA's Tag system works (Tagger)

Last updated on Sep 09, 2026

"Fluxos Controlados" (Controlled Flows) Report

When to use it - You want to know how many interactions went through each flow and how they ended - You want to find out at which step the customer gives up - You want to compare satisfaction and error rate across flows - You want to decide which flow to keep, adjust or turn off Prerequisites - Account with ClaudIA enabled - Administrator profile or user with reports permission - At least one "Fluxo Controlado" (Controlled Flow) in operation — with no flow, the report opens empty About this article The "Fluxos Controlados" report is under RelatĂłrio → ClaudIA → Fluxos Controlados (Report → ClaudIA → Controlled Flows) and shows the performance of each flow: volume, outcome, path taken, where the customer stops replying and the comparison between flows. It is different from the ClaudIA metrics dashboard: that one starts from the conversation, this one starts from the flow trigger. One conversation can set off more than one flow, and a trigger may not turn into any conversation at all. How it works The filters At the top of the screen: "Data" (Date — default: last 30 days), "Fluxo" (Flow — accepts more than one) and "Agrupar por perĂ­odo" (Group by period — day, week or month; it applies to the evolution chart). The "Caminho do atendimento" (Interaction path) chart has a filter of its own, right above it: "Fluxo (caminho)" (Flow (path)). This one accepts one flow at a time, because the drawing of the path only makes sense for a single flow. If you don't pick any, it opens on the highest-volume flow in your account. A trigger is not the same as a started interaction This is the most important distinction in the report. Every time a flow is set off, a trigger is recorded. But in a good share of the triggers the customer never replies — the conversation ends right there. These cases show up as "NĂŁo iniciou" (Did not start). :::info On average across accounts, "NĂŁo iniciou" is the largest slice of the triggers. That is why the cards at the top and the comparisons between flows count started interactions, not triggers. ::: In practice: - the "Resumo por fluxo" (Summary by flow) table brings both columns side by side — it is normal for the triggers one to be much larger than the started one; - a percentage calculated over triggers comes out artificially low. Always compare against the right denominator. Volume | Block | What it answers | |---|---| | "Iniciados" (Started) | Interactions in which the customer actually began to interact | | "Completos" (Completed) | Interactions the customer started and the flow finished successfully | | "Abandonos" (Drop-offs) | The customer stopped replying and the ticket was closed without going to a human | | "Transferidos" (Transferred) | Interactions that went to a human agent | Quality | Block | What it answers | |---|---| | "CSAT mĂ©dio" (Average CSAT) | Average satisfaction score from those who answered the survey | | "SLA 1ÂȘ resposta" (First response SLA) | Average time, in minutes, until the first reply | | "SLA resolução" (Resolution SLA) | Average time, in minutes, until the interaction is resolved | | "Recontatos (24h)" (Repeat contacts (24h)) | Customers who opened a conversation again within 24 hours | :::warning The four quality indicators consider the whole interaction, not only the stretch that went through the flow. If the customer was transferred to a human, the time and the score of that human service go into the calculation. ::: Change over time | Block | What it answers | |---|---| | "Evolução de uso" (Usage over time) | Started, Completed and Drop-offs over time. Use "Agrupar por perĂ­odo" to switch between day, week and month | | "Retenção por fluxo ao longo do tempo" (Retention by flow over time) | Percentage of interactions each flow resolved without escalating to human N2, period by period. It considers only interactions the customer started | Comparison between flows | Block | What it answers | |---|---| | "Desfecho por fluxo" (Outcome by flow) | How each trigger ended, in proportion. This is where the "NĂŁo iniciou" slice appears | | "CSAT mĂ©dio por fluxo" (Average CSAT by flow) | Score from 1 to 5 per flow, from those who answered the survey | | "% com erro por fluxo" (% with error by flow) | Percentage of the started interactions in which there was some failure in the flow | | "Handover para humano apĂłs erro" (Handover to human after error) | Interactions that went to a human and had some failure | "Caminho do atendimento" (Interaction path) | Block | What it answers | |---|---| | "Caminho do atendimento" (Interaction path) | Each band is the number of interactions that left one step and entered the next one. Pick a flow in the filter above the chart | To keep the drawing readable, stretches in which the interaction only moves forward — with no branching at all — appear summarized in a single band, from the beginning to the end of the stretch. The volumes remain correct: what disappears are the intermediate steps that were not deciding anything. "Detalhamento" (Details) All the tables in this section can be exported. | Block | What it answers | |---|---| | "Resumo por fluxo" (Summary by flow) | Counts per flow in the period, with triggers and started ones side by side | | "Detalhe por output (bloco de finalização)" (Detail by output (closing block)) | Where each interaction stopped inside the flow, with volume, satisfaction and times | | "Abandono por etapa" (Drop-off by step) | In which closing block the customer stopped replying — it points out where the flow loses the customer | | "Erros por fluxo" (Errors by flow) | Number of interactions with some failure in the flow | Frequently asked questions "NĂŁo iniciou" is huge. Is the flow broken? Probably not. "NĂŁo iniciou" means the flow was set off and the customer did not reply — it is the normal behavior of an outbound trigger and of a conversation the customer abandons before interacting. What is worth looking at is the proportion among the started ones: Completed, Drop-offs and Transferred. The "% com erro" is high. Is my flow failing? Not every failure blocks the interaction. Many of them are minor — a call that needed a retry, a field that came back empty — and the customer doesn't even notice. Use the "Erros por fluxo" table to see the real volume and the "Handover para humano apĂłs erro" block to find out whether the failure is costing you transfers. Does "Handover apĂłs erro" mean the error caused the transfer? No. It shows interactions in which both things happened — there was a failure and there was a transfer. It is a hint of where to investigate, not a cause-and-effect relationship. Is the CSAT here from ClaudIA or from the whole interaction? From the whole interaction. The survey is answered at the end of the conversation, and the conversation may have gone through the flow, ClaudIA and a human agent. To separate them, use the Cloud Chat CSAT report. Why don't the numbers match the ClaudIA metrics dashboard? The two look at different cuts: the metrics dashboard starts from the conversation, this report starts from the flow trigger. Can I export the data? Yes. The blocks in the "Detalhamento" section are tables and can be downloaded. Notes - To understand the concept behind the flows, see What the "Fluxo Controlado" is - For ClaudIA's general metrics, see How to use ClaudIA's Metrics Dashboard - To build your own cut of the data, see How to use the Report Builder

Last updated on Sep 22, 2026

How to use the ClaudIA dashboard - "Agentes Especialistas" (Specialist Agents)

1. What this dashboard is for To follow the performance of ClaudIA's "Agentes Especialistas" (Specialist Agents): how much of the volume goes through them, how much they resolve on their own, how much goes to human support, how the customer rates it, and where the tools are failing. 2. Three rules that apply to the whole dashboard 1. Only conversations that are already finished get in. An ongoing conversation has no outcome — counting it as resolved would inflate retention. 2. The most recent days read low. The Langfuse load is daily, so the day's agentic activity has not fully arrived yet. Do not read the last bar or the last point as a real drop. 3. Each metric has its own denominator. Retention and handover are about the conversations with a defined outcome; GenCX is about the scored conversations (today, roughly half of them); "Erros" (Errors) counts events, not conversations. That is why the numbers do not match one another. Every definition question is answered in the tile itself. Each chart has an ⓘ icon next to the title with the definition of the metric, the denominator used and whether or not it responds to the "Especialista" (Specialist) filter. 3. Filters | Filter | What it does | Note | | --- | --- | --- | | "Data da conversa" (Conversation date) | Period selection | Default: last 30 days | | "Agrupar por" (Group by) | Granularity of the time axis | Hour, Day, Week or Month (default: Week) | | "Tag" | Conversation tag | | | "NĂ­vel de resolução" (Resolution level) | N1 (ClaudIA) or N2 (human) | | | "Motivo de resolução" (Resolution reason) | Reason recorded at closing | | | "Teste A/B" (A/B test) | Key of the gradual rollout experiment | | | "Especialista" (Specialist) | Filters by sub-agent — it sits inside the "Especialistas" (Specialists) section, not in the top bar | Applies only to the specialist, tool and tool error tiles. See the question in section 6 | 4. Glossary | Term | Meaning | | --- | --- | | "Agente Especialista" (Specialist Agent, or sub-agent) | Agent with a specific scope that the supervisor triggers inside a ClaudIA conversation | | "Ativação" (Activation) | Share of ClaudIA conversations that went through the Agentes Especialistas | | N1 / "Retenção" (Retention) | Conversation resolved by ClaudIA, with no human | | N2 / "Transbordo" (Handover) | Conversation transferred to human support | | Tool | Tool that the specialist calls to look something up or to execute something | | CSAT | Rating given by the customer: positives Ă· responses received | | GenCX | A 1 to 5 score generated by a model that reads the conversation — it is not the customer answering | | "Erro" (Error) | Occurrence of an agentic failure (a single conversation may have several) | | Median / P80 / P90 | Typical case / the value that 1 in every 5 exceeds / the one that 1 in every 10 exceeds | 5. What is in each section "KPIs principais" (Main KPIs) | Tile | What it shows | | --- | --- | | "Ativação" (Activation) | How much of ClaudIA's conversations went through the Agentes Especialistas. The numerator is the same number as the "Conversas" (Conversations) card, which is why it is not repeated | | "Retenção" (Retention) | Conversations resolved by ClaudIA (N1) over the ones with a defined outcome. Below it, how many they were | | "Transbordo" (Handover) | Conversations transferred to a human (N2) over the ones with a defined outcome. Below it, how many they were | | "Conversas" (Conversations) | ClaudIA conversations that went through the Agentes Especialistas in the period | | CSAT | Positive ratings over ratings received. Below it, how many responses | | "Conversas com erro" (Conversations with an error) | Conversations with at least one agentic error, and how much that represents of the period's total | | "Ativação por perĂ­odo" (Activation by period) | The line is the activated share; the bars are how many conversations ClaudIA handled | | "Tempo atĂ© a primeira resposta" (Time to first response) | How long the customer waited, in seconds. Bottom line = median; top line = P80 | "Retenção e Transbordo (N2)" (Retention and Handover) | Tile | What it shows | | --- | --- | | "Volume de conversas — agĂȘntica" (Conversation volume — agentic) | Conversations per period that went through the specialists | | "Retenção e transbordo — agĂȘntica" (Retention and handover — agentic) | Split of each period between resolved by ClaudIA (N1) and transferred to a human (N2) | | "Motivo do transbordo — agĂȘntica" (Handover reason — agentic) | Why the conversation went to the human. "Repasse automĂĄtico" (Automatic handoff) is the default handoff, which records no reason | CSAT | Tile | What it shows | | --- | --- | | "Taxa de resposta do CSAT — agĂȘntica" (CSAT response rate — agentic) | How many of the conversations handled by ClaudIA answered the survey | | "CSAT — agĂȘntica" (CSAT — agentic) | CSAT of each period in the line; in the bars, how many responses back the number | | "Notas de CSAT por perĂ­odo" (CSAT scores by period) | Distribution of the 1 to 5 scores given by the customer | | "Notas de GenCX por perĂ­odo" (GenCX scores by period) | Distribution of the 1 to 5 scores generated by the model that reads the conversation | "Especialistas" (Specialists) | Tile | What it shows | | --- | --- | | "Especialistas — latĂȘncia tĂ­pica e P90" (Specialists — typical latency and P90) | Median and P90 latency of each specialist. Top 12 by P90 | | "Especialistas — retenção e transbordo" (Specialists — retention and handover) | Each specialist split between retained (N1) and handed over (N2); the third slice is the conversation that did not match with ClaudIA. Top 12 | | "Especialistas — acionamentos" (Specialists — triggers) | How many times each specialist was triggered. Top 12 | | "Especialistas — agĂȘntica" (Specialists — agentic) | Triggers, latency and error rate of each specialist | | "Conversas por especialista — agĂȘntica" (Conversations per specialist — agentic) | Handover, retention, CSAT and GenCX of each specialist | Tools | Tile | What it shows | | --- | --- | | "Tools — volume × latĂȘncia" (Tools — volume × latency) | Each dot is a tool: calls × average latency, size by the number of errors and color by the error rate | | "Profundidade da conversa — agĂȘntica" (Conversation depth — agentic) | Messages, tool triggers and distinct tools per conversation, over the period | | "Especialista × tool — mapa de calor" (Specialist × tool — heat map) | Error rate of each specialist × tool pair. The 80 most called pairs | | "Tools — volume e latĂȘncia — agĂȘntica" (Tools — volume and latency — agentic) | Most triggered tools, with average latency and error rate | | "Falha por tool — agĂȘntica" (Failure by tool — agentic) | Calls, failures and failure rate of each tool | "Erros" (Errors) | Tile | What it shows | | --- | --- | | "Erros ao longo do tempo — agĂȘntica" (Errors over time — agentic) | Errors per period. "Erros" (Errors) counts each occurrence; "Conversas com erro" (Conversations with an error) counts distinct conversations | | "Motivos de falha de tool — ranking" (Tool failure reasons — ranking) | Error reasons sorted by volume. Top 10 | | "Motivos de falha de tool — agĂȘntica" (Tool failure reasons — agentic) | Table of the reasons. "Conversas afetadas" (Affected conversations) is how many conversations felt the error | Tickets | Tile | What it shows | | --- | --- | | "Amostra de conversas com agĂȘntica" (Sample of conversations with agentic activity) | One row per conversation that triggered specialists. The link opens the conversation in the Hub | 6. Frequently asked questions Retention + Handover adds up to 100%, but it does not match the "Conversas" (Conversations) card. Why? Retention and handover are calculated only over the conversations that matched with ClaudIA and have a defined N1/N2 outcome. The "Conversas" card is the total that went through the specialists. Different denominators. Why does summing the "Conversas por especialista" (Conversations per specialist) column give more than the total number of conversations? A conversation that went through two specialists counts in both. The sum is of triggers, not of unique conversations. What is the difference between CSAT and GenCX? CSAT is the customer answering the survey. GenCX is a 1 to 5 score generated by a model that reads the conversation. GenCX covers more conversations (it does not depend on the customer answering), CSAT is the customer's voice. Why is "Erros" (Errors) higher than "Conversas com erro" (Conversations with an error)? "Erros" counts each occurrence; "Conversas com erro" counts distinct conversations. A conversation with 3 failures adds 3 to one and 1 to the other. The number of distinct conversations matches the card at the top. Why does last week's activation appear close to zero? The Langfuse load is daily, so the agentic activity of the most recent days has not fully arrived yet. Wait for the day to close before concluding anything. Where is the specialist filter? I cannot find it in the top bar. It sits inside the "Especialistas" (Specialists) section, right below the section title. Why does the "Especialista" (Specialist) filter not change the Retention card (nor CSAT, nor the charts by period)? Because those numbers are per conversation, and one conversation may go through more than one specialist — there is no way to attribute it to a single one. The filter applies to the tiles built on specialist triggers: the ones in the Especialistas section, the ones in Tools and the tool error ones. Each tile says in its own tooltip (ⓘ) whether it responds or not. Do Cloud Humans' internal tools count towards tool failure? No. The "Falha por tool" (Failure by tool) tile excludes the tools that operate the Cloud Humans platform — they are not customer operation and they fail much more often. What is "(tool nĂŁo identificada)" (unidentified tool)? Old failures in which the tool name was not recorded. It decreases as the load moves forward. What is "Repasse automĂĄtico" (Automatic handoff) in the handover reason? It is the default handoff to a human, which does not record a specific reason. Does an ongoing conversation show up in the dashboard? No, in no tile. Without an outcome, it would distort retention, handover and conversation depth. Where do I see a specific conversation? In the Tickets section, tile "Amostra de conversas com agĂȘntica" (Sample of conversations with agentic activity) — the link on each row opens the conversation in the Hub.

Last updated on Sep 09, 2026