What using YundaDesk looks like day to day

YundaDesk brings conversations from connected customer channels, customer context, AI customer service, and human collaboration into one workspace. This article explains what a team sees from initial setup through daily operations, what can happen automatically, and what still requires human judgment.

Available channels, apps, plan entitlements, and executable actions can vary by workspace. The current product page is always authoritative.

From setup to the first real customer conversation#

A team normally gets started in this order:

1. Open Channels, connect the website widget, Telegram, or another channel available in the current workspace, and complete a real send-and-receive test.
2. Create a customer-facing Agent, select one Role, and configure its service style, knowledge scope, skills, and required capabilities.
3. Bind the Agent to the channels where AI should respond. A channel is handled by no more than one Agent at a time. An unbound channel can still receive messages, but AI does not answer automatically.
4. Upload documents, sync a website, or connect an available knowledge source. Wait until the page shows that the content is ready to participate in answers.
5. Start a Test conversation from AI Agents and review the answer, knowledge, skills, and handoff boundary used for the response.
6. Send another message from a real customer client and confirm that the reply actually arrives. A successful configuration or sent status does not replace final verification in the real channel.

After this setup, the team no longer needs to work from every channel's separate administration screen for routine conversations. Customer messages enter the shared Inbox.

How the team handles conversations every day#

Agents use the Inbox to review new messages, ownership, conversation state, and customer context. Search and filters help them find conversations by the dimensions available on the page. Assignment, transfer, and internal notes keep responsibility clear when several people collaborate.

When a channel is bound to an enabled Agent, AI customer service can respond first. Human agents can inspect customer messages and AI replies, take over when needed, and continue with the same customer context. A conversation moves to the team when the customer asks for a person, the AI lacks reliable grounding, or a handoff condition applies.

The customer profile brings together confirmed channel identities, conversation history, tags, notes, and customer memory. Connected business apps may also provide order or activity summaries. When the relevant data source is not connected, the product does not invent that information.

What AI customer service handles automatically#

AI customer service generates replies from the current Agent's knowledge scope, skills, service style, available capabilities, and customer context. Different Agents can own pre-sales, after-sales, or other reception responsibilities and serve different channels.

AI can participate automatically in the customer conversations currently allowed for it, but one strong human reply does not immediately change every future answer. Strong replies, corrections, and unresolved questions can become improvement suggestions. They affect knowledge, skills, or another explicit destination only after the team reviews and adopts them. A skill must also complete its required tests and be enabled before it is active.

When knowledge is missing, required data is unavailable, permission is insufficient, or an external capability fails, the AI should explain the limitation, ask for the missing information, or hand the conversation to a person instead of guessing about an order, policy, or customer fact.

What Yuna does for the internal team#

Yuna serves the merchant team. It is separate from the customer-facing Agent that is speaking with a customer. You can ask Yuna to:

• explain conversations, customers, reports, and current workspace settings;
• find product instructions and describe the entry point, steps, and verification;
• help prepare channel, Agent, knowledge, and team settings;
• create and manage workspace tasks that run on demand or on a schedule;
• organize a business rule into reviewable learning content;
• summarize tasks that need confirmation, failed, or require follow-up.

Read-only questions can usually return immediately. An action that sends a message, affects a customer, or makes an important configuration change shows the target, content, or change preview first. It runs only when the current member has permission and completes the required confirmation. After an important action, review the result and verify it again on the related page or in the real channel.

How knowledge enters an answer#

Knowledge can come from uploaded documents, synchronized websites, connected document sources, and team-maintained questions and answers. The product shows processing or synchronization status. Only content that is ready and belongs to the current Agent's knowledge scope should participate in its answers.

The team can inspect matched knowledge and skills in a Test conversation or answer details. When an answer is wrong, first confirm that the source is current, the correct knowledge is bound to the Agent, and processing has finished. Correcting an existing fact and teaching a new handling method are different operations; both need an explicit target and follow-up verification.

What the team can measure#

Reports show the dimensions available on the current page, such as conversation volume, message trends, team reception, AI participation, automatic resolution, human handoff, customer satisfaction, and AI usage. Reporting definitions can differ, so review the time range, filters, and sample size together with each metric.

These results help identify frequent questions, knowledge gaps, channel-delivery issues, and team workload. They do not prove that every conversation was resolved correctly. Important conclusions should be checked against specific conversations, answer details, and real delivery results.

Channels, apps, and proactive outreach#

Channels send and receive customer messages. Apps provide order, customer, or other business context. Available data and actions depend on the channels and apps connected to the current workspace, the member's permission, and their current status. A live order or shipment question should use an available business capability rather than rely only on a static knowledge article.

Proactive outreach can prepare follow-ups, broadcasts, or messages for a segment. To repeat prepared outreach, open Yuna → Automations and use Customer outreach when that task type is available on the page. Every send still freezes the audience and message for human approval. Only a real delivery result or failure reason proves what happened. Changing frequency limits or quiet hours does not create or send a message by itself.

A reliable operating rhythm#

A team can use this routine:

1. Review new conversations, unanswered messages, and tasks that need a person.
2. Handle AI handoffs, channel failures, and customer follow-up.
3. Review improvement suggestions, update outdated knowledge, and retest critical questions.
4. Look for changes in reports, then inspect the underlying conversations to understand why.
5. Preview and test a new channel, capability, or automation before gradually exposing it to real customers.

This lets AI handle repetitive reception while people retain control over knowledge, risk, customer commitments, and real delivery outcomes.