AI Recommendations

Turn transferred conversations into practical AI Agent improvements. AI Recommendations now helps your team identify Content Gaps, Data Gaps, and Action Gaps, then guides admins to publish reusable Q&A content, create Data Connectors, or build Skills that close the missing capability.

What AI Recommendations Do

When customers are transferred to a human agent, the conversation often reveals why the AI Agent could not finish the job. AI Recommendations groups repeated failures into gap types and recommends the next admin action: add missing knowledge, connect live customer or business data, or create an automation Skill for an external operation.

Gap type	What it means	Typical next step
Content Gap	The AI Agent is missing reusable knowledge content, such as a policy, product explanation, or troubleshooting answer.	Review and approve the generated Q&A draft into the Knowledge base.
Data Gap	The AI Agent needs real-time or customer-specific data, such as order status, account details, delivery progress, or subscription information.	Create a Data Connector, then use it in a Skill so the AI Agent can look up the required data.
Action Gap	The AI Agent needs permission and workflow steps to complete an external action, such as canceling a subscription, changing an address, issuing a refund, or updating access.	Create a Data Connector if needed, then create a Skill with confirmation, checks, the action step, and fallback handling.

Who Should Use It

This feature is designed for support operations, customer experience operations, AI Agent admins, and anyone responsible for keeping the AI Agent’s Knowledge base accurate and complete. Use AI Recommendations as a regular review queue: open the page, review the most important suggested gaps, and take the recommended action for each type.

Where to Find It

Open your AI Agent configuration.
In the left sidebar, select Recommendation.

The page opens with a review queue for AI-generated recommendations from repeated transferred conversations.

YClouder Recommendations page with a Content gap card, recommendation evidence, Q&A draft, and recommendation details — The YClouder Recommendations page shows the recommendation queue and the selected Q&A draft.

How to Read the Page

Area	What it means
Impact summary	The top of the page summarizes the most impactful ways to improve the agent’s resolution rate.
Gap distribution	The stacked bar shows how many recommendations are Content gaps, Data gaps, and Action gaps, plus each type’s share of the total.
Recommendation count	The item count shows how many recommendations match the current filters and search terms.
Recommendation list	Each card shows the gap type, impact level, recommendation title, summary, transferred conversation count, and generated time.
Detail page	The right side changes by gap type. Content Gaps open Review Q&A draft. Data and Action Gaps open Connect data and create skill.
Recommendation details	The details card shows gap type, expected impact, transferred conversations, missing capability, and status.

Find the Right Recommendations

Control	How to use it
Sort	Use the sort control, such as Impact, to prioritize recommendations that are likely to reduce the most handoffs.
Filters	Use Gap type, Status, and Generated time to narrow the queue. For example, keep Status set to Active when you only want recommendations waiting for review.
Search recommendations	Search by keywords from the recommendation title, topic, missing capability, or customer issue when you want to find a specific recommendation.

YClouder Recommendations page with the Filters panel open for gap type, status, and generated time — Filter recommendations by gap type, status, and generated time.

Review a Recommendation

Choose a recommendation from the list. Start with high-impact items when you want to reduce repeated handoffs quickly.
Check the gap type so you know whether the recommendation should become Knowledge content, a Data Connector, or an automation Skill.
Read the summary, gap evidence, AI Agent failure reason, and missing capability. These fields explain what the AI Agent could not answer or do.
Click View source conversations to inspect the transferred conversations selected by AI. These sources help you verify whether the recommendation is based on real customer needs and useful human replies.
For a Content Gap, review the AI-generated Q&A draft. You can edit the title and answer before publishing.
For a Data Gap or Action Gap, review the suggested Data Connector fields and Skill steps before creating the automation.
If the recommendation is accurate, use the primary action for its type: Approve as Q&A for Content Gaps, or Create a data connector, Create a skill, and Mark as done for Data and Action Gaps.

The source conversations are read-only. They are there to help reviewers understand why the recommendation was created and whether the suggested Q&A, connector, or Skill is trustworthy.

Source conversations panel for a YClouder Content gap recommendation with customer names and phone numbers blurred — Source conversations are read-only and show the evidence selected by AI.

Resolve Data and Action Gaps

Data and Action Gap recommendations open the Connect data and create skill page instead of a Q&A draft. Use this page to review the evidence, understand what capability is missing, and create the automation the AI Agent needs.

Page area	What to review	What to do next
Recommendation	Read the generated summary and status. Use View source conversations if you need to verify the transferred conversations behind the suggestion.	If the suggestion is not useful, reject it. If it is useful, continue to the automation setup.
Gap evidence	Review Agent failed because, Missing capability, and Data the agent needs. Data fields appear as chips, such as account balance, disputeId, or status.	Confirm that the missing capability and required fields match the real operational need.
Create the automation	The page recommends two setup cards: Create a data connector and Create a skill. The connector card describes the API capability to add; the skill card describes the workflow that will call the connector.	Create and test the connector first, then select it in the skill builder and finish the Skill.
Mark as done	After the connector and Skill are configured and tested, return to the recommendation.	Click Mark as done so the recommendation moves out of the active queue.

The Skill depends on the Data Connector. Configure and test the connector first, then select it in the skill builder.

Data and Action Gap recommendations do not publish a Q&A answer directly. They guide you through the missing automation setup so the AI Agent can retrieve the right data or complete a safe external action.

Step	What you configure	Why it matters
1. Create a data connector	Open the existing Data Connectors module with the recommended connector name and return fields prefilled. Add your API endpoint, authentication, request parameters, and response mapping.	This gives the AI Agent access to live customer or business data instead of relying only on static Knowledge content.
2. Create a skill	Open Skill Builder with recommended steps that call the Data Connector. For Action Gaps, include identity confirmation, eligibility checks, user confirmation, the action step, and a fallback to human support when the action cannot be completed safely.	This turns the missing capability into a repeatable workflow the AI Agent can follow.
3. Mark as done	After the connector and Skill are configured and tested, return to the recommendation and click Mark as done.	The recommendation moves to Approved, while still remaining available through status filters.

Build and test the Data Connector before finishing the Skill. The Skill depends on the connector response fields to look up data or complete the recommended action reliably.

Edit the Draft Before Approval

The generated Q&A draft appears for Content Gap recommendations. Before approving it, make sure the question sounds like something a customer would actually ask and the answer is clear enough for the AI Agent to reuse.

Q&A title	Q&A answer
Use a natural customer question. Keep it specific enough to match the repeated issue, but broad enough to apply to future customers.	Write a direct, complete answer. Remove anything that only applies to one customer, such as private account details or one-off troubleshooting notes.

Reject a Recommendation

If a recommendation is not useful, click Reject. You might reject a recommendation when the evidence does not support the suggestion, the missing capability is not worth automating, the source conversations are too customer-specific, or your team does not want the AI Agent to handle that topic. Rejected recommendations leave the active queue but can still be found with the status filter.

Recommendation action area with View source conversations, Reject, and Approve as Q&A buttons — Use Reject to dismiss the recommendation or Approve as Q&A to publish the reviewed draft.

What Makes a Good Recommendation

A good recommendation usually comes from multiple transferred conversations with a clear pattern and enough evidence to define the next improvement. The best recommendation type depends on what the AI Agent was missing: reusable knowledge, live data, or a safe action workflow.

Gap type	Good fit	Poor fit
Content Gap	A repeated product, policy, billing, or troubleshooting question that can become stable Knowledge base content.	The answer depends on one customer’s private account, order, payment, or delivery state.
Data Gap	The AI Agent needs to look up customer-specific or real-time information, and the required fields can be returned by an API.	The data source is not available, cannot be safely exposed, or changes too unpredictably to map into a connector response.
Action Gap	The AI Agent should perform a repeatable operation with clear eligibility checks, confirmation, success handling, and fallback rules.	The operation requires manual judgment, unsupported permissions, or policy approval that should stay with a human agent.

Overview

Quick Start

WhatsApp Business Platform

Channels

Inbox

Contact

Calling

Campaign

Journey

Agents

CTWA (Click to WhatsApp Ad)

Shop

WhatsApp pay

Integrations

Developer

Account Management

Pricing and Billing

Partnership

Support

What AI Recommendations Do

Who Should Use It

Where to Find It

How to Read the Page

Find the Right Recommendations

Review a Recommendation

Resolve Data and Action Gaps

Edit the Draft Before Approval

Reject a Recommendation

What Makes a Good Recommendation

​What AI Recommendations Do

​Who Should Use It

​Where to Find It

​How to Read the Page

​Find the Right Recommendations

​Review a Recommendation

​Resolve Data and Action Gaps

​Edit the Draft Before Approval

​Reject a Recommendation

​What Makes a Good Recommendation

What AI Recommendations Do

Who Should Use It

Where to Find It

How to Read the Page

Find the Right Recommendations

Review a Recommendation

Resolve Data and Action Gaps

Edit the Draft Before Approval

Reject a Recommendation

What Makes a Good Recommendation