Insight Agent Product Overview
1. Background & Positioning
1.1 Urgent Demand for General Analysis Scenarios
In advertising operations, there is an increasing need for a highly responsive analysis tool: management needs quick answers to ad-hoc questions, daily data anomalies require direct cause diagnosis, and any questions that come to mind need to be queried instantly.
Meanwhile, as customers deepen their use of the Reporting Agent, their needs have evolved from "generating reports" to "genuine analysis." User feedback is primarily focused on two points:
-
"Cannot directly answer my questions" — Insufficient intent recognition.
-
"Current reports are lengthy but lack deep insights" — A lack of in-depth insights makes it difficult to meet the analytical requirements of senior operators and brand decision-makers.
This requires us to gradually transition from programmatic Reporting to a general Insight model, demanding an Agent that is broader, more flexible, and capable of providing deeper insights.
1.2 User Insights
The "reports" previously generated by the Reporting Agent were highly programmatic. While valuable for improving efficiency, their "insights were relatively thin." Both brands and agencies urgently require genuine analytical capabilities to identify key issues and find solutions based on data performance.
-
Data accuracy is the cornerstone of trust and cannot be compromised. Users explicitly point out that "incomplete data = incorrect data." Before the Agent can provide credible advice, its "data querying" capability must be rock-solid. Finding all relevant entities and explaining the data clearly is the absolute prerequisite for building user trust.
-
Immediate Q&A (Quick-Query) capability is a strict necessity. Multiple users noted the need for a tool to "quickly query whatever comes to mind or is asked." This is crucial for handling sudden management inquiries and rapidly validating strategic ideas.
-
Demands vary across roles, requiring a balance between "macro" and "micro" perspectives. Management (e.g., Operations Directors) needs a top-down, global perspective to quickly identify key takeaways, while frontline operators need bottom-up, granular data and clear next steps for specific Campaigns or ASINs.
Summary of Demand Scenarios:
-
Instant Queries: Quickly query the performance of specific entities (ASINs, Campaigns) and obtain preliminary explanations when questioned by management or upon noticing data anomalies.
-
Global Health Scanning: Scan the overall account to quickly locate the best and worst-performing areas, bringing the Action Summary forward for rapid decision-making.
-
Deep Attribution: Address specific issues (e.g., "How to lower the ACOS of a certain ad") by combining multi-dimensional data for root-cause diagnosis and providing comprehensive solutions.
-
Opportunity Mining: Focus on specific goals (e.g., "Scaling up a high-potential ASIN") by providing clear recommendations for budget increases or reallocation.
Insight Agent is the next-generation analysis assistant launched exactly to address these needs.
2. Product Positioning & Core Capabilities
Product Positioning: We explicitly position Insight Agent as an intelligent assistant for advertising analysis and strategic advice.
It is not just a data querying tool, but an operational partner capable of assisting users with diagnostic attribution and providing high-quality decision-making recommendations.
Core Features: Flexible and broad intent recognition, reliable data querying capabilities, and a seamless multi-turn dialogue experience—it answers whatever you ask and supports continuous follow-up questions and in-depth analysis.
3. Data Scope
Insight Agent currently supports full-funnel data for Sponsored Ads (SA), including SP (Sponsored Products) / SB (Sponsored Brands) / SD (Sponsored Display), covering the following dimensions:
Entity Dimensions
| Entity | Description |
|---|---|
| Store (Profile) | Account-level analysis entry point |
| AI Managed Group (aiGroup) | Overall performance of the managed group and AI actions |
| Portfolio | Budget and performance at the portfolio level |
| Campaign | Deep analysis at the campaign level |
| Ad Group | Ad group dimension |
| Promoted Product (productAd) | Sponsored product dimension |
| Target | Includes keywords, product targeting, audiences, and auto-targeting |
| Search Term | Sourced from both keywords and product targeting |
| Product Line / ParentASIN / ASIN / SKU | Product-level analysis |
(Note: Labels dimension is not currently supported.)
All Performance Data
Includes all performance metrics: Spend, Sales, Orders, Clicks, Impressions, ACOS, ROAS, CTR, CVR, CPC, TACOS, etc.
Most Business Settings (New)
Properties, budgets, bidding strategy settings, and toggle statuses for all SA ad entities. (Note: Currently does not support detailed queries for automation rules or budget groups.)
All Operation Logs (New)
All modification logs for budgets, bids, targeting, ad structures, statuses, and managed group settings, including:
-
Manual operation logs by users on Xnurta
-
Logs triggered by AI automation rules
-
AI managed group setting change logs
Currently Unsupported Data: DSP / AMC / SQP data, etc. Currently Unsupported Capabilities:Directly modifying ad settings or issuing execution commands (e.g., changing budgets, modifying bids, pausing ads). Stay tuned!
4. Classic Scenario Examples
Scenario 1: Simple Queries (D1)
"Answer what is asked" — Quickly and accurately query data for specific entities or metrics. Typical Prompts:
-
"List all ASINs with a TACOS below 8% over the past month."
-
"Which of my store's campaigns currently have an 'out of budget' status?"
-
"What keywords brought me more than 10 orders last month?"
-
"Find all ad groups within campaigns that generated more than 5 orders in the past 7 days."
Scenario 2: Overview Analysis (D2)
Overall health scanning, identifying the best and worst performers, and comparing performance. Typical Prompts:
-
"Scan my account and identify problematic ASINs that require attention."
-
"Scan my account and identify problematic ASINs that require attention, mainly looking at ACOS and TACOS."
-
"Which campaign performed the best last month? Look comprehensively at CVR and ROAS."
-
"Compare ad performance between April and May, highlighting the best and worst-performing campaigns."
-
"Analyze the health of Campaign-X over the past 30 days."
Scenario 3: Deep Attribution (D3)
Root cause analysis for specific issues, combining multi-dimensional data and operation logs. Typical Prompts:
-
"The ACOS for campaign SP-Auto-New-Launch spiked from 20% last week to 50% this week; please diagnose the cause."
-
"Please identify the campaigns with the most significant month-over-month decline from April, analyze the reasons, and provide optimization suggestions."
-
"I noticed the conversion rate across the entire account is slowly declining. Please analyze if this is being dragged down by a few problematic ASINs."
-
"Why did the AI raise the bid for my keyword 'running shoes' from $1.0 to $1.5 yesterday? What is the underlying logic of the algorithm's optimization?" (Explanation of Xnurta's algorithm logic)
Scenario 4: Strategic Recommendations (D3)
Providing actionable optimization directions based on data analysis. Typical Prompts:
-
"My total monthly ad budget for Product Line A is $20,000. Please create a detailed budget allocation plan to maximize overall ROAS."
-
"Campaign A has performed very well over the past 14 days. Are there any other campaigns whose budget can be partially reallocated to it?"
-
"Please identify the keywords that require immediate bid adjustments among the top 10 ASINs by sales, and specify whether to raise or lower them."
-
"For the HKKay brand, identify keywords that should be negated and help me build a negative keyword library."
-
"Please analyze the campaigns that did not exhaust their budgets yesterday and the primary reasons why."
Scenario 5: AI Management Explanation & Review
Explaining the AI's behavioral logic and evaluating the effectiveness of AI management. Typical Prompts:
-
"Review the performance of this managed group over the past 14 days. How does it compare before and after enabling AI?"
-
"On January 1st, the AI paused the phrase 'Summer-Pro'. What was the basis for this?"
-
"What is the logic behind your AI's dayparting bid adjustments?"
-
"Summarize all AI actions for this managed group over the past week, focusing particularly on keyword additions and negations, and analyze this in conjunction with ad performance."
-
"Review the situation of all managed groups last month, analyzing them according to their different management goals."
5. Future Iteration Directions
Insight Agent is the core entry point of the Xnurta Agent ecosystem and will continue to evolve:
Short-term (Ongoing Optimization)
-
Continuously supplement various data sources and business knowledge bases to expand Insight Agent's capability scope and improve answer quality.
-
Upgrade the underlying query architecture and models to enhance accuracy and response speed.
-
Enhance the depth and actionability of strategic recommendations.
Mid-term
-
Deeply integrate with the AI optimization engine, enabling it to explain AI algorithmic behaviors and produce high-quality insights consistent with the optimization engine.
-
Connect to DSP / AMC data to achieve comprehensive analysis across multiple data sources.
-
Implement long-term memory and custom instructions, evolving into a dedicated assistant that truly understands your business.
Long-term
-
Evolve from "Insights & Recommendations" to "Execution & Action" — The Agent will directly generate and execute operational commands for ad management (e.g., creating campaigns, adjusting budgets, enabling AI management).
-
Integrate more off-site e-commerce data (organic rankings, inventory, market competition trends) to provide commercial decision-level insights.
If you have any questions or would like to try it out, please contact your corresponding CSM to request access.