Preface
Reporting Agent is a new milestone for the SparkX Agent product system in 2025. It supports fully free conversation input, enables comprehensive query and analysis across pages, modules, and underlying data, and generates reports with a single click.
Model Used
During the internal testing phase, we will continuously evaluate and selectively use multiple large models, including but not limited to: GPT-5, Claude-Opus-4, Claude-Sonnet-3.7, Gemini-2.5-Pro, Grok-4, DeepSeek-R1.
Models used in the current version: Claude‑Sonnet‑3.7, Claude‑Sonnet‑4.
Data Security
We access various large model capabilities through Amazon Web Services (AWS) Bedrock. Your data is only used to generate your own analysis results and will not be used to train any large models; data between different customers is strictly isolated.
Patience and Exploration
During use, the model may still have deficiencies and errors. SparkX will continue to optimize product performance based on our experience in the Amazon advertising field, as well as our solid capabilities in AI Agent development, testing, and governance.
Currently still in the Close Beta phase of intensive iteration. Thank you for your understanding and support, and we look forward to your valuable suggestions!
Topic Overview
You can start by trying these topics
Overview of Advertising Campaign Performance
Supports overall trends, core ad performance, anomaly diagnosis, comparison between working days and weekends, etc.
Contribution of advertising to the overall business
Compare the contributions of organic traffic and advertising traffic; evaluate the impact of advertising on the store's total sales and TACOS; compare the proportion of advertising sales in each product line
Product Analysis
Diagnose all store products, identify key/high-potential/inefficient products; analyze the advertising performance of each child ASIN under each Parent ASIN and provide resource allocation recommendations
Targeting Type Analysis
Conduct in-depth comparisons of targeting types such as automatic, keyword, product placement, and audience; view the proportion of spend and sales, and drill down to each matching method
Advertising Structure Analysis
Compare the performance and business share of the three types of advertisements (SP, SB, SD); display key metrics and budget/sales share, and generate budget allocation recommendations
Budget Usage Analysis (Current Month)
Monthly budget utilization progress for the entire store or product line; identify advertising campaigns with budget anomalies
Note: Currently only "Monthly Budget Usage" (estimated monthly planned budget based on the current daily budget) is supported.
SB Special Topic
Regular Overview, Targeting Type × Matching Method
Comparison between SB and SBV; Effect of SBV video creatives
New-to-Brand Analysis
Automated Bidding vs Custom Bidding
Performance differences among different Placements
Comparison of Markets in Different Countries
Compare the core advertising metrics and budget/expense ratios across different national markets; provide cross-market budget allocation recommendations
Search Term Analysis
Word Frequency and Root Analysis
Find TOP N by data; Trends; Four quadrants (High efficiency/High potential/Low efficiency/Invalid)
SP Special Topic
Regular Overview, Orientation Type × Matching Method; Linked Analysis of Different Bidding Strategies and Placement
Advertising Campaign Analysis
TOP N, Trend, Quadrant (High Efficiency/High Potential/Low Efficiency/Inactive)
Keyword Analysis
TOP N, Trend, Quadrant (High Efficiency/High Potential/Low Efficiency/Inactive)
SparkX's underlying layer has Text-to-SQL capabilities, not limited to the above topics: it can combine multiple topics, continue in-depth exploration, and more usage scenarios await your discovery.
Note: Text-to-SQL refers to a system that, based on your natural language request, understands the intention on the spot and "writes SQL on the fly". It does not reuse fixed templates but instead matches tables and fields, infers conditions and aggregations based on your question, dynamically generates, and executes SQL.
The following data is not yet available for internal testing. Please stay tuned.
Detailed Analysis of AI Hosting
Operation Log Level Data
Category data benchmark
...
New Users Must Read ⭐️
Happy Path
First select the data range
Before starting any analysis, it is recommended to first set the data range:
All topics in the report will be calculated within this scope.
The reporting period serves as the primary time window; the comparison period can be selected as "sequential/year-on-year/custom," and all topics will follow this time setting.
The stores and products selected here serve as the data filtering scope for the entire report. The analysis granularity and grouping of each topic can be specified in natural language during the conversation (see 3.2.1).
Whenever you need to change these settings, simply inform the AI, and the system will intelligently evoke the settings component.
Q: Why do we need to specify these ranges through settings?
A: When specifying time, store, and product in natural language, ambiguity is likely to occur, and the threshold for precise input is relatively high. Through settings, a better balance can be achieved among convenience, accuracy, and permission control.
Don't know how to Prompt?
If there are no specific issues for the time being, you can directly provide a thematic instruction, for example: "Conduct targeted analysis for me" or "Create a version of product analysis". Even when the requirements are relatively vague, SparkX Agent will still provide you with some recommended analysis directions based on industry best practices.
The Agent will intelligently select the most relevant indicators based on the topic, and generally you do not need to specify them item by item.
Observe the outline recommended by AI and input your requirements in a similar structure to reduce communication rounds and speed up generation.
Advanced Journey: Refine Your Individual Needs
Freely specify the analysis dimension for each topic
All topics support specifying aggregation dimensions and comparison methods using natural language, for example:
Specify to view trends with "week" as the granularity
Specify to compare features by "Product Line"
Specify analysis with "Parent ASIN" as the granularity
Conduct multi-dimensional cross-analysis
Tips: You can simply say "For a certain topic, analyze it according to XX dimension", "Compare the characteristics of different XX ", "Find the best and worst performing XX ".
XX can be: Parent ASIN, ASIN, product line, advertising campaign, daily, weekly, monthly (the supported dimensions may vary for different topics)
There is no need to type the name of a specific entity
Define your business definition and preferences
Example 1: Professional operations often use the "four-quadrant" analysis method, which involves distinguishing different objects such as high-efficiency, high-potential, low-efficiency, and ineffective ones, and analyzing their performance
The challenge is that the definitions of "high efficiency/high potential/low efficiency/ineffectiveness" vary across different business stages. For example, during the maturity stage, cost efficiency is emphasized, where high efficiency = high spend + low ACOS; during the new product ramp-up period, it may be defined as high spend + high sales.
You can directly tell the Agent your definitions and concerns, and the system will generate analysis according to your specifications.
What kind of "search terms" do you think are efficient?
What kind of "advertising campaign" do you think is effective
How should it be determined when placing more emphasis on "conversion rate"?
Example 2: Anomaly analysis, tell the AI what "anomaly" is, and the system will screen and explain according to your rules.
Example 3:Logic of root analysis (recommended)
"Root analysis in groups of three words, outputting lists of high-contribution and high-cost word groups and cleanup suggestions."
"Create another version of root analysis with two words as a group."
More usage scenarios await your discovery! 👏🏻
You can also choose not to make any presets, and the Agent will first generate a version according to best practices, then iterate quickly based on your follow-up questions.
Capability Boundaries and Evolution
Currently, the Reporting Agent focuses on in-depth analysis and thematic reports. Depending on the complexity of the requirements, producing a report typically takes 6–8 minutes; even for simple issues, the complete process of "setup—outline—report" will be followed.
Lightweight analysis scenarios for "Quick Q&A" are coming soon. Stay tuned!
Sometimes you may only ask about one point, but the AI will output relatively long content. The principle is the same as above: we prioritize generating structured in-depth analysis, which combines best practice prompts and RAG knowledge in the background. Sometimes the AI may be too "immersed" in its own thinking and fail to fully respond to the details in your language, and we are continuing to optimize.
You can emphasize key points through follow-up questions, and the Agent will regenerate a version accordingly. In the short term, the full process will still be triggered, and we will record the scenarios and promote lightweight solutions.
Follow-up example: (For the already generated draft report, continue to raise your requirements or derivative questions)
Following up on the basis of the report results is the best way to meet the current demand for enhanced expression.
Management Report and Ideal Report for Reuse
Report Library
All reports will be automatically collected into the "Report Library".
It is recommended to mark reports with ideal results as "Favorite" for easy and quick reuse.
Create a periodic plan
When you have repeatedly refined an ideal report with AI, you can choose "Create Cycle Plan", and the Agent will automatically generate it according to the set cycle.
Entrance
Entry 1: Report Library - Intelligent Rendering
Entry 2: Within Session - Report Details
Periodic Plan Settings
Manageable plans in the report library
After successful execution, the report library will display a new report with the source "Periodic Plan".
If "Automatically create a session for new reports" is checked, it will be visible in the session list
The Agent will automatically execute tasks as planned
Batch Reuse
Reuse ideal reports in batches across different scenarios such as products, product lines, etc. The system will reuse the outline and prompt of the original report to the greatest extent possible and support the addition of differentiated requirements.
Entrance
Entry 1: Report Library - Intelligent Rendering
Entry 2: Within Session - Report Details
Batch Reuse Process
Prompt Window
After successful production, the report library will display a new report with the source "Batch Reuse".
New Meeting Session
Starting from any report, "create a new branch" to continue the conversation and optimization.
All reports can serve as templates, being generated periodically, reused in batches, and continuously optimized. In the future, capabilities such as in-team sharing will be supported.
After obtaining the ideal analysis/report through the methods described in Chapter 3, it is recommended to expand its applicable scenarios to allow experience to flow and be reused within the team.
Conclusion: Open co-construction, continuous iteration
SparkX Reporting Agent does not take the "Function List" as its boundary; instead, it focuses more on co-creating with you in real business scenarios. Using it does not require templates—simply articulate your goals and background, and feel free to experiment and explore. Your real problems and attempts will directly drive our continuous iteration of the "understanding and generation" capabilities.
Co-build and Share
Welcome to share your cases and high-quality prompts. We will give preferential usage quotas to high-quality examples and distill them into "excellent cases" to inspire more users.
Feedback and Support
If you encounter inaccurate understanding or deviation in responses from the Agent, feel free to provide feedback and share your thoughts. Currently, we are still in the intensive debugging phase of the close beta, and your opinions will directly drive the repair and optimization process. Thank you for your participation and support.
Method 1:
Score at the end of the report and add specific improvement suggestions (the more specific, the more helpful)
Method 2:
Like/Dislike a single Topic (the more specific, the more helpful)
Method 3: Contact the Customer Manager
Send us the direct link to the problem conversation (copy the web page URL)
If there are multiple reports in the conversation, indicate which one