Mission: Principal Data Analyst – Find Zava's Hidden Product-Quality Crisis

Mission/open goal Description

Note: I originally created this Promptathon submission in my fork by mistake. I am reposting the same submission here in the official repository so it can be reviewed.

I completed the Principal Data Analyst mission.

The objective was to investigate the PromptathonDb database and identify the highest-risk product-quality cluster by combining multiple sources of evidence instead of relying on a single metric.

The investigation included:

• SalesOrders and SalesOrderLines to measure revenue and quantity sold
• SupportTickets to evaluate support burden and customer satisfaction
• SupportChats.MessagesJson to identify recurring complaint themes
• Docs to analyze customer reviews and support documents
• Vector similarity search using find_similar_docs_by_doc_id to validate whether similar complaints clustered around the same products or categories

The goal was to produce an evidence-based executive recommendation identifying the product or category that should be prioritized for immediate quality investigation.

Harness and model

GitHub Copilot Agent Mode with GPT-5.5

Turn-by-turn journey

1. Prompt: Explore the PromptathonDb schema and identify the relevant entities.
   Agent action: Used SQL MCP tools to inspect Products, SalesOrders, SalesOrderLines, SupportTickets, SupportChats, Docs, Customers, and Employees.
   Result: Identified the data sources required for the investigation.

2. Prompt: Rank products and categories by revenue and quantity sold.
   Agent action: Queried sales data and identified the highest-sales-impact products and categories.
   Result: B2B products dominated revenue, with Premium and Elite apparel also showing significant sales.

3. Prompt: Compare the highest-sales candidates against support tickets.
   Agent action: Analyzed support ticket count, satisfaction, priority, and status.
   Result: High-revenue B2B products showed no meaningful support burden.

4. Prompt: Analyze SupportChats and Docs for recurring complaint themes.
   Agent action: Examined chat messages and document content.
   Result: Identified recurring themes around fabric quality, fit, comfort, and color, primarily affecting Premium and Elite apparel.

5. Prompt: Run vector similarity search using find_similar_docs_by_doc_id.
   Agent action: Selected a representative quality complaint and performed semantic similarity search.
   Result: Found a cluster of similar complaints centered on Premium and Elite apparel sold through the B2C online channel.

6. Prompt: Produce the final executive brief.
   Agent action: Combined evidence from sales, support tickets, chats, documents, and vector search.
   Result: Identified the Premium Short Sleeve Men's Top as the strongest evidence-backed product-quality risk cluster and recommended a targeted quality investigation.

Completion

• Yes, the agent completed the mission or goal.
• No, the agent did not complete the mission or goal.

Bonus work

Beyond completing the mission, I validated the findings step by step instead of accepting the first result. During the investigation, the SQL MCP entity tools encountered naming issues, so I analyzed how the agent adapted by discovering the database connection and querying the live SQL Server directly. I also verified that the final recommendation was supported by multiple independent sources, including sales data, support tickets, chat transcripts, document analysis, and vector similarity search, rather than relying on a single metric.

Start by inspecting the PromptathonDb entities named in the report: SalesOrders, SalesOrderLines, SupportTickets, SupportChats, Docs, and Customers. Run the documented SQL investigation, including find_similar_docs_by_doc_id, and confirm that the completed work combines sales, support, document, chat, and similarity evidence into an executive recommendation.