Ask Question

The Ask a question feature is the primary entry point for interacting with structured data in natural language via Connecty AI. This functionality allows users to pose questions related to their databases and receive context-aware answers, generated through multi-layer reasoning.

---

Prerequisites

Before asking a question, ensure the following prerequisites are met:

Select a Workspace

A data source must be connected and authorized. This determines the scope of available tables, columns, and metrics

Read about how to connect and setup your data sources in the “Getting Started’ section

Choose an AI Agent (Coming Soon)

Connecty AI offers multiple role-specific agents. Each focuses on a different layer of the data analysis stack:

• 🟣 Analyze Agent – performs natural language to SQL generation with step-by-step logic.

• 🟠 Discover Agent – explores metadata, semantic relationships, PII detection, and lineage mapping.

These agents can be selected depending on your goal — e.g., ask the Analyze Agent to calculate metrics, or the Discover Agent to explore the structure of your data.

For selecting an agent, click the forward slash button (/) in the dialog box:

Select a Reasoning Type

• Simple Mode: Ideal for direct lookups or aggregations. This mode is optimized for speed and simplicity, making it suitable for straightforward use cases like metric summaries, basic filters, or one-table queries.

• Deep Reasoning Mode: Recommended for multi-step, explainable reasoning involving joins, metric definitions, or hierarchy traversal. This mode breaks complex questions into sub-intents that are logically connected, processed in sequence, and individually explained. It supports advanced queries such as comparisons over time, root cause analysis, and drilldowns across multiple dimensions.

---

How to Formulate Questions Correctly

Formulating clear and meaningful questions increases the accuracy and value of the AI-generated output.

Best Practices

• Be specific: Use concrete business terms (e.g., revenue, orders, regions, last month)

• One objective per question (Especially in Simple Mode) — e.g., “Show average cart size by day”

• Use natural language: Don’t write SQL — say what you need in plain English

• Specify filters: Add timeframes, segments, or regions to narrow scope

• Use known metrics/dimensions: Ensure the term exists in your connected catalog

• For complex logic, use Deep Reasoning Mode to get step-by-step answers

Question Types (Do's and Don'ts)

Supported Question Types	Unsupported Question Types
✅ Metric Lookups “What is the total revenue in Q1 2024?”	❌ External KPIs (not in your database) “How much is the GDP of the USA?”
✅ Time-Based Aggregations “Show sales by month this year”	❌ Real-Time World Data “What’s the weather like today?”
✅ Top/Bottom Filters “Show top 10 products by revenue”	❌ Industry Benchmarks “How does my sales compare to the industry average?” (unless stored in your own data)
✅ Comparisons “Compare revenue this year vs last year”	❌ Access Limitations / System Questions “Why aren’t you giving me access to external data?”
✅ Anomaly or Trend Detection “Identify top 3 churn spikes in 2023”	❌ Social Prompts “How are you today, Connecty AI?”
✅ Customer Segmentation “List high-value customers by region”	❌ Live Market Data “Tell me the price of Tesla stock”
✅ Funnel/Drop-off Analysis “What is the drop off rate in the onboarding funnel from trial to conversion page?”	❌ UI Troubleshooting “Why is this question not working?”
✅ Drilldowns “Break down revenue by campaign grouped by region”	❌ General Business Advice “Can you help me write a business plan?”
✅ Metric Definitions “What is Customer Lifetime Value?”	❌ Cross-Workspace Data Access “Give me data from another data workspace”
✅ Filters “Only show customers from Germany”	❌ Sentiment/Opinion Analysis (outside DB) “What do people think of our pricing model?”

Pro tip: Deep Reasoning Mode Support

When activated, Deep Reasoning Mode allows you to ask complex or multi-part questions that will be internally decomposed into sub-tasks or “sub-intents.”

More examples from TPCH dataset

Topic	Metric Question
Supplier analytics	Average fulfillment time by supplier and customer nation for parts from Asia region suppliers grouped by part type
Supplier analytics	Gross margin per part category from suppliers who shipped parts with discount over 5%
Supplier analytics	Late shipment rate by supplier region and part size category grouped by nation and year
Supplier analytics	Total number of parts supplied by high-value-order suppliers segmented by part brand
Supplier analytics	Profit margin from parts supplied by top 10 European suppliers segmented by customer segment
Customer analytics	Average order value per part category for 'Automotive' segment customers ordering from European suppliers
Customer analytics	Total profit per supplier from orders by customers who only ordered in the last 3 years
Customer analytics	Average order value by part type for delayed shipments to Asian customers ordering German parts
Customer analytics	Number of unique suppliers per customer segment in orders containing discounted parts
Customer analytics	Customer lifetime discount amount grouped by region
Order analytics	Average line item count per part category in orders from 1995 with Japanese suppliers and European customers
Order analytics	Total discount amount by supplier nation for orders over $20,000 with average line item count analysis
Order analytics	Average part cost grouped by region in multi-supplier orders with at least 3 suppliers and higher cancellation rates
Order analytics	Supplier count per order for fast-shipped orders (under 10 days) containing at least 5 distinct parts with average part cost analysis
Order analytics	Gross margin segmented by part brand for orders with line items sourced from multiple regions
First/last order analytics	Total number of suppliers engaged by customers active for over 5 years and their first-to-last order time gap
First/last order analytics	Average line item discount in first orders per customer segment by supplier region
First/last order analytics	Total part cost grouped by part category in last orders for customers with decreasing order frequency and high discount rates
First/last order analytics	Difference in gross margin and order value between first and last orders per supplier segmented by customer segment
First/last order analytics	Average part discount grouped by part type in first orders from new suppliers in each region