How Scoop Investigates

Purpose: Understand how Scoop's "Digital Data Analyst" thinks so you can ask better questions. Audience: Business Users, Analysts

The "Digital Data Analyst" Difference

Unlike standard BI tools that just run the SQL query you ask for, Scoop acts like a human analyst. It breaks your question down, formulates hypotheses, checks them, and synthesizes an answer.

The 3-Phase Investigation Process

When you ask a question like "Why is Store 42 underperforming?", Scoop doesn't just look up "Store 42 performance." It follows this process:

Phase 1: Understanding & Planning

Scoop first decides what to investigate.

Context: It identifies that this is a "Financial" and "Operational" question.
Entities: It locks onto "Store 42" as the subject.
Time: It sets a relevant time period (e.g., "Last 14 days" vs "Previous Period") based on your question.

Phase 2: The Investigation Loop

This is where Scoop acts like a detective. It runs multiple rounds of queries:

Round 1 (Broad Scan): "Show me revenue, transaction count, and margin for Store 42."
- Finding: Revenue is down 22%, but Margin is stable.
Round 2 (Hypothesis Testing): Scoop notices the revenue drop and spawns new questions automatically:
- Hypothesis A: "Is it a district-wide issue?" (Checks other stores).
- Hypothesis B: "Is it a competitive issue?" (Checks for new competitor openings).
Round 3 (Drilling): It finds a new competitor opened nearby. It drills down to confirm the timing matches the revenue drop.

Note: Scoop uses "Bounded Exploration." It limits itself to ~4 levels of depth to ensure it gives you an answer quickly rather than investigating forever.

Phase 3: Synthesis

Scoop collects all the findings—both the "Smoking Guns" and the "Dead Ends"—and writes a summary.

Result: "Store 42's revenue decline is likely caused by the new competitor X that opened 2 weeks ago. This is an isolated incident; other stores in the district are healthy."

How to Help Scoop Help You

Because Scoop thinks like an analyst, you can guide it with "Executive Hints" in your prompt.

1. Give Context, Not Just Keywords

Bad: "Store 42 revenue"
Good: "Store 42 revenue is down. Investigate if this is an operational issue or external market factor."
- Why: This cues Scoop to load the "Operational" and "Competitive" investigation playbooks immediately.

2. Compare to Baselines

Bad: "Show me sales."
Good: "Show me sales compared to the district average."
- Why: Comparisons allow Scoop to detect anomalies (e.g., "This store is 20% below average") which triggers deeper investigation.

3. Ask "Why", Not Just "What"

Bad: "List top 10 products."
Good: "Why are these top 10 products selling better than the others?"
- Why: This triggers a "Driver Analysis" where Scoop looks for correlations (e.g., "Top products are all discounted 15%").

Key Concepts

"Drilling"

Scoop automatically "drills" into data. If it sees a spike in "Total Costs," it will automatically break that down by "Cost Category" (Labor, Goods, Rent) to find the driver. You don't need to ask for the drill-down; Scoop does it to explain the top-level number.

"Entity Focus"

Scoop investigates one "primary" entity at a time to stay focused.

Single Mode: Deep dive on one store/manager.
Group Mode: Pattern finding across many stores/managers.
- Tip: Don't mix them. Ask "Analyze Store 42" (Single) or "Analyze all Mall Stores" (Group). Don't ask "Analyze Store 42 and all Mall Stores" in one breath; it's harder to synthesize.

"Freeform" vs "Playbook"

Playbook: For common questions ("Why is revenue down?"), Scoop follows a structured, expert-defined path.
Freeform: For wild questions ("Does rain affect bagel sales?"), Scoop improvises a new investigation path on the fly. It is flexible!