20 Questions

Questions to Ask When Analyzing Data

A practical checklist of questions for analysts and decision-makers to define the problem, validate data, choose methods, and translate findings into clear actions.

1

What decision will this analysis inform—and who needs to make it?

Click to see why this works

Why this works

Anchors the work to a real decision owner and prevents aimless exploration.

2

What is the minimal useful output (metric, threshold, ranking)?

Click to see why this works

Why this works

Clarifies deliverables so scope, method, and timelines match the business need.

3

How was the data generated and what biases could it contain?

Click to see why this works

Why this works

Surfaces selection, measurement, and survivorship biases that distort conclusions.

4

What are the unit of analysis and time windows?

Click to see why this works

Why this works

Avoids aggregation errors and mismatched denominators that mislead stakeholders.

5

Where are missing values, outliers, and duplicates—and why?

Click to see why this works

Why this works

Forces data quality triage before modeling, improving reliability of results.

6

What baseline or prior should we compare against?

Click to see why this works

Why this works

Frames results relative to expectations or controls, not in isolation.

7

What assumptions does our method require, and do they hold?

Click to see why this works

Why this works

Prevents invalid inference by checking normality, independence, stationarity, etc.

8

What are plausible alternative explanations?

Click to see why this works

Why this works

Encourages counterfactual thinking and reduces false certainty in causal claims.

9

What sensitivity checks or ablations will we run?

Click to see why this works

Why this works

Tests robustness by varying inputs, features, or time slices.

10

What’s the simplest model that would be good enough?

Click to see why this works

Why this works

Right-sizes complexity to interpretability and deployment costs.

11

How will we validate results (holdout, cross-validation, backtesting)?

Click to see why this works

Why this works

Ensures generalization and guards against overfitting to historical noise.

12

What error bars, confidence intervals, or prediction intervals matter here?

Click to see why this works

Why this works

Communicates uncertainty transparently for risk-aware decisions.

13

If we’re wrong, how will it fail—and what’s the impact?

Click to see why this works

Why this works

Focuses on downside risk and mitigation plans upfront.

14

How will we make this explainable to non-technical stakeholders?

Click to see why this works

Why this works

Prioritizes simple narratives, visuals, and decision-ready artifacts.

15

What decision thresholds trigger action—and who owns the next step?

Click to see why this works

Why this works

Turns findings into a playbook with clear responsibilities.

16

What data would most improve this analysis next time?

Click to see why this works

Why this works

Creates a learning loop that compounds value across projects.

17

What are the privacy, security, or ethical considerations?

Click to see why this works

Why this works

Reduces legal and reputational risk by designing for responsible use.

18

What’s the ROI of acting on these insights?

Click to see why this works

Why this works

Frames impact in dollars, hours, or risk reduction to prioritize execution.

19

How will we monitor drift or model decay over time?

Click to see why this works

Why this works

Plans for maintenance, alerts, and retraining to preserve performance.

20

What is the one-slide summary a leader needs to say yes?

Click to see why this works

Why this works

Forces ruthless synthesis so the recommendation is easy to approve.

From Question to Decision: A Data Analysis Playbook

Expert tips and techniques for getting the most out of these questions.

Analysis That Drives Action

Start With the Decision

Define decision, owner, and deadline before opening a notebook.

Make Assumptions Explicit

List and test assumptions; document what breaks if they fail.

Quantify Uncertainty

Leaders trust ranges more than point estimates—show your error bars.

Stakeholder-Ready Artifacts

Decision Memo Template

1
Context
2
Recommendation
3
Evidence
4
Risks
5
Next Steps

Common Pitfalls

Analysis Paralysis

Timebox exploration; ship the minimal useful answer first.

Causal Overreach

Avoid claiming causality without design or instruments to support it.

End-to-End Workflow

Five-Step Loop

1
Step 1: Define decision
2
Step 2: Audit data
3
Step 3: Choose method
4
Step 4: Validate & stress test
5
Step 5: Synthesize & handoff

Further Reading

The Signal and the Noise by Nate Silver
Validating Statistical Models (various sources)

Find Your Perfect Questions

Search our collection of thoughtful questions for any conversation or situation