Best Practices

Learn proven techniques for creating accurate, reliable analyses that drive better decisions.

Foundation Principles

Start with Clear Objectives

Before building any model, clearly define:

• What decision(s) are you trying to make?
• When do you need to make them?
• What specific questions need answers?
• What level of accuracy is needed?
• Who will use the results and how?

Embrace "Good Enough" Modeling

Perfect models don't exist. Focus on building models that are:

• Useful: Address the key decision
• Understandable: Stakeholders can follow the logic
• Actionable: Lead to clear next steps
• Timely: Available when the decision needs to be made

Model Building Best Practices

Keep It Simple

Start with the simplest model that addresses your question, then add complexity only if needed:

• Begin with 3-5 key Areas of Uncertainty or Decision
• Test simple versions before building complex ones

Validate Your Inputs

Your outputs are only as good as your inputs:

• Use multiple sources: Don't rely on single data points
• Check historical data: How accurate were past estimates?
• Get expert input: Validate assumptions with people who know
• Test extreme values: Make sure your ranges are realistic

Document Everything

Future you (and others) will thank you:

• Record data sources and assumptions
• Explain why you chose specific alternative values for each Area of Uncertainty or Decision as well as the probability assigned to each uncertaintycase
• Note any limitations or concerns
• Keep track of model versions and changes

Avoid the Overconfidence Trap

People tend to be overconfident in their estimates. Be aware of this and try to challenge your assumptions:

• Make ranges wider than your first instinct
• Ask "What could go wrong?" and "What could go better than expected?"
• Challenge assumptions with devil's advocate thinking

Sanity Check

Always test your model:

• Extreme value testing: Set variables to min/max and see if results make sense
• Order of magnitude checks: Are results in the right ballpark?
• Trend analysis: Do results change in expected directions when inputs change?
• Historical comparison: How do results compare to past similar situations?

Sensitivity Testing

Understand what drives your results:

• Identify the most influential variables
• Test different distribution assumptions
• Vary correlation assumptions
• Check robustness to outliers

Communication Best Practices

Lead with Insights

Structure your communication around key insights:

1. Bottom line up front: What's the recommendation?
2. Key insights: What did you learn?
3. Supporting evidence: What analysis supports this?
4. Methodology: How did you do the analysis?

Address Uncertainty Honestly

Don't hide uncertainty - embrace it:

• Show ranges of outcomes and cumuliative likelihoods, not just point estimates
• Discuss key assumptions and their impact
• Acknowledge what you don't know
• Suggest ways to reduce uncertainty if needed

Make It Visual

Use charts effectively:

• Choose chart types that support your message
• Highlight key insights with annotations
• Use consistent colors and formatting
• Avoid chart junk that distracts from the message
• Don't put too many charts in a single slide

Organizational Best Practices

Build Modeling Capability

Create sustainable modeling practices:

• Train multiple people, not just one expert
• Develop standard templates and approaches
• Create review processes for important analyses
• Build a library of past analyses for reference

Establish Governance

For critical decisions, establish clear processes:

• Define who can approve model assumptions
• Require peer review for high-stakes analyses
• Document model validation procedures
• Archive models and results for future reference

Learn from Results

Track how your models perform:

• Compare predictions to actual outcomes when possible with lookback analysis efforts
• Identify patterns in modeling errors
• Update future models based on lessons learned
• Share learnings across the organization

Ethical Considerations

Avoid Bias

Be aware of potential biases:

• Confirmation bias: Looking for results that confirm preconceptions
• Anchoring bias: Being overly influenced by initial estimates
• Availability bias: Overweighting recent or memorable events
• Optimism bias: Being unrealistically positive about outcomes

Consider Stakeholder Impact

Think about who is affected by your analysis:

• Are there groups who might be harmed by the decision?
• Have you considered unintended consequences?
• Are you transparent about limitations and assumptions?
• Do stakeholders understand the uncertainty in your results?

Final Recommendations

Remember that the goal of analysis is better decisions, not perfect models. Focus on:

• Clarity: Make your analysis easy to understand and act on
• Honesty: Be transparent about uncertainty and limitations
• Relevance: Address the actual decision being made
• Timeliness: Deliver insights when they're needed

The best analysis is the one that leads to better decisions. Keep that goal in mind, and you'll create valuable, actionable insights.