Decision Intelligence
A framework that combines data science, social science, and managerial science to turn data into better decisions at scale β augmented by AI and GenAI.
π What is Decision Intelligence?
Decision Intelligence (DI) is an engineering discipline that augments human decision-making with data, analytics, and AI. Coined by Cassie Kozyrkov (Google's former Chief Decision Scientist), it bridges the gap between raw data and actionable business outcomes by providing frameworks to make better, faster, and more reliable decisions.
Decision Framing
Define the right question
Data Analysis
Evidence-based insights
AI Augmentation
Models + reasoning
Action
Outcomes, not reports
π‘ Key Insight: Decision Intelligence is not about replacing human judgment β it's about augmenting decisions with data and AI so decision-makers can act with greater confidence and speed.
π Types of Decisions
β‘ Type 1: Automated Decisions
High-volume, low-risk decisions that can be fully automated by AI/ML models.
- Fraud detection alerts, email spam filtering
- Dynamic pricing adjustments
- Content recommendations, ad targeting
π€ Type 2: Augmented Decisions
AI provides recommendations and analysis; humans make the final call.
- Hiring decisions with AI-screened candidates
- Medical diagnosis with AI-assisted imaging
- Investment decisions with GenAI market analysis
π§ Type 3: Strategic Decisions
High-stakes, low-frequency decisions requiring deep context and human judgment.
- Mergers & acquisitions
- Market entry strategies
- Organizational restructuring, pivots
ποΈ Decision Intelligence Framework
A structured approach to transform data into decisions:
Frame the Decision
Define what decision needs to be made, who the decision-maker is, what actions are available, and what outcomes are desired. This is the most critical and most often skipped step.
Gather & Analyze Data
Collect relevant data, apply statistical analysis, and build predictive models. Use GenAI for unstructured data analysis (documents, emails, conversations).
Model Decision Options
Create causal models, decision trees, or simulation models to understand the impact of different choices. Use LLMs to generate and evaluate scenarios.
Decide & Act
Make the decision with confidence. For automated decisions, deploy the model; for augmented decisions, present insights to the decision-maker.
Measure & Learn
Track the outcomes of your decisions. Create feedback loops to improve future decision quality. Monitor for bias and drift.
π€ GenAI in Decision Intelligence
Large Language Models transform Decision Intelligence by making data analysis conversational and accessible to non-technical stakeholders:
π£οΈ Conversational Analytics
Ask questions in natural language: "What caused the drop in Q3 revenue?" and get AI-generated analysis with supporting data visualizations.
π Scenario Simulation
Use LLMs to generate "what-if" scenarios: "What happens if we increase prices by 10%?" with multi-variable impact analysis.
π Document Intelligence
Extract decision-relevant information from contracts, reports, emails, and meeting transcripts using RAG pipelines.
βοΈ Bias Detection
Use AI to identify cognitive biases in decision-making: anchoring bias, confirmation bias, sunk cost fallacy, and groupthink.
π» Example: Decision Support with LLM
from openai import OpenAI
client = OpenAI()
def analyze_decision(context: str, options: list[str]) -> str:
"""Use an LLM to analyze a business decision."""
prompt = f"""
You are a Decision Intelligence analyst. Analyze this decision:
Context: {context}
Options:
{chr(10).join(f'- {opt}' for opt in options)}
For each option, provide:
1. Pros and Cons
2. Risk assessment (Low/Medium/High)
3. Expected outcome
4. Confidence level
5. Recommended action with justification
"""
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": prompt}],
temperature=0.3 # Low temperature for analytical tasks
)
return response.choices[0].message.content
# Example usage
result = analyze_decision(
context="Our SaaS product has 10K users. Growth has plateaued at 2% MoM.",
options=[
"Launch a freemium tier to increase top-of-funnel",
"Invest in enterprise sales team",
"Build AI features to differentiate from competitors",
"Expand to new geographic markets"
]
)
print(result)
π οΈ Key Techniques
| Technique | Purpose | AI Role | Example |
|---|---|---|---|
| Causal Inference | Understand cause & effect | Model relationships | Did the campaign cause sales increase? |
| Decision Trees | Map decision paths | Auto-generate branches | Approval workflow routing |
| Monte Carlo Simulation | Quantify uncertainty | Run thousands of scenarios | Project ROI under uncertainty |
| Multi-Criteria Analysis | Weighted scoring | Optimize trade-offs | Vendor selection, hiring |
| A/B Testing | Experimentation | Statistical significance | Feature rollout decisions |
| Reinforcement Learning | Sequential decisions | Adaptive optimization | Dynamic pricing, inventory |
π― Industry Applications
Financial Services
Credit scoring, fraud detection, portfolio optimization, risk assessment.
Healthcare
Treatment selection, resource allocation, clinical trial design, diagnosis support.
Retail
Inventory management, pricing strategy, store location, assortment planning.
Manufacturing
Predictive maintenance, supply chain optimization, quality control decisions.
Marketing
Campaign optimization, budget allocation, channel selection, customer segmentation.
HR & People
Talent acquisition, retention risk, workforce planning, compensation strategy.
π§ Tools & Platforms
| Tool | Type | GenAI Features | Best For |
|---|---|---|---|
| Google Looker | BI + Analytics | β Gemini AI | Data exploration, dashboards |
| Palantir Foundry | Decision Platform | β AIP | Enterprise decision ops |
| Databricks | Data + AI | β GenAI apps | ML pipelines, lakehouse |
| Tableau (Salesforce) | Visualization | β Einstein AI | Visual analytics, storytelling |
| Power BI (Microsoft) | BI + Analytics | β Copilot | Microsoft ecosystem, self-service BI |
β Best Practices
Do's
- Start with the decision, not the data
- Define success criteria before analysis
- Include diverse perspectives to reduce bias
- Create feedback loops to learn from outcomes
- Document decision rationale for accountability
Don'ts
- Let data analysis replace decision framing
- Blindly trust AI outputs without validation
- Ignore uncertainty and present false precision
- Automate high-stakes decisions without human oversight
- Confuse correlation with causation
π Learn More
Related Topics
Test Your Knowledge
Score 8/10 or higher to pass
You need to be logged in to take this quiz.
Login to Continue