# Cassie Kozyrkov — Decision Science, Data Strategy & AI Adoption

## Snapshot
- Founded Decision Intelligence discipline at Google, served as Chief Decision Scientist
- Trained 17,000+ Google employees in applied statistics, ML, and decision-making
- Background: statistics, economics, mathematics, psychology (U of Chicago, Duke)
- Domain: data-driven decision-making, bridging data science and business

## Core Philosophy
- **Decision intelligence**: Decisions should be engineered products, not afterthoughts. The biggest gap isn't data — it's clarity about what decisions to make.
- **Bayesian reasoning**: Update beliefs based on evidence. Decisions are bets on incomplete info. Mental agility — willingness to revise beliefs — is a competitive advantage.
- **Signaling theory**: Every decision sends messages to stakeholders. Leaders unknowingly create misaligned signals if they ignore symbolic impact.
- **Trade-off thinking**: No perfect decisions. Articulate constraints clearly. Distinguish "must-have" from "nice-to-have." Make trade-offs visible.
- **Human-centric**: Psychological factors (biases, emotions, storytelling) must integrate into analysis. Decision tools should empower people, not replace them.

## Key Frameworks

### Decision Workflow
1. **Frame the question** — What decision? Who owns it? What does success look like?
2. **Design the decision** — Stakeholders, constraints, trade-offs. Map possible outcomes.
3. **Collect evidence** — Gather relevant data. Interpret probabilistically.
4. **Update beliefs** — Revise assumptions with mental agility. Avoid binary thinking.
5. **Decide & communicate** — Choose best option. Communicate reasoning transparently, including trade-offs.
6. **Monitor & iterate** — Track outcomes, learn, adapt.

Most analytics projects fail because the question is ill-defined.

### Bayesian Decision-Making
1. Define prior beliefs (existing assumptions)
2. Gather evidence (experiments, surveys, research)
3. Update beliefs (adjust probability based on evidence)
4. Decide (highest expected value given updated probabilities)

**Apply**: Whenever decisions involve uncertainty and evolving evidence.
**Skip**: Routine operational decisions with clear deterministic outcomes.

### Signaling Framework
1. Identify stakeholders who will interpret the decision
2. List potential signals each group might infer
3. Align signals with company values and mission
4. Communicate explicitly to prevent misinterpretation

**Apply**: Decisions affecting brand perception (price changes, partnerships, promotions).
**Skip**: Purely internal operational decisions with no external visibility.

### Trade-Off Design Canvas
Columns: Objectives | Constraints | Options | Costs | Benefits
- Forces articulation of goals and must-have criteria
- Reveals implicit trade-offs
- Encourages prioritization debate

**Apply**: Strategic decisions (supplier selection, product design, marketing strategy).

### Decision-Ready Data Pipeline
1. Identify the decision question the data will inform
2. Specify success metrics and measurement methods
3. Ensure data quality (clean, validate)
4. Create dashboards that support probabilistic decision-making

## Tactical Application
- **Strategy**: Frame business questions precisely before collecting data
- **Marketing**: Treat hypotheses as probabilistic. A/B test. Update based on results.
- **Product**: Use trade-off canvas for formulation decisions (cost vs. quality vs. shelf life)
- **AI adoption**: Design data pipelines that produce decision-ready insights
- **Culture**: Build a culture where acknowledging costs and limitations is strength

## Warnings
- Data without a decision question is waste
- Binary thinking kills nuance — treat decisions as probability ranges
- Ignoring signaling creates brand/culture misalignment
- Perfect decisions don't exist — optimizing one objective costs another