Module 5 — Decision Frameworks & Ethics in Analytics

Decision trees provide a visual framework for evaluating options under uncertainty. Each branch represents a possible decision or outcome, with probabilities and payoffs assigned to each path. The expected value of each option is calculated by multiplying each outcome's probability by its payoff and summing the results.

For example, a firm deciding whether to launch a new product might face: 60% chance of success (profit £500K) and 40% chance of failure (loss £200K). Expected value = (0.6 × 500K) + (0.4 × -200K) = £220K. Compare this to the expected value of not launching (£0) to make the decision.

Multi-Criteria Decision Analysis (MCDA) extends beyond financial considerations. When decisions involve multiple, sometimes conflicting objectives (cost, quality, speed, risk, environmental impact), MCDA provides a structured approach: identify criteria, assign weights reflecting their relative importance, score each option against each criterion, and calculate weighted scores to rank alternatives.

MCDA is particularly valuable for complex strategic decisions where pure financial analysis is insufficient — technology vendor selection, location decisions, partnership evaluations — because it makes trade-offs explicit and transparent.

As analytics becomes more pervasive, ethical considerations become critical. Key issues include:

Privacy: GDPR and similar regulations give individuals rights over their data — the right to access, correct, delete, and restrict processing. Analytics professionals must ensure data collection is lawful, transparent, and proportionate.

Algorithmic bias: AI and statistical models can perpetuate or amplify existing biases. Hiring algorithms trained on historical data may discriminate against minorities; credit scoring models may disadvantage certain communities. Bias can enter through biased training data, biased feature selection, or biased interpretation of results.

Transparency: Stakeholders should understand how data-driven decisions are made, especially when those decisions significantly affect people's lives. The "black box" problem of complex ML models raises questions about accountability.

Communicating uncertainty is an ethical obligation. Every prediction has a margin of error; every model has limitations. Presenting analytics results without acknowledging uncertainty is misleading. Best practices: always include confidence intervals, state assumptions explicitly, highlight limitations, and present multiple scenarios rather than single-point forecasts.

The responsible analyst asks: Is the data collected ethically? Are there biases in the data or model? Who benefits and who is harmed by this analysis? Have we communicated uncertainty honestly?