Introduction

TODO: 2-3 paragraph introduction explaining what the Analytics Decision Framework is and the problem it solves. Most analytics investments produce dashboards that get ignored. The root cause: teams build analytics without first deciding which decisions the analytics should improve. This framework starts with decisions, not data.

TODO: Second paragraph covering the four analytics types — descriptive (what happened?), diagnostic (why did it happen?), predictive (what will happen?), and prescriptive (what should we do?) — and the decision-making contexts where each is appropriate. The framework maps business decisions to the right analytics type before any data or tooling work begins.

TODO: Third paragraph on practical application — use this framework in a discovery session with business decision-makers before starting any analytics project. The output is a prioritized map of decisions, their required analytics type, data requirements, and estimated value. This map drives the analytics roadmap.

Step 1: Define the Decision

TODO: Overview of Step 1 — before choosing an analytics approach, the decision must be precisely defined. This is the most frequently skipped step and the most common cause of analytics investments that produce no behavior change.

Identify the Decision Maker

TODO: Every analytics initiative must have a named decision-maker who will use the output to make different decisions than they currently make. Anonymous decision-makers produce shelf-ware dashboards. Name the person, their role, and the specific decision they own.

Specify the Decision

TODO: A specific decision has: a clear trigger (when is the decision made?), a finite set of options (what choices are available?), and a defined outcome (what changes as a result of the decision?). Walk through how to take a vague request ("we need better analytics") to a precise decision specification.

Quantify the Decision Value

TODO: Estimate the value of making this decision better. Cover: how often is this decision made? What is the cost of a wrong decision? How much better would analytics make this decision? The value estimate justifies the analytics investment and sets expectations for what improvement is realistic.

Step 2: Choose the Analytics Type

TODO: Overview of Step 2 — mapping the defined decision to the appropriate analytics approach. Each type has different data requirements, technical complexity, and decision-making value.

TODO: Descriptive analytics — use when the decision-maker needs to understand current or historical state. Required data: historical records. Complexity: low. Best for: operational monitoring, performance reviews, status reporting.
TODO: Diagnostic analytics — use when the decision-maker needs to understand why something happened. Required data: multi-dimensional historical data with sufficient granularity for drill-down. Complexity: medium. Best for: root cause analysis, exception investigation.
TODO: Predictive analytics — use when the decision-maker needs to anticipate future outcomes. Required data: labeled historical examples with outcome data. Complexity: high. Best for: demand forecasting, churn prediction, risk scoring.
TODO: Prescriptive analytics — use when the decision-maker needs to know what action to take. Required data: outcome data with action history and counterfactuals. Complexity: very high. Best for: pricing optimization, resource allocation, treatment selection.

Step 3: Assess Data Requirements

TODO: Overview of Step 3 — verifying that the data needed for the chosen analytics type exists, is accessible, and has sufficient quality.

Map Required Data to Sources

TODO: For each analytics type, identify the specific data elements required. Trace each element back to its source system. Identify gaps between what's required and what exists today. Estimate the effort to close each gap.

Evaluate Data Quality

TODO: Analytics quality is bounded by data quality. For each required data element: assess completeness (what percentage of records have this field populated?), accuracy (is the value correct?), and timeliness (how old is the data when it reaches the analytics layer?). Document quality issues that will constrain analytics reliability.

Step 4: Define Success Metrics

TODO: Overview of Step 4 — establishing how you'll know whether the analytics investment improved decision-making. Most analytics projects are never evaluated against their goals.

Set Decision Quality Metrics

TODO: Define measurable proxies for decision quality improvement. Examples: reduction in time to decision, reduction in decision reversal rate, improvement in outcome metrics correlated with the decision (e.g., churn rate for a churn intervention decision). These metrics are established before the analytics project begins, not after.

Establish Adoption Benchmarks

TODO: Analytics that isn't used doesn't improve decisions. Define minimum adoption metrics: percentage of decisions made using the analytics output, frequency of dashboard access, user satisfaction score. Low adoption is an early warning sign that the analytics doesn't match the decision-maker's actual workflow.

Analytics Type Reference

Level 1Descriptive

TODO: Answers 'what happened?' Using aggregation, summarization, and visualization of historical data. Examples: revenue by region last quarter, daily active users, defect rate by production line. Decision context: monitoring, review, reporting. Data requirement: clean historical records. Build-vs-buy: BI tools (Tableau, Looker, PowerBI) are usually sufficient.

Level 2Diagnostic

TODO: Answers 'why did it happen?' Using drill-down, segmentation, and correlation analysis. Examples: why did churn spike in the northeast region? Why did conversion drop after the checkout redesign? Decision context: root cause analysis, investigation. Data requirement: multi-dimensional data with sufficient granularity. Build-vs-buy: BI tools with strong drill-down capabilities.

Level 3Predictive

TODO: Answers 'what will happen?' Using statistical models and machine learning on historical labeled data. Examples: which customers are likely to churn next month? What will demand be next quarter? Decision context: anticipating outcomes to take proactive action. Data requirement: historical labeled examples with outcome data. Build-vs-buy: usually requires ML engineering; managed platforms (SageMaker, Vertex) reduce infrastructure burden.

Level 4Prescriptive

TODO: Answers 'what should we do?' Using optimization, simulation, or reinforcement learning to identify the best action. Examples: what price should we set for this product to maximize margin? How should we allocate maintenance resources across facilities? Decision context: complex optimization with multiple constraints. Data requirement: outcome data with action history. Build-vs-buy: advanced capability usually requiring specialist ML engineering.

Common Failure Patterns

TODO: Building predictive analytics when descriptive is sufficient — complex models where a simple dashboard would suffice, creating unnecessary maintenance burden
TODO: Analytics without a decision-maker — 'we should have this data' requests that produce dashboards nobody uses
TODO: Data quality ignored until delivery — discovering data is incomplete or inaccurate after significant analytics investment
TODO: No adoption plan — analytics built without change management, onboarding, or workflow integration
TODO: Success undefined — no baseline metrics established before the project, making it impossible to demonstrate value
TODO: Wrong analytics type — attempting predictive analytics before descriptive and diagnostic analytics are working reliably

Next Steps

TODO: Guidance on applying the framework. Suggest starting with a decision mapping workshop (1 day, 4-8 decision-makers across the organization) to produce a prioritized analytics roadmap. Link to the Analytics service page for organizations that want Xephyr to facilitate the workshop or deliver the analytics roadmap.

Analytics Decision Framework