Glossary: Core Analytics & Data Terms

Glossary: Core Analytics & Data Terms

Use this glossary as a practical, team-friendly reference to align language across analytics, operations, product, and leadership. Each entry gives a concise definition, a short example, and a quick note on when to use the term. This is a starting place for shared vocabulary — adapt terms to your organization's master-data standards and specific practices before embedding them in production systems or formal reports.

How to use this glossary

• Read the short definition to get the core concept.
• Look at the example to see a concrete application in common work contexts.
• Check the “When to use” guidance to decide whether the term fits your decision, report, or analysis.
• For policy, compliance, or formal measurement work, adapt wording to local standards and document the authoritative definition you will use.

Entries

KPI (Key Performance Indicator)

Definition: A metric directly tied to a strategic objective and used to track progress toward that objective.
Example: Monthly active users (MAU) for a subscription product when growth is a top business goal.
When to use: Choose KPIs when you need a small, prioritized set of measures for leadership reporting or strategic decisions. Distinguish KPIs from broader metrics by documenting the objective they support.

Metric

Definition: Any quantifiable measure of a system, process, or outcome. Metrics are the raw building blocks of KPIs and analysis.
Example: Page load time, number of support tickets, or defect rate per batch.
When to use: Use metrics for monitoring, diagnostics, and building dashboards. Not every metric is a KPI — clarify role and target before elevating a metric to KPI status.

Cohort

Definition: A group of users, units, or observations that share a defining characteristic or experience within a defined time frame.
Example: Users who signed up in January 2026 or machines commissioned in Q3 2024.
When to use: Use cohort analysis to compare behavior or outcomes across groups over time (e.g., retention by signup month).

Drift

Definition: A change over time in the statistical distribution of data; common types include feature drift (input distribution changes) and label drift (target distribution changes).
Example: A new product layout causes a sustained change in click patterns compared with historic data.
When to use: Monitor drift to detect when models, alerts, or dashboards might no longer reflect current reality and need retraining, recalibration, or investigation.

Model drift

Definition: A decline in a model's predictive performance over time, typically caused by data drift, concept drift, or changes in the environment.
Example: A fraud-detection model's precision drops after fraudsters adopt new tactics.
When to use: Implement model performance monitoring and set thresholds that trigger review, retraining, or rollback when drift is detected.

p-value

Definition: In hypothesis testing, the p-value is the probability of observing data at least as extreme as the sample given that the null hypothesis is true.
Example: A/B test yields p = 0.03 for conversion difference; under a standard threshold, the result is considered statistically significant.
When to use: Use p-values as part of formal hypothesis testing, but not as the sole evidence. Report effect sizes, confidence intervals, and check assumptions (randomization, independence) before drawing conclusions.

Confidence interval

Definition: A range of values, derived from sample data, that is believed—at a stated confidence level (e.g., 95%)—to contain the true population parameter.
Example: A 95% confidence interval for mean revenue per customer might be [$45, $55].
When to use: Present confidence intervals alongside point estimates to communicate uncertainty and practical significance to decision makers.

ETL / ELT

Definition: ETL = Extract, Transform, Load (transform data before loading into target). ELT = Extract, Load, Transform (load raw data to a target first, then transform). Choice depends on architecture and processing needs.
Example: ETL used when staging cleansed data for legacy data warehouses; ELT used with modern cloud data lakes where transformations run on scalable compute.
When to use: Choose ETL or ELT based on data volume, latency requirements, tooling, and governance. Document which pattern you adopt for each pipeline.

Event

Definition: A recorded occurrence or action at a point in time (e.g., click, transaction, sensor reading).
Example: 'Add to cart' event logged with user ID, timestamp, and product ID.
When to use: Events are useful for behavioral analytics, funnels, and time-series analysis. Define event schemas and consistent naming for reliable analytics.

Feature

Definition: An input variable used by a model or analysis. Features may be raw (e.g., age) or engineered (e.g., rolling 7-day average of spend).
Example: Customer tenure in months used as a feature for churn prediction.
When to use: Track feature lineage, data quality, and freshness. Maintain a feature catalog to avoid duplicated or inconsistent calculations.

Data product

Definition: A packaged data capability designed for reuse by others—this can be a cleaned dataset, an API, a model, or a dashboard with defined SLAs and documentation.
Example: A customer-360 API that returns unified customer attributes for use in marketing, support, and analytics.
When to use: Treat reusable analytics assets as products when they serve multiple stakeholders; include ownership, versioning, and documentation.

OEE (Overall Equipment Effectiveness)

Definition: A composite manufacturing metric equal to Availability × Performance × Quality; used to measure the productive time of equipment relative to ideal.
Example: If a machine runs 90% of scheduled time (Availability), performs at 95% speed (Performance), and produces 98% good parts (Quality), OEE = 0.90×0.95×0.98 ≈ 83.7%.
When to use: Use OEE to summarize equipment effectiveness, but analyze its components separately to find improvement opportunities.

Leading indicator

Definition: A metric that tends to change before a target outcome and can be used to predict future performance.
Example: Number of qualified leads as a leading indicator for future sales revenue.
When to use: Monitor leading indicators to take preventive or proactive actions. Validate predictive relationship statistically where possible.

Lagging indicator

Definition: A metric that reflects outcomes after they have occurred; useful for verifying results but not for predicting them.
Example: Quarterly revenue or defect rate reported after production.
When to use: Use lagging indicators for accountability, trend confirmation, and evaluation of past initiatives.

Notes, cautions, and next steps

This glossary is a practical team reference, not a substitute for domain-specific measurement standards or formal statistical consulting. When you build dashboards, reports, or automated systems, document the authoritative definitions you will use (naming, calculation, unit, aggregation window, and owner). Consider creating a central data glossary or feature catalog that your teams can tailor and version.