Data Analytics Glossary — Key Terms Explained Simply

Recorded facts — numbers, text, dates, images. In analytics, "data" usually means structured data: tables of rows and columns, like a spreadsheet or a SQL table.

A collection of data — often a single table (e.g., "orders") or a few related tables (orders, customers, products). "Dataset" and "table" are often used interchangeably in casual conversation, but a dataset can be a whole collection.

Data organised as rows and columns. Each row is one record (one order, one user, one event). Each column is one field (name, price, date). In SQL, tables are the fundamental structure.

The kind of value a column holds: integer (whole number), float (decimal), string (text), date, boolean (true/false). Types matter because they determine what you can do with the data — you can't SUM a text column.

The structure of a database — the tables, their columns, their types, and how they relate. "Reading a schema" means understanding the shape of the data before you query it. Good analysts always start here.

A primary key uniquely identifies a row (e.g., user_id). A foreign key references a primary key in another table (e.g., orders.user_id → users.id). They're how tables relate to each other, and how JOINs work.

A system that stores and serves data — usually with querying, transactions, access controls. Common ones: PostgreSQL, MySQL, SQL Server, Oracle for relational; MongoDB, Firestore for document stores. Most analytics work happens on relational databases via SQL.

A database optimised for analysis rather than transactions — e.g., Snowflake, BigQuery, Redshift. Warehouses store cleaned, joined data from many sources, so analysts can query efficiently without loading production systems. The warehouse is usually where "the data team" actually works.

The language of relational databases. You write queries to filter, join, group and calculate data. Even if you use a BI tool or Python, SQL is under the hood — and most analytics job descriptions ask for it explicitly.

The building blocks of a query. SELECT picks columns, WHERE filters rows, GROUP BY aggregates (e.g., "revenue per month"). A SQL query is essentially: pick these columns, from this table, filter this way, group by this, order by this.

Combining two tables on a related column. "Get every order, plus the user's name" requires joining the orders table to the users table. Joins are why foreign keys exist, and why understanding the schema matters — bad joins produce wrong numbers fast.

Summarising data across rows: SUM, COUNT, AVG, MIN, MAX. "Revenue by month" is an aggregation — sum the revenue column, group by month. Almost every business metric is an aggregation.

A named subquery — a way to break a complex query into readable steps. Written with WITH. Modern analytics SQL is mostly chains of CTEs, each doing one thing. If you're nesting subqueries five levels deep, you need CTEs.

An aggregation that doesn't collapse rows — e.g., "revenue plus the running total" or "each row plus the user's average". Uses OVER (PARTITION BY ...). They're the difference between beginner and intermediate SQL.

Getting data from a source system (extract), cleaning and reshaping it (transform), and loading it into a warehouse where analysts can query it. Modern variants include ELT (transform inside the warehouse) and reverse ETL (warehouse → app). "The data pipeline" usually refers to this whole process.

Fixing messy data — handling nulls, deduplicating, standardising formats, correcting obvious errors. Data analysts spend anywhere from 40% to 80% of their time here. It's not glamorous and it's where most of the real value is created.

A number that measures something about a business — revenue, active users, conversion rate, churn, NPS. A well-defined metric has a clear definition, a data source, and a reason to exist. Metrics without definitions cause endless meeting arguments.

A metric that's genuinely important to the business — a metric tied to a goal. "Monthly revenue" is a metric; "monthly revenue vs. target" is a KPI. Teams usually track 5–10 KPIs, not 100 metrics, because attention is limited.

The single number that best represents customer value for a product — the one everyone else cascades from. For Spotify it might be "minutes listened"; for Airbnb "nights booked". One metric, agreed at the top, that aligns the whole company.

Daily Active Users and Monthly Active Users — the counts of distinct users active in a period. The ratio (DAU/MAU) is a common engagement metric: a high ratio means users come back often; a low ratio means they visit rarely.

The fraction of users or customers who leave in a period. "2% monthly churn" means you lose 2% of your users each month. Critical for SaaS — small changes compound dramatically.

The fraction of users who come back — the inverse of churn but more usefully broken into cohorts. "Day 7 retention" measures how many new users are still active a week later. High retention usually signals a product people genuinely want.

The fraction of users who do a desired action — sign up, make a purchase, subscribe. "3% conversion on the pricing page" means 3 of every 100 visitors bought. Conversion is usually measured per funnel step, not just end-to-end.

CAC (Customer Acquisition Cost) is what you spend to get one customer. LTV (Lifetime Value) is what they're worth over their lifetime. The LTV/CAC ratio is a standard business health check — above 3:1 is usually good.

A visual summary of key metrics, updated regularly. Usually built in Power BI, Tableau, Looker or Metabase. Good dashboards answer a specific question for a specific audience — not "all the data we have".

A dashboard is live, self-serve, always-on — leaders check it weekly. A report is a one-off or periodic document that explains something specific — "the Q3 revenue drop: what happened and why". Both matter; they serve different purposes.

Grouping users by a shared characteristic — usually by when they signed up — and tracking each group over time. Cohort analysis answers "are newer users retaining better than older users?" It's the standard way to measure product improvements without confusing them with growth.

A sequence of steps users go through — visit → sign up → add to cart → purchase. Funnel analysis tells you where users drop off, so you know where to focus. It's one of the most common analyst deliverables.

Breaking users or data into meaningful groups — by geography, device, plan, acquisition channel. Segment analysis often reveals that a "problem" is confined to one group, or an "opportunity" only exists in another. Most useful insights come from segmentation.

Presenting findings as a clear narrative — problem, analysis, recommendation — not a wall of charts. A good data story has a beginning (what question), a middle (what we found), and an end (what to do about it). Analysts who tell good stories get promoted faster than analysts who don't.

Splitting users randomly into two groups (A and B), changing one thing for group B, and comparing outcomes. It's how you prove a change actually caused an improvement — rather than assuming it did. Standard in product and marketing.

The likelihood that an A/B test result isn't just random noise. Usually measured by a p-value — if p < 0.05, we say the result is "significant". It doesn't mean the effect is large or important, just that it's probably real.

The chance of detecting an effect if one actually exists. Low power means you might miss real improvements. Power depends on sample size — small samples produce unreliable tests, which is why "let's just run it for a week" is often too short.

Correlation means two things move together. Causation means one causes the other. Ice cream sales correlate with drowning — but ice cream doesn't cause drowning (hot weather does). Analysts spend a lot of time distinguishing the two.

When your data sample isn't representative of the population you're trying to understand. Surveying only active users about why people churn is biased — you're missing the people who already left. Good analysts always ask "who is in this data, and who is missing?"

A range around an estimate that likely contains the true value — e.g., "conversion is 3.2% ± 0.4%". Wider intervals mean more uncertainty (usually from smaller samples). Reporting a number without its uncertainty is often misleading.

A metric is any measurable number about the business — page views, revenue, signups. A KPI is a metric that's tied to a goal and genuinely important to track. All KPIs are metrics; not all metrics are KPIs.

Cohort analysis groups users by a shared characteristic (often signup date) and tracks each group over time. It's useful because it separates "we're growing" from "our product is improving" — which growth metrics alone can't do.

ETL is Extract-Transform-Load — the process of getting data from source systems into a warehouse where analysts can query it. Without ETL, analysts can't get consistent, clean data. Most "why is the data wrong?" questions trace back to ETL issues.

Correlation means two things move together. Causation means one causes the other. You can't infer causation from correlation alone — you need experiments (A/B tests) or very careful observational studies to prove causation.

The single number that best represents customer value for a product — the one everyone else's metrics cascade from. It's a strategic choice, not a technical one. Good North Stars are customer-centric, not vanity metrics like "total signups".

No — you need the first three sections (foundations, SQL, metrics) to start, plus the basics of dashboards and one visualisation tool. Cohort analysis and A/B testing become important as you progress. The statistics section matters if you go into product analytics or data science.

Build something. Get a small public dataset (Kaggle has thousands). Load it into a database or Excel. Write SQL. Build a dashboard. Try one A/B-test-style comparison. The vocabulary sticks because you'll use each term in context.

No. Data analysts focus on interpreting existing data — SQL, dashboards, reports. Data scientists build statistical and ML models, usually in Python or R. The roles overlap and titles vary by company, but the skills are different enough to matter.

Probably "KPI" — many beginners use it as a synonym for "metric", which it isn't. Others: "cohort" (sounds abstract, is actually just "a group") and "correlation vs causation" (well-known in principle, badly applied in practice).

Our Data Analytics programme covers SQL, Excel, Power BI and Python with real datasets. The blog tutorials cover individual terms in depth — start with "SQL for beginners" or "Your first dashboard".