This glossary is written for business users, product managers, and data-adjacent stakeholders who need to understand what their data engineering teams are talking about. Each term includes a plain-English definition, the business impact it has, and a real example from everyday organisational life.
For the technical deep-dive on these concepts, see Data Jargon 101.
Storage & Schema Models
Schema-on-Read
What it means: Data structure is applied when you query the data, not when you store it. Raw data lands first; shape is defined at analysis time.
💼 Business Impact: Faster data ingestion, more flexibility to explore data in different ways without upfront design decisions.
Example: Storing customer feedback as raw text files, then analysing sentiment or extracting topics when needed — without needing to pre-define which fields matter.
Schema-on-Write
What it means: Data structure is enforced before data enters the system. Non-conforming data is rejected at the door.
💼 Business Impact: Consistent, reliable reports — but slower to adapt when new data types or fields appear. Better for compliance-sensitive data.
Example: A CRM system that forces every customer record to have a valid email and phone number before it can be saved.
Columnar Storage
What it means: Data is stored by column rather than by row. All values for a single field (e.g. "Revenue") are stored together on disk.
💼 Business Impact: Analytical queries that only read a few columns from millions of rows run 10–100× faster and cost less compute.
Example: A "Total Revenue by Region" query on a 1B-row table only reads the Revenue and Region columns — not all 50 columns in each row.
Row-Oriented Storage
What it means: Data is stored by row — all fields of a record live adjacent on disk. Optimised for reading and writing individual records quickly.
💼 Business Impact: Ideal for operational systems (point of sale, CRM, order management) where you read/write one record at a time.
Example: Looking up a single customer's full profile in a CRM — row storage returns everything about that customer in one read.
Processing Patterns
ETL (Extract, Transform, Load)
What it means: Data is extracted from source systems, transformed/cleaned in a staging area, then loaded into the destination (warehouse).
💼 Business Impact: Data is validated before it enters the warehouse — analysts always work with clean data. But new data shapes require pipeline changes, which take time.
Example: A nightly job that pulls sales data from Salesforce, strips test records, converts currencies, and loads it into Snowflake by 6am.
ELT (Extract, Load, Transform)
What it means: Raw data is loaded first into the destination, then transformed inside it. The destination's compute power does the heavy lifting.
💼 Business Impact: Faster time to data — analysts can access raw data immediately while transformation happens in parallel. Easier to iterate on business logic.
Example: Fivetran dumps raw Hubspot data into Snowflake every hour; dbt models then clean and model it for the analytics team.
Batch Processing
What it means: Data is processed in scheduled chunks — e.g. hourly, nightly, or weekly — rather than continuously.
💼 Business Impact: Lower infrastructure cost, simpler to manage, and ideal when decisions don't need to be made in seconds. Most BI reports run on batch data.
Example: Daily sales dashboard refreshes at 7am using data from the previous day.
Streaming / Real-Time Processing
What it means: Data is processed continuously as events arrive — with latency measured in seconds or milliseconds.
💼 Business Impact: Enables real-time fraud alerts, live inventory updates, and instant personalisation. Higher infrastructure cost and complexity.
Example: A bank's fraud detection system that flags a suspicious transaction within 2 seconds of it being initiated.
Data Access & Analytics Types
OLTP (Online Transaction Processing)
What it means: Systems optimised for high-volume, high-speed transactional operations — inserts, updates, and point reads.
💼 Business Impact: Powers your day-to-day operational applications — order management, CRM, inventory, banking. Must be fast and reliable 24/7.
Example: The database behind an e-commerce checkout — must handle thousands of concurrent order writes per second.
OLAP (Online Analytical Processing)
What it means: Systems optimised for complex analytical queries — aggregations, groupings, and joins across millions to billions of rows.
💼 Business Impact: Powers your dashboards, reports, and business intelligence. Slower for individual record lookups, but fast for "total sales by region and month" queries.
Example: A Power BI dashboard that shows revenue trends across 50 markets over 3 years, querying a cloud data warehouse.
Data Virtualisation
What it means: A unified view of data from multiple sources — without physically moving or copying the data. Queries are pushed down to each source at runtime.
💼 Business Impact: Analysts can query data from CRM, ERP, and marketing tools in one place without waiting for ETL pipelines to be built.
Example: A finance analyst writing one SQL query that joins Salesforce revenue data with SAP cost data without either being moved to a central warehouse.
Governance & Quality
Data Catalog
What it means: A searchable inventory of all data assets — tables, files, reports, ML models — enriched with descriptions, owners, and quality scores.
💼 Business Impact: Analysts spend less time hunting for data and more time using it. Reduces duplication of datasets and "whose number is right?" debates.
Example: A data analyst searches "customer churn rate" in the catalog and finds the official definition, the table it comes from, the team that owns it, and how fresh it is.
Data Lineage
What it means: A map of where data comes from, how it flows through transformations, and where it ends up — from source to report.
💼 Business Impact: "Where did this number come from?" becomes answerable in seconds. Essential for regulatory audits and impact analysis when source systems change.
Example: A CFO asks why Q3 revenue changed overnight. Data lineage shows the finance team that a currency conversion bug in a pipeline 3 hops upstream was the cause.
Data Contract
What it means: A formal agreement between a data-producing team and consuming teams — specifying the schema, freshness SLA, and quality guarantees of a dataset.
💼 Business Impact: Prevents "the pipeline broke and nobody told the analytics team" scenarios. Creates accountability for data quality.
Example: The Orders team commits to delivering a clean `orders_daily` table by 7am with <0.1% null rates and schema versioning notice of 30 days for any breaking changes.
Data Quality
What it means: The degree to which data is accurate, complete, consistent, timely, and fit for its intended use.
💼 Business Impact: Poor data quality costs organisations an average of $12.9M per year (Gartner). Decisions made on bad data are worse than no decisions at all.
Example: A marketing campaign targeting "customers with 0 purchases in the last 90 days" accidentally includes active customers due to a join bug — wasting budget and confusing customers.
Architecture Concepts
Medallion Architecture (Bronze / Silver / Gold)
What it means: A pattern for organising data into three progressive quality zones — Bronze (raw), Silver (cleaned), Gold (aggregated, business-ready).
💼 Business Impact: Analysts always work with Gold-quality data. Engineers can reprocess historical data from Bronze when business logic changes.
Example: Raw IoT sensor readings arrive in Bronze every second. Silver applies noise filtering. Gold produces hourly averages by machine — what the maintenance team actually uses.
Data Product
What it means: A curated, documented, and SLA-backed dataset treated like a software product — with an owner, quality guarantees, and a versioned interface.
💼 Business Impact: Other teams can trust and consume data without needing to understand how it was built. Reduces ad-hoc data requests to engineering.
Example: The Logistics team publishes a "Delivery Performance" data product — updated every 4 hours, with documented fields, owned by a named engineer, with a support channel.
Data Mesh
What it means: An operating model where each business domain owns its data end-to-end, treating data as a product, supported by a shared self-serve data platform.
💼 Business Impact: Eliminates the central data team bottleneck. Business domains move faster. Better data quality because owners understand the data.
Example: Instead of the Finance team waiting 3 months for the central data team to build a pipeline, they own and build their own "Financial KPIs" data product using the shared platform.
Data Lakehouse
What it means: An architecture that combines the low-cost, flexible storage of a data lake with the reliability and governance of a data warehouse — using open table formats.
💼 Business Impact: One platform for BI, data science, and ML — without copying data between a lake and a warehouse. Saves cost and eliminates data freshness delays.
Example: A retailer's data scientists train ML models on the same Delta tables that power the BI dashboards — no separate export step, no version mismatch.
ACID Transactions
What it means: A guarantee that database operations are Atomic (all-or-nothing), Consistent (always valid state), Isolated (concurrent operations don't interfere), and Durable (committed data survives crashes).
💼 Business Impact: Prevents partial writes and data corruption. Critical for financial transactions, inventory updates, and any operation where partial failure is unacceptable.
Example: A bank transfer debiting Account A and crediting Account B must either fully complete or fully fail — never leave one account updated and the other not.
Time Travel
What it means: The ability to query a table as it existed at a previous point in time — any historical version is accessible.
💼 Business Impact: Enables debugging ("what did the data look like before the pipeline bug?"), audit trails, and regulatory point-in-time reporting without maintaining expensive snapshots.
Example: An auditor asks for the customer list as it existed on 31 March. With Delta Lake time travel, the data team queries `SELECT * FROM customers TIMESTAMP AS OF '2025-03-31'` in seconds.
Technology Terms
Apache Spark
What it means: An open-source distributed computing engine for large-scale data processing. Runs workloads in parallel across hundreds of machines.
💼 Business Impact: Processes terabytes of data in minutes rather than hours. Powers the ETL pipelines and ML training jobs for most large organisations.
Example: A global retailer processes 10 years of transaction history (2 terabytes) in 4 minutes using Spark on a 50-node cluster.
dbt (data build tool)
What it means: A SQL-first transformation framework that lets analysts write transformation logic in SQL, with built-in testing, documentation, and version control.
💼 Business Impact: Analysts can own and ship data models without waiting for engineers. Every transformation is tested, documented, and reviewable like code.
Example: A marketing analyst writes a dbt model that defines "monthly active users" in SQL, with automated tests that verify the result is never null or negative.
Apache Kafka
What it means: A distributed event streaming platform that acts as a high-throughput, durable message broker — receiving, storing, and distributing events in real time.
💼 Business Impact: The backbone of real-time data architectures. Enables different systems to react to business events (order placed, payment failed) without tight coupling.
Example: Every product page view publishes an event to Kafka. Three independent systems consume it: the recommendation engine, the analytics pipeline, and the A/B testing platform.
Delta Lake
What it means: An open-source storage format that adds ACID transactions, schema enforcement, and time travel to data stored as Parquet files in object storage (like S3 or Azure Data Lake).
💼 Business Impact: Turns a raw data lake into a reliable, queryable database without the cost of a proprietary warehouse. Enables safe concurrent writes and data corrections.
Example: A data engineer accidentally loads duplicate records. Because the table is Delta Lake, they run `RESTORE TABLE events TO VERSION AS OF 5` and recover in 30 seconds.