ETL
ETL (extract, transform, load) is the process of pulling data out of source systems, reshaping it to fit a target schema, and loading it into a warehouse or database where reports and AI can rely on it.
No jargon, no math
In plain words, what is it?
ETL moves data from where it is created to where it is used, in three named stages. Extract pulls raw rows out of your ERP, spreadsheets, sensors and invoicing tools. Transform is the workshop: dates to one format, currencies to one unit, duplicate customers merged, codes mapped to a shared catalogue. Load writes the clean rows into warehouse tables, replacing or appending by rule. Runs are scheduled, so last night's data is ready by breakfast. The breaking points are known: a source that quietly changes a column name, a unit left unconverted, a duplicate key, and the whole report is wrong with a straight face. That is why every serious ETL job validates before it loads, and alerts a human when checks fail.
- 01
Extract. Raw rows are pulled from the ERP, spreadsheets and sensor feeds.
- 02
Transform. Formats, units and codes are standardised; duplicates are merged.
- 03
Validate. Rules check the shape: totals match, dates parse, required fields are filled.
- 04
Load or alert. Passing rows land in the warehouse; a failed check alerts a human before anything is written.
- 05
Repeat. The run is scheduled, so yesterday's data is waiting every morning.
Terms that live nearby.
Data Pipeline
A data pipeline is the connected chain of steps that carries data from where it is produced to where it is consumed: ingest, validate, transform, store and serve, each stage handing clean, well-formed data to the next.
ReadStructured Output
Structured output is a mode where the model is forced to answer in a machine-readable format (JSON, a table, a fixed schema) instead of free text, so its answer can be validated, stored and passed to other software without a human retyping it.
ReadUnstructured Data
Unstructured data is information with no fixed schema: documents, emails, images, audio, chat logs, PDFs. Most of what a company knows lives in this form, but software must first parse, transcribe or embed it before it can be searched or counted.
Read§ 04
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.