ETL

ETL stands for Extract, Transform, Load: the three-step process that moves data from its source systems to a central repository where it can be analyzed. First it extracts information from scattered sources (a CRM, an ERP, spreadsheets, a payment gateway), then it transforms it to clean it and give it a consistent format, and finally it loads it into a destination such as a data warehouse or a data lake.

ETL is the invisible plumbing of analytics: without it, each system speaks its own language and reports end up being built by hand in spreadsheets that never reconcile. A well-built ETL process is the foundation of a single source of truth, and it is part of what lays the groundwork for the dashboards and models Metrix offers.

The value of ETL is that it solves the most expensive problem in analytics: data almost never arrives ready to use. The e-invoice says "Customer: ACME S.A.", the CRM stores it as "Acme SA" and the collections spreadsheet records it as "ACME". To a human they are the same company; to a computer they are three different customers. ETL exists to reconcile those differences automatically and repeatably, every night, without anyone having to paste columns by hand.

The three steps, in practice

• Extract: data is read from each source. Typical sources at an Argentine company are the CRM (for example, Salesforce), the ERP, the e-invoicing system of ARCA (Argentina's federal tax agency, formerly AFIP), sales reps' Excel files, a payment gateway API and marketing exports. Extraction can be full (everything every time) or incremental (only what changed since the last run).
• Transform: this is the step that takes the most work. Here null values are cleaned up, date and currency formats are standardized (pesos versus dollars), names are normalized, business rules are applied (for example, calculating margin) and data quality is validated. This is where "raw data" becomes "reliable data".
• Load: the processed data is written to the analytical destination. From there it feeds dashboards, reports and models without touching the operational systems again, which stay free for their day-to-day work.

ETL versus ELT: order matters

With the arrival of cloud data warehouses, a variant appeared, ELT, which reverses the last two steps: it first loads the raw data and only then transforms it, inside the warehouse, taking advantage of its computing power. One does not replace the other; they coexist depending on the case.

Why it matters to the business

A consumer goods company that wants a sales dashboard by channel needs to cross-reference each distributor's sell-out with its own sell-in, with prices and with collections. Each of those data sets lives in a different system. Without ETL, someone builds that spreadsheet every Monday morning, it takes hours and last month's number never matches finance's. With ETL running, the cross-referencing happens on its own, overnight, and by morning the dashboard is already updated and shows the same number for everyone. That is the leap: from arguing about whose data is right to arguing about what to do with the data.

Common mistakes

• Transforming without documenting: when the business rules live only in the head of whoever built the process, ETL becomes a black box that is impossible to maintain.
• Loading without validating: if quality is not checked before loading, errors spread to every dashboard and trust in the data is lost.
• Overly rigid processes: an ETL that does not account for new fields or new sources breaks with every change in the business.
• Forgetting the cadence: a process that runs whenever someone remembers is useless; the value lies in it being automatic and predictable.

In practice, you rarely notice ETL when it works well: the measure of success is that nobody argues about where a number comes from. Once that plumbing is solid, it starts to make sense to invest in advanced analytics, conversational BI or predictive analytics, because they all rely on data that already arrives clean.