What is a data platform? The layers, the choices, the pitfalls

Ask ten people what a data platform is and you get ten brand names. That is exactly why projects go wrong: technology gets chosen before anyone has written down which problem it should solve.

A data platform is the layer between your source systems and the place where people decide. It pulls data in, makes it comparable, records what a term means, and delivers that to every report, every agent and every application that needs it.

The layers it consists of

Each layer carries the one above it. An answer at the top is exactly as trustworthy as the meaning below it.

Bottom to top: ingestion pulls data out of systems. Storage keeps the raw landing and the cleaned versions. Modelling turns it into facts and dimensions, with history. Meaning records what margin, order or customer precisely is and who owns it. Use is everything that leans on all of that.

Most organisations have the bottom three layers. What is missing is meaning. So every report calculates for itself, and the outcome differs per department.

Why separate tools do not fix it

Without shared meaning, a new report moves the problem rather than solving it.

A new visualisation tool makes a wrong number prettier, not more correct. As long as the definition sits in a formula inside a report, that definition belongs to whoever built the report. When that person leaves, the meaning leaves too.

What a data platform must be able to do

Four requirements come back every time:

• Lineage. For every number you can retell which source it came from and which steps it went through.
• One place for definitions. A term is recorded once and reused everywhere, not reinvented per report.
• Access per user. Who may see what holds in the platform, not only in the report. See security in Power BI for how that works at report level.
• History. You must be able to see what a number looked like last month, even if the source has since changed.

Where it usually fails

Not on technology. On ownership. A platform forces the question of who owns a term, and that is an organisational question. Without those owners the platform stays a collection of tables, and the argument shifts from which report is right to which table is right.

So start small and concrete: one decision, one term, one owner. See start with one number for that approach, and what data architecture is for the layers drawn out.

In short

A data platform is not a purchase, it is a layer with owners. The technology is the easy half. Who owns which term decides whether the platform produces trust or just more tables.