What is a data platform, and does your company need one?

A data platform is the set of systems that collects your company’s data from the tools it lives in, stores it in one place, and turns it into numbers everyone agrees on. Your company needs one when teams argue about whose figures are right, when people build reports by hand every week, or when the answers you need are stuck inside software you cannot query. If none of that sounds familiar, you probably do not need one yet, and a few well-built reports will serve you better. Below we explain the parts in plain English, how to tell which side of that line you are on, and how to start without committing to a large build.

What does a data platform actually do?

Most companies already have plenty of data. It is spread across the tools they use to run the business: a CRM for sales, an accounting package, an order or booking system, a support desk, ad accounts and a few spreadsheets. Each tool has its own reports, and each one answers questions about itself.

The questions that matter usually cross those lines. Which marketing channel brings customers who stay? Which product lines make money once returns and support time are counted? A data platform exists to answer questions like these, by bringing the data together and keeping it in a shape people can trust.

The parts of a data platform

A data platform is less a single product than a chain of parts, each doing one job.

Sources. The systems where data is created: your CRM, accounting, e-commerce, product database, ad platforms and so on. You do not replace these. The platform reads from them.

Pipelines. Scheduled jobs that copy data out of each source into one central store. The common pattern is called ELT, short for extract, load, transform: pull the data out (extract), copy it as it is into the central store (load), then clean and reshape it there (transform). Connector tools such as Fivetran and Airbyte handle the extract and load steps for many popular systems, and custom code covers the rest.

A warehouse. A database built for analysis rather than for running an app. It holds the copied data and is designed to answer questions across large amounts of it quickly. BigQuery and Snowflake are common choices, and for smaller volumes a well-run Postgres database can do the job.

Transformations. The step that turns raw copies into tidy tables: removing duplicates, matching the same customer across systems, applying business rules. Tools such as dbt let a team write these steps as code, so they are versioned, reviewed and tested like any other software.

A semantic or metrics layer. This is where each business term gets one definition. “Active customer”, “net revenue” and “churn” are written down once, and every report uses that definition. Some dashboard tools have this built in, and some teams keep it in a separate layer. Either way, it is the part that stops two teams from reporting two different revenue numbers.

Dashboards and reports. The screens people actually look at, in tools such as Looker Studio, Power BI or a custom app. They read from the tidy tables and the shared definitions, so they agree with each other.

AI on top. Once the data is clean and defined, it becomes useful to AI features too: a weekly summary written from real figures, a forecast, or a question box that answers from your own numbers. AI built on scattered, inconsistent data gives scattered, inconsistent answers. Built on a platform, its output can be checked against figures you already trust.

Signs your company needs one

These are the patterns we see most often when a company is ready.

• Numbers disagree between teams. Sales and finance bring different revenue figures to the same meeting, and the first part of the meeting goes on working out why.
• Reports are built by hand. Someone exports from several systems every Monday, pastes into a spreadsheet and fixes the formulas that broke. If that person is away, the report does not happen.
• Data is trapped in SaaS tools. The answer exists, but only inside a vendor’s report screen that cannot be combined with anything else, or only through an export someone has to remember to run.
• Simple questions take days. A leader asks something reasonable, and answering it becomes a small project every time.
• You want AI features but cannot say what data they would use. That is a data problem before it is an AI problem.

Signs you don’t need one yet

A data platform is a real commitment. It needs building, and it needs looking after. You may be better off waiting if:

• Most of your important data lives in one system, and its built-in reports answer your questions.
• A small group uses the numbers, and they agree on what the numbers mean.
• Your reports rarely change and take little time to produce.
• You have not yet settled how the business measures itself. Building a platform around metrics that will change next quarter means building it twice.

In those cases, good reports straight from your main system, or one carefully built spreadsheet, may be all you need for now. We would rather tell you that than build something you will not use.

How to start small

You do not need every part on day one. The safest way in is to pick one question that matters and build the shortest path that answers it well.

1. Choose one decision. For example, “which channels bring customers who come back”. Write down who makes that decision and how often.
2. List the sources it needs. Usually a few systems, not every system you own.
3. Agree the definitions. What counts as a customer, a repeat purchase, a channel. Have the people who own those numbers sign off.
4. Build the thin slice. Pipelines for those sources, a warehouse, the transformations and one dashboard. Nothing more.
5. Use it, then extend it. Once people rely on the first dashboard, add the next question. Each new one reuses the pipelines and definitions you already have, so the platform grows from real use rather than from a plan drawn up in advance.

Our article on replacing spreadsheet reports with dashboards covers how to roll out that first dashboard so people actually stop opening the old spreadsheet.

What to ask a vendor

Whether you hire us or someone else, these questions will tell you a lot about how the platform will hold up.

• Where will our data live, and who owns the accounts? You want the warehouse and tools in accounts your company controls, so you can change partners without losing your data.
• Where are business definitions written down? Look for one shared place, rather than formulas copied into each dashboard.
• How will we know the data is fresh and correct? Ask what checks run, what happens when a pipeline fails, and who gets told.
• What does it take to add a new source or a new report? The answer shows whether the design will grow or need rebuilding.
• What will it take to run day to day? Every platform needs some care. Ask what that work is and whether your own team could take it on.
• Can we start with one question? A vendor who insists on building everything at once is asking you to take a lot on trust.

If you are still weighing who should do the work, our piece on choosing a consultant, an agency or an in-house team goes through the trade-offs.

Where to start

Write down the one question your leadership asks most often that takes too long to answer, and list the systems that hold the data for it. That short note is enough to plan a first slice. If you want a partner to design and build it alongside your team, our dashboards and data platforms service explains how we work, and you are welcome to get in touch with your question.

Questions people ask

What is the difference between a data warehouse and a data platform?

A data warehouse is one part of a data platform: the database where your data is stored for analysis. The platform also includes the pipelines that fill the warehouse, the shared definitions that keep numbers consistent, and the dashboards and AI features that use the data.

Does a mid-sized company need a data platform?

Not always. If most of your data lives in one system and its built-in reports answer your questions, you may not need one yet. A platform starts to earn its place when teams disagree on numbers, reports are built by hand, or important data is spread across several tools.

How should a company start building a data platform?

Start with one important business question, connect only the few systems that question needs, agree the definitions with the people who own those numbers, and build one dashboard. Add the next question once people rely on the first.