Building a data platform: components, approach and costs

Data platform bouwen: onderdelen, aanpak en kosten
Date
November 20, 2026
Author
Isatis Group
Category
Blog
Read time
9 min read

Learn how to build a data platform: the five layers, a six-step approach, what drives the cost and the mistakes to avoid. Start with one question.

It's Monday morning and the management team is going over last month's revenue. Finance has one figure from the accounting package, sales has a different one from the CRM, and operations has its own spreadsheet that lands somewhere in between. The first twenty minutes go on working out which number is right. The conversation about what to do with the numbers only starts after that, or not at all. That's the moment when organizations start looking for a better way to work with their data.

A data platform brings the data that's now scattered across separate systems, spreadsheets and databases together in one reliable source. You build it step by step: you start with a concrete question, connect the sources you need to answer it, sort out storage and agreements on definitions, and only then build reporting and AI on top. The cost depends on the number of sources, the quality of your data and how far you want to go. So there's no fixed price for a platform.

What is a data platform and when do you need one?

A data platform isn't a product you buy and switch on. It's a coherent whole of connections, storage, agreements and applications that lets people in your organization work with data reliably.

You need one when the same question gets a different answer depending on which system you ask. Other signs are:

  • Manual work every month: someone pastes exports from three systems together to build a report.
  • Nobody dares to use the numbers: there's no agreement on what a customer, an order or an active user actually is.
  • AI plans that stall: a forecasting model or document processing needs clean, findable data, and it isn't there.

Most organizations have more data than they think. The problem is rarely a shortage of data; it's usually fragmentation.

The components: five layers you'll see again and again

Whatever the technology is called, the structure is almost always the same. You can divide a data platform into five layers:

  • Sources: your business software, databases, spreadsheets, third-party APIs and, where relevant, sensors or machines.
  • Ingestion: the connections that pull data out of the sources, at set times or continuously. This is where most things go wrong in practice, for example when something changes on the source side.
  • Storage: a data warehouse for structured data, a data lake for raw or mixed data, or a combination of the two.
  • Processing and modeling: cleaning, combining and fixing definitions so that revenue means the same thing everywhere.
  • Use: dashboards, reports, analyses and AI applications that people and processes work with.

Across all the layers sits one that's often forgotten: access rights, quality checks and monitoring. Without it, the platform loses its users' trust within a year.

Warehouse, data lake or something simpler

The storage choice is often made bigger than it needs to be. A data warehouse fits when you mainly have structured business data and want fixed reports. A data lake fits when you want to keep many different kinds of data, such as documents, log files or sensor data, and don't yet know what you'll use them for.

For a smaller organization with five or six sources, something simpler is sometimes enough: one well-set-up database with scheduled connections and clean modeling. It makes sense to match the setup to where your organization is now and where it's heading, not to what a company with ten times the data needs. You can always expand later, as long as the foundation is laid properly.

The choice of cloud platform plays a part here too. AWS, Azure and Google Cloud all offer the building blocks for storage and processing. Which one fits best depends on what your organization already uses, where your data is allowed to be stored and the knowledge in your own team.

A six-step approach

A workable sequence looks like this:

  1. Pick one question that matters. For example: what's our revenue per customer group, or where do orders get delayed? A platform that tries to answer everything at once never gets finished.
  2. Map out the sources. Which systems hold the data for that question, who manages them and how good is the quality? This is where the surprises come up, such as duplicate customer numbers or fields that have been used differently for years.
  3. Define your terms. Together with people from the business, you decide what the core concepts mean. This isn't technical work, but it determines whether the platform will be trusted later.
  4. Build the first end-to-end line. From source to dashboard, for that one question. That way you find out early whether the connections, storage and reporting work well together.
  5. Test it and let users take a look. Check the figures against what people already know and adjust where they differ. A difference almost always points to an unclear definition or an error in the source.
  6. Expand and take care of maintenance. Add sources and questions, record who is responsible for what and set up monitoring, so you spot a broken connection before a user does.

The first line is the most important moment. If people work with it and learn something from it, support for the rest grows on its own.

What determines the cost of a data platform?

You can be matter-of-fact about costs: without knowing your situation, nobody can give you a reliable figure. Isatis doesn't use fixed rates either. What you can do is name the factors that determine the amount:

  • Number and type of sources: connecting to a modern API is less work than an old system with no way to connect to it or a folder full of spreadsheets.
  • Data quality: the more cleaning and deduplication that's needed, the more time it takes.
  • Size of the initial scope: one question and one dashboard is something different from a full company-wide platform.
  • Security and access requirements: these weigh heavily, especially in healthcare, government and education.
  • What gets built on top: reports, forecasts or AI applications are separate work packages.
  • Maintenance after go-live: connections change, sources get replaced and new questions come up. Factor this in from the start.

If you want a first indication, you can choose an indicative, non-binding budget range in the Isatis contact form: under €10,000, €10,000 – €25,000, €25,000 – €50,000, €50,000 – €100,000 or €100,000+. That's not a quote, but it helps us work out together which scope fits your budget.

Common mistakes when building one

Most data platforms don't fail because of the technology. They fail because of a few recurring mistakes:

  • Starting with the tooling instead of the question. If you pick a platform first and look for an application afterwards, you often end up building something nobody uses.
  • Skipping definitions. Without agreements on what a term means, all you get on screen is prettier versions of the same discussion.
  • Wanting everything at once. A big year-long program with no interim results loses its sponsor along the way.
  • Not appointing an owner. Data without someone responsible quickly becomes outdated or unreliable.
  • Delivering a black box. If only the builder understands the platform, you depend on that one person or supplier.

That last point is also a reason to choose an architecture your own team can maintain. It's the same thinking as with other custom-build choices, as described in our blog on how to avoid vendor lock-in with custom software.

Where AI fits on a data platform

AI isn't a separate step you add later, but an application that can run on a good platform. Document processing, classification, forecasting and anomaly detection all need reliable, findable data. If you don't have that foundation, you'll see a model work well in a demo and disappoint in practice.

Practical AI means using it where it saves time or catches errors, and making sure the solution is explainable, testable and maintainable by your own team. You can read what that delivers in practice in our blog on data analysis with AI.

Isatis builds both the AI solutions and the data platforms they run on. On the page about data and AI solutions you can see how it all fits together: from data strategy and platforms to reporting and automation. The advantage of one team is that the people who design the platform keep improving it after go-live, with performance monitoring and regular check-ins. Isatis is also ISO 9001 and ISO 27001 certified, which matters if you work with sensitive data.

Frequently asked questions

What layers does a data platform consist of?

At its core, five layers: sources, ingestion, storage, processing and modeling, and use in the form of dashboards, analyses or AI. Across them sits a layer for access rights, quality control and monitoring.

Which data platform is the best?

There's no such thing as the best one in general. It depends on your sources, your data, your security requirements and the knowledge in your team. A good choice starts with the question you want to answer and with what your organization already uses, not with a list of popular tools. It can run on AWS, Azure or Google Cloud, and sometimes a simpler setup is enough.

What does it cost to build a data platform?

There's no fixed amount for that, and Isatis doesn't quote fixed rates. The cost follows from the number of sources, the data quality, the size of the initial scope and what you want to build on top. In the contact form you choose an indicative, non-binding budget range, so you quickly know what fits within your budget.

Do I always need a data lake?

No. A data lake mainly makes sense if you want to keep many different kinds of data without having settled yet on what for. For a limited number of structured sources, a data warehouse or a simpler setup may be a better fit and easier to maintain.

Talk first, then build

Not sure whether your organization is ready for a data platform, or want to know which first step would deliver the most? Start with a 15-minute conversation: you tell us what you want to achieve and we'll honestly tell you whether we can help. You can schedule that conversation via the contact page, and we respond to every message within 24 hours.


About the author: Jeroen Penninx, Director and co-founder of Isatis.

Share this article

Jack van Poll

Jack van Poll

Co-Founder, Isatis

Writes about nearshore engineering,
software partnerships and building teams that last.

Get in touch

Read next.

All articles