A data silo is a set of data locked in one system or one department, while others need that data too. Data silos usually form because systems were bought separately from each other. As a result, the same data exists in several places, without one version being leading.
What exactly is a data silo?
A data silo is data the rest of the organisation needs but can only reach through manual work. Sales works in the CRM, finance in the ERP, the warehouse in its own system. Each system does its job, but they don't share their data.
So the defining feature is not that data is stored separately. That is true in every organisation. The defining feature is that the separation causes a problem: someone has to export, retype or ask around to get a complete picture. A data warehouse is the opposite of a data silo: a place that brings data from several sources together.
How do data silos form?
Technical causes.
Data silos often form because every department picks its own software. A marketing tool, a planning package, a PIM system for product data: each is a good choice on its own, but nobody decided up front how they exchange data. Older systems without an API, the agreed entry point through which software shares data, make it harder.
A separate category is shadow IT: spreadsheets and tools employees start using because the official system falls short. That spreadsheet with the real pricing agreements is often the most underestimated data silo in an organisation.
Organisational causes.
Technology is rarely the only cause. Departments with their own budget buy their own systems. Nobody owns the data that several departments depend on. And sometimes data is shielded on purpose: "our" data stays with us. Without agreements on ownership, a new data silo appears as soon as an integration drops out.
How do you recognise a data silo in your organisation?
You recognise a data silo by the manual work needed to bring data together. These are the most common signs:
The same account appears under different names in different systems.
Someone makes an export every month to compare two systems.
Ask how much was delivered to a client and each department gives a different answer.
An address change has to be made in three places.
Reports are merged in a spreadsheet before they go to management.
A new colleague has to learn which system holds the right version of which data.
What does a data silo actually cost?
A data silo costs time, and that cost can be calculated. Add up the hours per month spent exporting, retyping and checking, and you have the direct part. Then come the errors: invoices sent to an old address, orders at an outdated price, stock that differs between the web store and the warehouse.
The indirect part is harder to measure but often bigger. Decisions are made on last week's figures. And projects that rely on data, such as reporting or AI, stall because the data quality has to be fixed first.
When is a data silo not a problem?
A data silo is not a problem when nobody outside the team needs the data. A test lab with its own measurement data, or the planning of one location that appears nowhere else: centralising costs money and delivers nothing.
Sometimes separation is even required. Personnel files and medical data should not be accessible to everyone. So the question is not how to break down every silo, but which separation gets in the way of a process. Start there.
Breaking down data silos: four routes.
You break down a data silo by making systems work together or by making one source leading. In practice there are four routes.
Integrate. Systems exchange data automatically through an API or an integration. This is the most common route, and the basis of data integration.
Centralise. Data from several sources comes together in one data platform, where it is combined for reporting or further processing. That is the role of a Data Management Platform.
Agree on one source. For each data type you decide which system is leading. All other systems read from it. This is the principle of a single source of truth, and for master data you work it out with master data management.
Consolidate. Sometimes it is simpler to replace two systems with one. If product data is spread across the ERP, the web store and spreadsheets, a PIM system for product data brings it together in one place.
In short.
A data silo is data locked in one system or department while others need it. Data silos form through separate software choices and unclear ownership. You recognise them by manual work: exports, retyping, different answers to the same question. Not every silo has to go. Break down only the separations that hinder a process, by integrating, centralising, agreeing on one leading source or consolidating.
Recognise data silos in your organisation? Schedule a no-obligation call.
Related reading...
What is Single Source of Truth (SSOT?
Learn what SSOT means and how to implement it for reliable data integration in your organization
What is data quality?
The six dimensions, how to measure them and how to improve data quality for good.
What is master data management (MDM)?
Managing master data in practice, with or without a dedicated package.
Questions about data silos? No problem.
Look at the manual work. If exports are compared every month, changes have to be entered in several places or departments give different figures for the same question, you have data silos.
No. Data only one team uses does not need to be central. Sometimes separation is even required, as with personnel data. Break down only the silos that demonstrably hinder a process.
A data silo keeps data locked in one system. A data warehouse brings data from several systems together, so you can run reports and analyses on it.
Partly. An ERP brings finance, stock and purchasing together, but systems outside it, such as the CRM or the web store, remain separate sources. Those still need to be integrated.
Often nobody, and that is exactly the problem. Appoint one owner per data type who decides which system is leading and who may change the data.