# Data Warehouse and ETL Explained: Architecture Options and When Your Business Needs One

> What a data warehouse is, how ETL works and when you need one. Learn the layers, ETL vs ELT vs CDC and 8 steps to bring ERP and CRM data into one source.

Source: https://digitalbridge.com.tr/en/blog/data-warehouse-etl-guide.html
Publisher: Digital Bridge Teknoloji (https://digitalbridge.com.tr)
Topic: Data & Analytics
Published: 2026-09-24 · Updated: 2026-10-30

A data warehouse is a central database that brings together data from separate systems — ERP, CRM, e-commerce, production, HR — for reporting, with shared definitions and full history. [ETL](https://digitalbridge.com.tr/en/glossary.html#etl) (extract, transform, load) pulls, cleans and loads data from each source into the warehouse, so reports run from one source without slowing live systems and everyone sees the same number.

## Key takeaways

- A data warehouse combines many systems' data in one reporting database, with shared definitions and history, so reports do not slow the live ERP.
- ETL transforms data before loading it into the warehouse, ELT loads raw data first, and CDC moves only changed records.
- According to TurkStat, in 2025 28.3% of Turkish enterprises with 10+ employees used ERP, but only 6.5% used business intelligence (BI) software.
- A project starts from business questions: build one data mart, such as sales or finance, end to end, then expand.
- Because a warehouse gathers personal data from many systems, permissions, masking and KVKK retention periods are designed from the start.

## Why does company data end up scattered across systems?

As a company grows, every job gets its own system. Accounting and inventory live in the ERP (enterprise resource planning), quotes in the CRM (customer relationship management), orders on the webshop and marketplace portals, production data in the [MES (manufacturing execution system)](https://digitalbridge.com.tr/en/blog/mes-vs-erp.html), staff records in the HR software. Each system produces data in its own format, on its own schedule, with its own codes.

When a report is needed, the typical routine goes like this: someone exports from each system, stitches the files together in Excel with lookups, and fixes unmatched records by hand. Sales and finance end up with different revenue figures; heavy report queries slow the ERP during working hours; and when someone wants a like-for-like comparison with last year, prices and the organisation chart have long since changed. We set out the way out of this manual routine in [moving from Excel reports to BI](https://digitalbridge.com.tr/en/blog/from-excel-reports-to-bi.html), and how to decide who owns a definition such as "revenue" in our [data governance framework](https://digitalbridge.com.tr/en/blog/data-governance-framework.html).

The ERP alone cannot fix this, because it is built to record transactions, not to analyse them. A quote in the CRM, a marketplace order or a downtime event in the MES may never reach the ERP at all.

## What does scattered data cost a business?

In Türkiye, most businesses record their data but do not analyse it:

> According to TurkStat, in 2025 28.3% of Turkish enterprises with 10+ employees used ERP software, while only **6.5%** used business intelligence (BI) software. ([TurkStat ICT Usage Survey in Enterprises 2025](https://veriportali.tuik.gov.tr/tr/press/54012))

The gap between ERP and BI use represents companies whose data sits in systems but is never combined for decisions. The cost of scattered data never shows up as a single line item; it surfaces in a wrong report, a delayed decision, stock counted twice. Data quality expert Thomas C. Redman estimates the cost of bad data at 15% to 25% of revenue for most companies, and reckons two-thirds of those costs can be identified and eliminated permanently ([MIT Sloan Management Review, "Seizing Opportunity in Data Quality", 2017](https://sloanreview.mit.edu/article/seizing-opportunity-in-data-quality/)).

The infrastructure picture is shifting as well. TurkStat reports that 20.2% of Turkish enterprises used paid cloud computing services in 2026 — 57.0% of those with 250+ employees against 16.7% of those with 10-49 ([TurkStat ICT Usage Survey in Enterprises 2026](https://veriportali.tuik.gov.tr/tr/press/58021)). The same survey found that only 15.2% employed ICT specialists. In short, a data warehouse is no longer only for large corporates, but building and running one takes expertise most companies do not have in-house. To judge whether your data really is at big data scale, see [what is big data for business](https://digitalbridge.com.tr/en/blog/what-is-big-data-for-business.html).

## How does a data warehouse work? Layers and the ETL process

A data warehouse is made up of several layers that run from source to report:

- **Source systems:** ERP (Logo, SAP, Mikro, Netsis or bespoke), CRM, e-commerce and [marketplace APIs](https://digitalbridge.com.tr/en/blog/api-integration-explained.html), MES/SCADA, HR and payroll, spreadsheets.
- **ETL / ELT processes:** extracting data from the source, transforming it according to business rules and loading it into the warehouse. In [ELT](https://digitalbridge.com.tr/en/glossary.html#elt) the raw data is loaded first and transformed inside the warehouse. We compare when each approach fits in [ETL vs ELT differences](https://digitalbridge.com.tr/en/blog/etl-vs-elt-differences.html).
- **Staging area:** where raw data is held temporarily and quality checks run before transformation.
- **Warehouse model:** a star schema built from fact tables (sales, production, movements) and dimension tables (customer, product, date, branch). It makes reports faster and easier to understand.
- **Data marts:** subsets of the warehouse focused on one area such as sales, finance or production.
- **Data lake:** separate storage for high-volume or unstructured data such as sensor streams, log files and images. We explain how it differs from a warehouse in [data lake vs data warehouse](https://digitalbridge.com.tr/en/blog/data-lake-vs-data-warehouse.html), and cover [time-series databases for sensor data](https://digitalbridge.com.tr/en/blog/time-series-database-iot-data.html) separately.
- **Consumption layer:** [management BI dashboards](https://digitalbridge.com.tr/en/blog/bi-dashboard-kpis.html), reports, [AI demand forecasting](https://digitalbridge.com.tr/en/blog/ai-demand-forecasting.html) and other AI models.

Spatial data can be a source too. Municipalities and utilities keep pipe, valve, subscriber and zone layers in their [municipal GIS](https://digitalbridge.com.tr/en/blog/gis-for-municipalities.html), and water utilities in a [water network GIS inventory](https://digitalbridge.com.tr/en/blog/gis-for-water-utilities.html); once those layers are loaded into the warehouse by asset code or zone key, faults, consumption and billing can be reported zone by zone in one place.

### ETL, ELT or a direct connection?

| Approach | How it works | Good fit for | Watch out for |
|---|---|---|---|
| Direct reporting connection | The BI tool queries the source database directly | One system, few users, simple reports | Slows the live system; history is not kept |
| ETL | Data is transformed outside the warehouse and loaded clean | On-premises servers, strict data rules | Changing a transformation rule may mean reloading |
| ELT | Raw data is loaded first, transformed inside the warehouse | Cloud warehouses, high data volumes | Access to personal data in raw tables must be controlled |
| CDC (change data capture) | Only changed records are transferred | Frequently updated critical data such as stock and orders | Check that the source database supports it |

In practice most data warehouse projects mix these methods: stock and order data moves within minutes via CDC, while finance data is loaded in a nightly batch.

### Why does a data warehouse keep history?

When a customer moves to a different sales rep this year, which rep should last year's sales be credited to? The answer changes your reports. Source systems usually hold only the current state. A warehouse uses a "slowly changing dimension" technique to store old and new values with their dates, so historical figures stay correct when prices, customer accounts or the organisation change.

## How do you build a data warehouse? 8 steps

1. **Collect the questions.** A warehouse is designed around the business questions it has to answer. List what management, sales, finance and production ask every month and the KPIs they use. We explain how to choose those indicators in [how to set KPIs](https://digitalbridge.com.tr/en/blog/how-to-set-kpis.html).
2. **Map the sources.** Which data lives in which system and table, how often it changes and who owns it. Parallel lists kept in spreadsheets belong on this map too.
3. **Write shared definitions.** Without a single definition of "net sales", "active customer" or "stock on hand", a warehouse simply moves the old argument into a new database. The single reference for customer and product codes across systems is set through [master data management (MDM)](https://digitalbridge.com.tr/en/blog/master-data-management-guide.html), and a [data catalog and data dictionary](https://digitalbridge.com.tr/en/blog/data-catalog-data-dictionary.html) is where everyone finds the definitions.
4. **Measure data quality.** Duplicate customer accounts, inconsistently coded products and empty mandatory fields are identified before loading. We cover this in detail in [data quality and duplicate records](https://digitalbridge.com.tr/en/blog/data-quality-duplicates.html).
5. **Choose the platform.** PostgreSQL, MSSQL or cloud warehouses such as BigQuery and Redshift; the choice depends on data volume, existing licences, where personal data must be held and your team's skills. We break down the cloud cost items in [cloud migration cost](https://digitalbridge.com.tr/en/blog/cloud-migration-cost.html) and the warehouse's own budget lines in [data warehouse cost factors](https://digitalbridge.com.tr/en/blog/data-warehouse-cost-factors.html).
6. **Start with one area.** Build one data mart — sales or finance, say — end to end, put the reports in front of users, then expand. A good first scope for finance is [cash flow reporting automation](https://digitalbridge.com.tr/en/blog/cash-flow-reporting-automation.html), which combines bank, receivables and payment plans.
7. **Automate and monitor.** ETL jobs should be scheduled and re-runnable; failed loads should be logged and reported to an owner. An ETL job that stops silently leads to decisions on stale data.
8. **Plan access and KVKK.** Under KVKK, Türkiye's Personal Data Protection Law No. 6698 (the local counterpart of GDPR), a warehouse needs more careful permissions than its sources because it brings many systems' data together. Masking, [retention and disposal periods](https://digitalbridge.com.tr/en/blog/data-retention-disposal-policy.html) and access logs for personal data are designed from the start.

## How does Digital Bridge build data warehouse and ETL projects?

On Digital Bridge's [data warehouse and ETL integration](https://digitalbridge.com.tr/en/data-analytics/data-warehouse.html) projects, the setup we deliver includes:

- **Scheduled, re-runnable ETL.** Data is pulled from the sources at the agreed interval and transformed to your business rules. Failed loads are logged, retried automatically and reported to the owner.
- **A central warehouse modelled as a star schema.** We choose PostgreSQL, MSSQL or a cloud platform based on your data volume and existing systems, and historical figures are preserved when prices, accounts or the organisation change. For the cloud-versus-on-premises decision our [cloud migration and infrastructure consultancy](https://digitalbridge.com.tr/en/consultancy/cloud-migration-infrastructure-consultancy.html) team joins the process.
- **Quality checks at load time.** Empty fields, duplicate records, format errors and broken references are caught during loading, and faulty records are held back before they reach a report. Where the root cause must be fixed at source, we continue with [data governance and quality](https://digitalbridge.com.tr/en/data-analytics/data-governance-quality.html) work.
- **CDC transfers that spare the live system.** Critical data such as stock and orders arrives within minutes, finance data daily in batch; because only changed records move, the ERP does not slow down.
- **Source system connections.** We build connectors for Logo, SAP, Mikro, Netsis and bespoke ERPs, CRMs and marketplace APIs, drawing on our [system integrations](https://digitalbridge.com.tr/en/technology-hardware/system-integrations.html) experience.
- **Consumption layer.** On top of the warehouse we build [BI dashboards](https://digitalbridge.com.tr/en/data-analytics/bi-dashboard.html) and models such as [demand forecasting analytics](https://digitalbridge.com.tr/en/technology-hardware/demand-forecasting-analytics.html), all fed from the same single source.

Operational systems can be sources too. [SmartPass](https://digitalbridge.com.tr/en/smartpass.html), Digital Bridge's access control, time and attendance and canteen product, exports each closed month's per-person attendance summary as CSV from the panel or as JSON via the API, and canteen reports as Excel or CSV. Loaded alongside ERP data, these files put headcount and attendance next to production and cost figures in the same warehouse.

We do not sell off-the-shelf packages; after a needs analysis we provide a written proposal covering scope, phases and cost. For neighbouring topics such as data quality, KPIs and GIS, see [our full Data & Analytics guide](https://digitalbridge.com.tr/en/blog/topic/veri.html).

## Does your business need a data warehouse? A three-question check

To judge whether you need a data warehouse, ask yourself three questions. How many systems do you export from to build the management report? Do two departments report the same KPI with different numbers? Can you make a reliable like-for-like comparison with last year? If even one answer bothers you, a warehouse project that starts with a single data area is worth considering. [Get in touch](https://digitalbridge.com.tr/en/contact.html) and we will map your sources with you and scope the first data mart.

## Frequently Asked Questions

### What is the difference between a data warehouse and a database?

An operational database (ERP, CRM) is designed to record day-to-day transactions quickly and usually holds only the current state. A data warehouse combines data from many systems for analysis, transforms it using shared definitions and keeps its history. When reports run from the warehouse, the live system is not slowed down. A warehouse also makes reliable year-on-year comparisons possible.

### What is the difference between ETL and ELT?

In ETL, data is transformed outside the warehouse and loaded clean; in ELT, raw data is loaded first and transformed inside the warehouse. ELT is common on cloud warehouses with plenty of processing power, while ETL suits on-premises servers and environments with strict data rules. In practice many projects combine the two approaches.

### Is a data warehouse the same as a data lake?

No. A data warehouse holds structured, cleaned and modelled data for reporting. A data lake stores raw and unstructured data — sensor streams, logs, images — at low cost. Many projects use both: raw data stays in the lake, and the part needed for reporting is cleaned and loaded into the warehouse. That gives flexibility and dependable reporting at the same time.

### Do SMEs need a data warehouse?

Not if simple reports come from a single ERP. If data from several systems has to be combined, reports slow the ERP down, or historical comparisons cannot be trusted, a small warehouse starting with one data area makes sense for an SME too. Starting with one data mart, such as sales or finance, ties the investment to the first report people actually use.

### How is personal data protected in a data warehouse?

Because a warehouse brings many systems together, permissions, masking and access logging must be designed from the outset. Personal data not needed for reporting is either left out or masked, and retention periods are kept consistent with your personal data inventory under KVKK, Türkiye's data protection law. Who queried which data is logged as well.

### What platform is used for a data warehouse?

A data warehouse can be built on on-premises databases such as PostgreSQL or MSSQL, or on cloud warehouses such as BigQuery and Redshift. The choice depends on data volume, existing licences, where personal data must be held and your team's skills. Cloud warehouses suit high-volume ELT, while on-premises servers and ETL can be the better fit where data rules are strict.

### What is a slowly changing dimension?

A slowly changing dimension is a data warehouse technique that stores changes to customer, product or organisation details with both old and new values and their dates. Source systems usually hold only the current state. With this technique, when a customer moves to another sales rep or a price changes, historical figures stay correct and year-on-year comparisons remain reliable.