The modern stack is not a list of tools. It is a chain of responsibilities.

When an SME starts discussing the modern data stack, the conversation can become vendor-driven very quickly: Snowflake or BigQuery? dbt or plain SQL? Fivetran or Airbyte? Power BI or Looker? These are valid questions, but they come after a more important one: what data operating system can support the business for the next three years?

A good stack must be robust enough to grow and simple enough for the internal team to understand. If it requires skills the company will never have, it creates dependency. If it is too weak, it becomes another centralised spreadsheet.

Why this matters now for SMEs

Eurostat reports that in 2024, 73% of EU SMEs reached at least a basic level of digital intensity, still around 20 percentage points below the EU 2030 target. The European Commission also notes that AI, cloud and big data adoption is improving but needs to accelerate.

Many SMEs are therefore in an intermediate phase: they use digital tools, but do not yet have a real data architecture. The next step is not buying another software product. It is connecting systems coherently.

The minimum structure of a data stack

1. Sources

ERP, CRM, e-commerce, production systems, spreadsheets and operational databases. The first job is to understand which sources contain decision-critical data and which are only noise.

2. Ingestion

This is how data enters the analytical environment. It can start manually, but should become automated and monitored. Tools such as Fivetran or Airbyte make sense when they reduce maintenance, not when they add complexity.

3. Data warehouse

This is the single source of truth. For an SME, cloud warehouses such as BigQuery or Snowflake may be appropriate, but PostgreSQL may be enough for lighter architectures. The choice depends on volume, skills, budget and security requirements.

4. Transformations

This is where raw data becomes metrics. dbt is useful when SQL logic needs to be reusable, versioned and tested. If the team is not ready, simpler transformations can work, provided they are documented.

5. BI and decision routines

Power BI, Tableau, Looker or Metabase are the last mile. The dashboard should not impress; it should be used. That is why it must be designed together with the management routine: who reviews it, when, and what decision is expected.

Three pragmatic architectures

Basic: essential management control

Main sources: ERP, accounting, sales.
Light warehouse or analytical database.
Management dashboard with financial, commercial and operational KPIs.
Goal: reduce spreadsheet dependency and monthly manual work.

Intermediate: multi-source recurring control

Automated ingestion from ERP, CRM, e-commerce or production.
Cloud data warehouse.
Documented and tested transformations.
Dashboards for leadership and functions.

Advanced: analytics and AI-ready

Monitored pipelines with quality alerts.
Documented and governed data model.
Enough history for forecasting and anomaly detection.
Permissions and lineage for AI and sensitive reporting.

The selection criterion: maintenance, not fashion

Every technology carries a hidden cost: maintenance, skills, monitoring, incidents and documentation. The right question is not “what is the best tool?”, but “who will maintain it when the project is over?”.

A boutique data agency should design for autonomy. If the stack remains understandable, the client grows. If it becomes a black box, even the best technology loses value.

Sources cited

Eurostat, Digitalisation in Europe 2025

European Commission, State of the Digital Decade 2025

OECD, SME digitalisation for competitiveness: The 2025 D4SME Survey