Databricks Lakehouse in private banking: what is really at stake behind the migration
In Swiss private banks, the question is no longer whether client and portfolio data flows will one day leave the legacy data warehouse. It is how to move them to a Lakehouse without losing regulatory control on the way.
Databricks is often the chosen tool, but the tool is not the issue. We see these programmes up close. Here is what is really at stake, and why success depends less on the platform than on governance and on the people who carry it.
Why banks are migrating now
Two regulatory deadlines explain the timing. FINMA Circular 2023/1 on operational risks and resilience has applied since 1 January 2024, with a dedicated chapter on critical data: inventory, classification and protection across the whole lifecycle.
Its operational resilience requirements, with impact tolerance thresholds per critical function, were expected by 1 January 2026. And the final Basel III framework has applied in Switzerland since 1 January 2025, with the first reports in the revised format.
In practice, a bank must prove where every figure in its prudential reporting comes from, who accessed it and how it was transformed. With ten source systems, Excel extracts and an ageing warehouse, that proof costs weeks at every closing.
The observation Databricks shared at CBA Live in April 2026 matches ours: banks do not have an AI problem, they have a data platform problem. Pilots stall because the foundation and the governance do not keep up.
What "Lakehouse" means in practice
A Lakehouse brings together, on one foundation, the raw storage of a data lake and the transactional guarantees of a warehouse. At Databricks, that foundation rests on the open Delta Lake format and on a layered organisation often called the medallion architecture.
Bronze: raw data as it leaves the core banking system, the PMS or the CRM, timestamped and immutable.
Silver: cleaned, deduplicated data, reconciled across systems, with quality rules applied.
Gold: business-ready datasets for regulatory reporting, risk, wealth management and self-service analytics.
The benefit for a bank fits in one sentence: the same gold dataset feeds FINMA reporting, relationship managers' dashboards and risk models. One definition of the client, the portfolio and the position, instead of three drifting copies.
Databricks cites a leading global bank that cut its liquidity reporting processing from ten hours to eight minutes after unifying its data. The order of magnitude speaks, even if every context differs.
Unity Catalog, the real heart of the matter
If one component deserves the executive committee's attention, it is Unity Catalog. It is the layer that technically answers the critical data requirements of Circular 2023/1.
Lineage: every gold table traces back to its sources, transformation by transformation. That is the auditability evidence FINMA asks for.
Attribute-based access control (ABAC), generally available since 2026: row filtering and column masking defined once, enforced everywhere.
Governed tags and automatic classification of sensitive data, which make the critical data inventory tangible.
Centralised access auditing, essential for banking secrecy and data protection.
At the Data + AI Summit in June 2026, Databricks extended the catalogue to AI agents, models and tools through Unity AI Gateway, and announced a Governance Hub in preview. The logic is clear: govern AI and data in the same place, under the same rules.
For a bank, this changes the nature of the project. Compliance stops being a documentation exercise run after the fact. It becomes a by-product of the architecture, provided it was designed in from the start.
The pitfalls we observe in the field
The first pitfall is the like-for-like migration. Copying the old warehouse tables without rethinking the data model moves technical debt into the cloud, with a compute bill on top.
The second is the catalogue installed too late. Configuring Unity Catalog after six months of pipelines means redoing permissions, schemas and lineage from scratch. It must be in place before the first line of PySpark.
The third concerns location. Client data stays in a Swiss Azure region, and some institutions keep part of their estate on premises. Databricks announced in June 2026 a hybrid storage ecosystem precisely to govern data that will never migrate.
The fourth is cost. Serverless compute and AI workloads escape budgets quickly if nobody sets caps per team. A serious bank treats cost governance the way it treats access governance.
None of these pitfalls is a tooling problem. They are architecture and organisation decisions, taken early or paid for later.
The skills that make the difference
On these programmes, the scarce resource is not the licence, it is the profile. Three roles carry the outcome, and they are hard to find in Switzerland with banking experience.
The Azure/Databricks Data Engineer: PySpark, Delta Lake, Azure Data Factory, SQL and Unity Catalog, with knowledge of a bank's client, position and transaction flows.
The platform architect: Lakehouse design, catalogue and lineage governance, DevSecOps applied to data, and a working dialogue with compliance and internal audit.
The Analytics Engineer: Databricks SQL, dbt, Power BI, Python, to deliver self-service to wealth management and risk teams without bypassing governance.
Our sourcing experience on these profiles is unambiguous: an architect who has already answered an auditor about the lineage of a regulatory figure is worth more than three certifications. Banking context is not learned in a notebook.
The combination that works pairs a core team in Geneva or Zurich, close to compliance and the business, with nearshore reinforcements to industrialise the pipelines. All under a single catalogue governance.
What we take away
A Databricks migration in private banking is a data governance project disguised as a technical one. Unity Catalog, lineage and critical data classification answer FINMA Circular 2023/1 directly and ease final Basel III reporting.
The platform is mature. What is most often missing are the architecture decisions taken early and the people able to hold the conversation between engineering, risk and audit.
If you are preparing or strengthening a programme of this kind, let's talk. We will tell you plainly where the risk sits, on the platform or in the team.
Comments