An established mid-tier UK wealth manager, part of an international group. The firm set out to modernise a data estate shaped by organic growth and acquisition — laying the foundations to capitalise on, and subsequently implement, a suite of AI offerings.

Organic growth and acquisition had left the client with a complex data estate — legacy systems sitting alongside modern platforms, each holding a different slice of the business. Nothing unusual for a firm of this scale, but it meant a consolidated view of the whole book required pulling data together from multiple tools rather than seeing it in one place.

We were engaged to lead a proof of concept assessing whether Snowflake could solve this — evaluated against an alternative platform already in use elsewhere in the group.

A small, focused team ran a five-week PoC. Rather than a broad technology exercise, it was built around specific business outcomes: proving ingestion from both legacy and modern sources, data transformation to data products, and testing whether a single platform could unify the estate.

The results:

• A single view of the whole book. Data from legacy and modern systems was brought together in one place — something the business had not been able to do easily before.

• Significant reduction in ingestion pipeline setup time, using Fivetran into Snowflake. From there, dbt handled the transformation layer, feeding into a semantic layer.

• Using Snowflake Cortex to allow natural language queries over portfolio data — not something the client had asked for, but a deliberate step beyond the brief to demonstrate what else the platform could unlock.

The PoC gave the business the evidence it needed to move forward with confidence — and set the direction for the security, implementation, and AI work that followed.

An established mid-tier UK wealth manager, part of an international group. Following a successful proof of concept, the client engaged us to lead the technical implementation of their new Snowflake data platform.

With the platform decision made, implementation began with a month spent designing how the platform would sit within the client's network and security landscape. Data modelling came next — the PoC version had given a good flavour of what was possible, but a considered, long-term approach was now needed. Working with subject matter experts, we found that a lot of the business's shared language had quietly been inherited from a single legacy system rather than agreed as a business standard. We proposed the start of a proper taxonomy to unpick this.

On the underlying model, we landed on a Medallion architecture with a Kimball-influenced dimensional model. Other modelling approaches were discussed, but the priority was how the business actually consumes data.

On the infrastructure side:

• dbt was implemented as the transformation layer, with the wider team brought in during a parallel run to build early familiarity

• Terraform was used to provision Snowflake infrastructure, with the client adopting Infrastructure-as-Code as part of the programme

• A multi-environment strategy was designed and agreed to fit the realities of the existing legacy estate

• Data quality was built in from the start — dbt tests, source freshness checks, and Snowflake Streams for change tracking, to catch issues before they reached the business

By the end of this phase, the wider team had moved onto Snowflake and dbt in daily use — a meaningful shift in how the team worked day to day. With the foundation in place and the team beginning to take ownership, focus moved to the client's first production AI capability.

An established mid-tier UK wealth manager, part of an international group. Security-conscious by nature and by regulatory obligation, the client needed a data platform built to best-practice standards from day one — without disrupting the processes the business already relied on.

Security is of paramount importance to the client, and was embedded in the network and access architecture from the outset. The approach followed Snowflake's recommended security model across three layers, while working within the client's existing operational processes rather than forcing wholesale change.

An established mid-tier UK wealth manager, part of an international group. Having built the platform's foundations, the client moved to launch its first wave of AI capability — this work forming one strand of a coordinated push into production AI.

Building on the platform, Snowflake Cortex and its built-in functions were used to process one million unstructured, multimodal documents in support of operations workflows. Transformation was handed to dbt, so the resulting pipeline could be maintained by the wider team, not just its author.

Cost and quality discipline shaped how this was delivered — each approach trialled on small batches before scaling to production volumes. The output was brought together with data from existing sources into one structured, consistent view. Validation was built in throughout to catch hallucinations, with every output traceable back to its source document — a deliberate step to build trust in AI-generated results, not just produce them.

Alongside the pipeline, a Streamlit application was built to let the business interact with the results directly. A further extraction architecture was also proposed, well received at senior level, and is now being taken forward by the client's own team.