“Last year, I walked around this conference. I was curious, but sceptical. At Ceres Pharma, we’d just had a failed data initiative, and I wanted to find a solution. One year later, I’m on stage.” – Henry De Rudder, Head of Data, AI & IT, Ceres Pharma at Data Innovation Summit 2026, Stockholm

The problem wasn’t data. It was trust.

In May 2025, Henry De Rudder walked into the Data Innovation Summit in Stockholm as a visitor. He browsed the booths, sat through the talks, and sized up the modern data stack. He was curious. He was also sceptical. At Ceres Pharma, the PE-backed pharmaceutical group he’d recently joined as Head of Data, AI & IT, a previous data initiative had just failed – and the executive team had no appetite for another.

Ceres Pharma specialises in over-the-counter products and prescription drugs in women’s and family health. Since 2018 the company has completed 14 acquisitions, expanding from Belgium into Hungary, Romania, Bulgaria and Italy. Growth was fast. Integration was not. Each acquisition brought its own systems: Exact Online and Salesforce in Belgium, Navision and SAP R3 in Hungary, Senior in Romania, Sell Matic in Bulgaria, Sage X3 in Italy. Layer on multiple CRMs, e-commerce platforms, third-party logistics providers and procurement tools, and you get an IT landscape where the same product appears under different codes in different countries and the same customer has different names depending on which entity you ask.

The result was predictable. Finance and commercial could not agree on what the numbers meant. There was no single source of truth – only fragmented sources, duplicate dimensions, and endless arguments about whose spreadsheet was right.

“We didn’t have a data problem. We had a trust problem.”

A previous initiative – built on Azure Synapse, SQL Database and Power BI with an external partner – had produced polished dashboards but untrustworthy data. The executive team wrote off the investment. As Henry puts it: “We burned quite some cash on a previous project.” The board became data-sceptical, and the data team found itself sidelined – a technical function that built things nobody trusted, then got pushed to the margins when those things failed.

Henry’s brief was blunt: prove the data is right before anyone sees a dashboard.

Best-of-breed, not lock-in

Henry faced two paths. He could centralise five ERPs into one system – a multi-year, multimillion-euro programme with enormous change management overhead – or he could build a governed analytical layer on top of the existing systems. He chose the second.

The stack: Snowflake as the cloud data warehouse, dbt Core for transformation and governance, and Astrato as the analytics and application layer. Each component was benchmarked independently. Snowflake was tested against Databricks and Microsoft Fabric; Astrato against Sigma and others.

“I benchmarked Astrato against Sigma and other solutions. For us, Astrato was the best – because you get the best of both worlds. You have classic analytical reporting, but you can also build apps with controlled write-back to Snowflake.”

The warehouse-native architecture was a decisive factor. Henry needed a BI tool that lives on Snowflake, not next to it – one that uses Snowflake’s own compute and query power rather than extracting data into a separate layer. That meant no duplication overhead, no sync issues, and no middle layer to maintain.

“If we ever need to change a component, we can swap it easily. Each tool does what it’s best at.”

The delivery model changed too. The previous initiative had relied on an external partner with validation only from corporate finance. Henry brought data architecture, modelling and governance in-house, with Proxify – an implementation partner he found at that same Data Innovation Summit in 2025 – providing specialist talent to accelerate. The POC began in July. By December, five months later, the executive team approved production.

Beyond Dashboards: Data Apps that create governed data

The conventional story of a BI modernisation ends with better dashboards. Ceres Pharma’s story is different. The most valuable things Henry’s team built in Astrato are not dashboards – they are operational data applications that actively create and govern data. This is the distinction that matters: these apps don’t just visualise what’s in the warehouse. They write back to it.

The Metadata Manager

Ceres Pharma’s central problem was dimensional chaos. The same product coded differently across five ERPs, with no corporate standard. Their Belgian ERP had only four product dimension fields. They needed twelve. A traditional solution would be an expensive product information management tool – which Ceres had, in fact, already purchased for one country at considerable cost. They were using perhaps 5 or 10 per cent of what they’d paid for.

Instead, Henry’s team built the Metadata Manager: a data application inside Astrato that displays all local product codes from every source system side by side, lets business users – the domain experts who actually know the products – map them to a single corporate standard, and writes the unified dimensions back to Snowflake. dbt then enforces governance rules on the clean data. Every change creates a new row in Snowflake, never overwriting history. The result is a full audit trail, from source system through to corporate code, visible to any user directly within the application.

“Our ERP had only four fields for product dimensions, but we needed twelve. We’d already invested in an expensive product information tool – and we were only using 5 or 10 per cent of it. So we built this app with write-back. Much faster, much cheaper, and it’s being used correctly now.”

The write-back capability is what makes this possible. The Metadata Manager does not just consume data from the warehouse – it creates governed data and sends it back. Astrato straddles the silver and gold layers of the architecture: it reads from silver, lets business users enrich and unify, and writes the result to Snowflake where dbt governs it into gold. The hard part of any data transformation, as Henry sees it, isn’t building dashboards – it’s building the layers underneath that make dashboards trustworthy.

The Reconciliation Tool

With dimensions unified, the next step was facts. Henry’s team built a reconciliation tool that benchmarks transactional data against the financial truth per entity. In a pharmaceutical group where finance is the final authority, this was non-negotiable: no analytical layer would be trusted until it matched the general ledger.

The tool traces discrepancies to their source, enabling structured conversations between finance and commercial about specific variances rather than blanket disagreements about whose numbers are right.

“Six months ago, this was impossible to think about. Now our finance and data teams are agreeing on the same numbers.”

Self-Service analytics

On top of the governed gold layer, the team built self-service analytical views that go well beyond static reporting. Users choose their own dimensions, their own measures, and build their own tables – then bookmark them for later. An information tab explains how every measure was calculated, when the last invoice data was available, and how mappings were applied. Transparency is built into the application, not bolted on after the fact.

The acid test came from the CFO’s office. Rather than requesting another set of pre-built dashboards, the finance team wanted the ability to explore the data on their own terms.

“I don’t care about dashboards. I want to choose my dimensions, my facts, and make my own analysis.”

That shift – from consuming pre-packaged reports to conducting independent analysis on trusted data — is what self-service actually means in practice. It’s not a feature. It’s an outcome of getting the foundation right.

“It’s not just static, classical BI reporting. You have self-service analytics where people can do their own ad hoc analysis.”

From technical team to strategic partner

The shift was not only architectural. A year earlier, the data team had been a back-office function – building tools nobody trusted, then getting sidelined when those tools failed. The new platform changed the team’s standing within the organisation.

“We’re no longer the technical team pushed to the side. We sit between finance and commercial. We understand what they need to see, and we bring the tools they can build on.”

Business owners now own their data. Henry’s team builds the tools and the governance framework; the business users – the people who actually understand the products, the customers, the invoices – input, validate and take responsibility for what goes into the system. The Metadata Manager is the clearest expression of this principle: domain experts, not data engineers, decide what the corporate product standard should be.

“Isn’t it too early?”

Henry posed the question himself, on stage, before anyone in the audience could. His CFO, sitting in the front rows, had asked him to manage expectations. His answer was disarming.

“Are we finished? Totally not. We need more user adoption, more use cases, more applications. But the hardest part is done – the foundation.”

The hard part of any data transformation, he argued, is not building dashboards. It’s building the layers underneath that make dashboards trustworthy. Bronze, silver, gold – all sources ingested, all dimensions unified, all facts reconciled against finance. That work is done.

The architecture is designed for the acquisition model. When Ceres acquires a new company – and there will be more – the process is now repeatable: connect the raw sources to the bronze layer, run them through the Metadata Manager, reconcile against finance, and analytics are live. What previously took years, or never happened at all, now follows a playbook. Belgium is already live on the full platform. Hungary’s corporate teams are next in the queue.

Usage monitoring inside Astrato shows growing adoption since September. Cost monitoring – using a template freely available within Astrato – tracks Snowflake compute and credit consumption, ensuring the platform scales economically as adoption grows.

“You can plug in the cost monitoring template and see exactly how many credits you’re using and what the average query timing is. For us, that’s essential.”

What’s next: AI on governed data

The same governed data layer that serves dashboards and data apps today will serve AI agents tomorrow. Ceres is enabling Snowflake Intelligence for conversational analytics – but with guardrails. There is a verified question set, a feedback loop that tracks the most-asked questions to optimise the data model, role-based access, an executive pilot before broader rollout, and cost monitoring against tokenisation.

“We built a business question loop – we see what the most-asked questions are, verify those queries, and optimise the model to get better answers.”

The principle is the same one that governed the entire programme: do the boring part first. Build the foundation. Earn trust. Then move forward.

‍

‍

“We didn’t set out to build a data platform. We set out to rebuild trust in data. The platform was how we got there.”

‍