How a Cybersecurity Giant Cut Data Pipeline Delivery Time Fourfold, Mid-Acquisition

Sophos protects 600,000 organisations from cyberattacks. Behind that mission sits a data platform that, until recently, was creaking under its own weight.

The company was running two legacy data integration systems side by side: Informatica, which had become expensive and hard to manage, and an earlier version of Matillion that had done its job well but was starting to hit real limits as the business scaled. Together, the two platforms carried more than 800 production pipelines feeding revenue forecasting, HR operations, and analytics across the business.

Then Sophos closed the largest acquisition in its history, and the migration workload doubled overnight. The company modernised its data infrastructure with Maia, an agentic data engineering platform, in the middle of that upheaval.

'We had a lot going on,' said Jason Mulvin, Director of Enterprise Data at Sophos. 'Frankly, if you think about trying to do an upgrade at the same time, it's not really a great idea. But we did it'.

The Problem Wasn't Just Volume; It Was Trust

Sophos had a strict internal rule, enforced by its GenAI Council: no customer or employee data goes to a third-party AI tool. That's a simple rule to apply to a chatbot. It's a much harder one to apply to ETL, where the pipelines themselves move PII, financial records, and workforce data every day.

The team found its answer in the way Maia runs: as a Snowflake Native App. Because pipeline execution happens entirely inside Sophos's own Snowflake environment, no data ever crosses to a third-party endpoint.

'Once the data lands in our environment, Maia is executing that within our Snowflake perimeter, and the data never has to leave until it's ready to be consumed,' Mulvin said. 'That was a huge win for us'.

The GenAI Council approved the platform in October 2025, and Sophos moved from pilot to production as an early adopter.

Skills, Not One Giant Context File

One early lesson shaped everything that followed. The team first tried to load all of Sophos's engineering standards into a single context file for Maia. It didn't hold up at scale. Breaking that context into modular, purpose-built Maia Skills, one set per pipeline type, is what let the team apply consistent standards across every build.

That structure now underpins three workstreams: migrating Informatica mappings to cloud-native pipelines, modernising more than 800 legacy Matillion pipelines, and folding generated pipelines into a Git-backed CI/CD process so they move to production as versioned, iterable artefacts rather than one-off builds.

The Numbers

Across the full pipeline lifecycle—design, development, documentation, testing, and deployment—Sophos now delivers in two to five hours what used to take 10 to 16. That's roughly 4x faster, end-to-end. First-time deployment success sits above 95%, and every Maia-generated pipeline meets 100% of Sophos's engineering standards, checked during the build rather than after it.

Robman Balisi, a Principal Data Engineer on the team, put it in more concrete terms: a proof-of-concept pipeline that used to take two to three days now takes under an hour.

'So many people talk about AI helping generate code faster,' Mulvin said. 'All that's true, but it's really the full life cycle where you start to see those gains'.

What the Freed-up Time Buys

With less time spent maintaining pipelines, Mulvin says the data engineering role itself is changing shape. 'The blocking and tackling of ETL development goes away,' he said. 'They're free to look more at the big picture and understand the business context of what they're trying to do'.

Sophos is putting that reclaimed time toward two next steps: opening pipeline creation to business teams as citizen developers under proper governance, and extending its Snowflake-based security perimeter to cover a wider set of data consumers, including AI applications and autonomous agents, not just reporting teams.

It's a platform built for whatever consumes its data next, whether that's a reporting team, an application, or an autonomous agent.

Visit maia.ai to read more customer stories.