Share with your CDO
Fonterra’s Global Ingredients division retired a decade-old on-premises analytics system, the Market Analytics System, that was consuming half its engineering bandwidth on maintenance alone, and rebuilt on Databricks with AI query agents sitting directly on top of the new data layer. The team completed the migration ahead of schedule and under budget, decommissioning physical basement servers and replacing fragile web scrapes with API connections. Within the first month on the new platform, business users were interrogating milk supply forecasts and pricing models in minutes, without touching Excel or filing IT requests.
What this means for your business
The Fonterra story is a clean diagnostic for any data organization still running legacy analytics infrastructure: when engineering teams spend half their time on maintenance rather than building, the platform isn’t supporting the business, it’s consuming it. That 50% maintenance tax is the tell. CDOs sitting on aging on-premises systems, even well-loved ones with years of institutional logic baked in, should read this as a benchmark, not a case study. If your ratio looks anything like Fonterra’s pre-migration numbers, the question isn’t whether to move, it’s how long the delay is costing you in deferred insight.
The architectural choice worth examining isn’t Databricks specifically, but the decision to rebuild rather than lift and shift. Migrating existing problems to new infrastructure is one of the most predictable failure modes in data modernization, and Fonterra explicitly avoided it by auditing legacy requirements and recutting the logic from scratch. That’s harder and slower up front, and it’s why so many teams skip it. But it’s also why they landed ahead of schedule: they weren’t inheriting technical debt on a faster machine, they were clearing it. The Databricks Genie layer, a natural-language AI interface that lets business users query datasets directly, only works cleanly when the underlying data model is trustworthy. You can’t bolt an AI interface onto a messy foundation and expect coherent answers.
The competitive pressure this creates is subtle but real. When analysts can validate a market hypothesis in minutes rather than days, the organizations that still require Excel exports and IT queue time are running a structurally slower decision loop. That gap compounds. CDOs who treat AI interfaces as a presentation layer to add later, once the data is “ready,” are likely to find that readiness is a moving target, and the business users have already found workarounds. I’d revise this read if Fonterra’s adoption numbers six months post-launch show the Genie tools collecting dust, but the two delivered use cases inside the first month suggest the demand was real and waiting.
Concept deep-dive: Natural-language data interfaces
A natural-language data interface, think of it as a search bar for your data warehouse, lets business users ask questions in plain English and receive structured answers drawn from the underlying dataset, without writing SQL or requesting analyst support. Databricks Genie is one implementation. These tools only return reliable answers when the data beneath them is clean, well-governed, and semantically consistent. Their business value is access democratization: the bottleneck shifts from “can IT build this report” to “is the data trustworthy enough to query directly.”
Based on reporting from Inside A Fonterra Division’s Data Overhaul: From Basement Servers to AI-Powered Insights, originally published 2026-09-03 21:06:00.
