Viticulture data pipeline for a Surf Coast winery
End-to-end data pipeline for a mid-sized Surf Coast winery — field sensors, weather, lab results and stock — unifying five tools into one reporting view.
- Client
- Winery (confidential)
- Sector
- Agriculture
- Location
- Surf Coast, Victoria
- Services
- Systems review · Data engineering · Reporting & BI · Documentation
The context
Our client operates a mid-sized winery on the Surf Coast with vineyards across Geelong and the Bellarine. Over a decade of growth, each part of the operation had adopted a different tool: a soil-moisture sensor network, a cellar inventory application, an Excel-based lab results log, a POS for the cellar door, and Xero for finance.
Each was fine individually. Together they couldn\u2019t answer questions as basic as “what\u2019s the margin on the 2023 Pinot by month?” without a full afternoon in spreadsheets.
The shaping phase
This began as an advisory engagement. Before recommending any build, we:
- Mapped every source system, its owner, its update cadence, and its fragility.
- Interviewed the viticulturist, the winemaker, the cellar door manager and the owner-operator.
- Sat with the accountant to understand what questions the board actually asks.
Our recommendation was not to build a warehouse. Instead: a lightweight, auditable data pipeline and a handful of dashboards that answered the questions that mattered \u2014 no more.
The build
- Ingestion: Cloudflare Workers pulling from each source on a schedule, normalising into a single Postgres schema.
- Warehouse: Postgres, not Snowflake. The data set fits comfortably and will for a decade.
- Reporting: Metabase with a curated set of dashboards, owned by the accountant and the owner-operator.
- Documentation: a plain-English data dictionary and a one-page runbook that a new bookkeeper could follow on day one.
Outcomes
- The monthly close was reduced from three days to half a day.
- Lab-to-bottle traceability became a single report, satisfying a due-diligence request from an export partner in two days instead of three weeks.
- The owner cancelled two overlapping SaaS subscriptions, paying for the engagement inside twelve months.
What we\u2019d do differently
The soil-sensor network\u2019s API was unreliable. We built retries and a dead-letter queue early, which we\u2019d do again \u2014 but we\u2019d also set a hard contractual expectation with the sensor vendor at the outset.