NorviTech

NorviTech / Consulting / DA2 into a warehouse

Getting DA2 data into Google, Azure or your warehouse

Discern Analytics 2.0 is good at producing an extract. Reporting from somewhere you own means moving that extract into a warehouse, on a schedule, without losing or doubling a day.

The short version

A scheduled DA2 report writes a file. A small, boring pipeline picks it up, checks it, reshapes it and loads it into the place your organisation already reports from: Google Sheets or BigQuery, Azure SQL, Fabric or a lakehouse, Snowflake, or an on-premises warehouse. Power BI, Tableau or Looker Studio then read the warehouse, not the extract.

The destination matters less than people expect. The same four decisions make any of them dependable.

Four decisions that matter more than the destination

1. Load by day, and make a reload harmless

Key each load on the day the data describes and replace that day whole, rather than appending rows. Then re-running yesterday, or loading a late file, corrects the warehouse instead of doubling it. Append-only loads are how a dashboard slowly drifts from the source.

2. Check the shape before you load

DA2 reports get edited: a column renamed, a prompt changed, a filter added. Validate every file against the columns the load needs, and refuse it with a clear message when they are missing. A load that quietly reads a renamed column as empty produces zeros that look like a real bad day.

3. Keep the raw file

Land each extract untouched before transforming it, and keep it for as long as your retention policy allows. When a number is questioned in three months, the raw file is how you prove what the source said on that day, and how you rebuild history after fixing a transformation.

4. Treat the destination as a place PHI now lives

The moment patient-level rows leave Millennium, the cloud tenant holding them needs the same care as the source: a Business Associate Agreement with the provider, access through groups rather than personal shares, a service identity with the narrowest scope that works, and aggregation wherever the report does not need the patient. Many dashboards need no identifiers at all.

By destination

Google

Google Sheets suits a daily dashboard of a few thousand rows that people already live in; a service account writes to a shared drive the organisation owns. Past that size, BigQuery holds the history and Looker Studio or Sheets read summaries from it.

Azure

Land the files in storage, load them with a scheduled pipeline into Azure SQL, Fabric or a lakehouse, and report in Power BI. It fits organisations already on Microsoft 365 and Entra ID, because access follows the groups they already manage.

Snowflake and others

Stage the files and copy them into tables, then build a modelled layer with a clear grain, one row per encounter, charge or appointment, so every report agrees on what it is counting.

Whichever you choose, the reliability rules for the extract itself still apply: pick the file by its header, map the file date to the data date, and catch up on missed runs. Automating scheduled DA2 extracts covers those.

Next

Oracle, Oracle Health, Cerner, Millennium and Discern are trademarks of Oracle and/or its affiliates. NorviTech is independent and is not affiliated with, sponsored by or endorsed by Oracle.