Manufacturing analytics, and AI agents that ask a person before they act.

By designing the controls in before the agent is built. The agent reads only the governed tables it is granted in Unity Catalog, every action it proposes, such as matching a receipt to a scale ticket or flagging a supplier lot, is checked against policy before anything changes, and exceptions go to a named person to approve.

Each decision is recorded, so quality and finance can trace it later, and the same checks run in testing as in production. The Field Guide's chapter on governed actions walks through this kind of check, and our AI agent services build it.

Whatever leadership argues about on a Monday, measured the same way every week. For a plant in ramp-up that's usually utilisation against nameplate, backlog in days at the current rate, downtime by cause, yield by supplier and receiving exceptions.

Each one needs a single agreed definition before it gets a chart, which is most of the work. The plant leadership study at /case-studies/plant-leadership-reporting-databricks shows nine of them.

If every number you need is in Business Central, Power BI on the ERP is enough and cheaper. Once a question needs the line historian, the lab, the scale house or market prices, a lakehouse is how those join the ERP under one definition.

Both readings of the same metric then come from the gold layer rather than from two copies of a measure. The trade-off is laid out at /expertise/erp-reporting.

Less than you'd expect. The NMBT work at /case-studies/nmbt was UX research into how engineers search an enormous electro-magnetic parts catalogue, and findability under that kind of volume behaves the same way in distribution, retail and lending. The domain shapes the vocabulary and the compliance surface; the engineering underneath transfers.

From treating lineage as a product feature rather than a logging setting. SmartCert at /case-studies/smart-cert replaced emailed certificate attachments with a traceable exchange between manufacturers, distributors and their customers. On Databricks, Unity Catalog gives you the substrate and someone still has to ship the workflow people will use.

We read from them and don't build them. A line historian, a downtime log or a lab system is a source we land in the lakehouse and join with the ERP. Industrial control itself, the PLCs and SCADA that run the equipment, is a separate discipline, and the people who build it should keep owning it.

The plant leadership study at /case-studies/plant-leadership-reporting-databricks has a ten-week plan to go-live, and the build behind it is 12 bronze, 15 silver and 10 gold tables from seven sources.

Week one is read access to Business Central and exports from the plant systems, because the dashboards were built and tested on sample data first. Starting from nothing, the six-week Lakehouse Launchpad at /databricks/launchpad stands the governed foundation up.

Yes. We're an Insight Works partner, and we implement Warehouse Insight, Shop Floor Insight, License Plating and Dynamic Ship on Business Central and support the warehouse and production transactions that run through them, on site where that helps. The detail is at /expertise/business-central-insight-works.