Opportunity Radar

How it works, and what it needs to run

It turns the quality-reject data you already keep into a ranked, dollar-first list of the five fixes worth chasing now. No new data collection, no new system of record.

The flow, start to finish
Step 1
Your reject data

The same quality-reject records you already track on the board today. Pulled in as a table.

Power BI / export
Step 2
The AI agent does the analyst's work
  • Groups similar defects across similar parts
  • Ranks by yearly savings, discounted by how chase-able each one is
  • Flags what is rising and projects the cost if ignored
  • Writes the plain-English why and the first step to check
Azure AI
Step 3
Your Top 5 board

The five most valuable fixes, money first, each with the reason it ranks there and where to start.

Web app
Step 4
You decide, it re-ranks

Mark one handled or skip it, and the next-best problem moves up. The board always shows five.

Live
Where it runs: this demo uses synthetic data and is hosted on Cloudflare, so it is safe to show anywhere. In production it runs inside your Azure tenant on the real data, so nothing ever leaves company systems. That is the version IT and ISO would sign off on.
What to ask Chris for
"Can you export the reject-detail board for the last 6 to 12 months as one flat table (CSV or Excel) with these columns?"
It is small and simple: one table, no joins, no geometry or dimensions. Even a few hundred rows is enough to build and test from. Everything below already lives on that board.

The seven marked Need are the minimum to rank opportunities. The three marked Nice just sharpen it.

FieldExampleWhat the tool builds from it
date Need2026-05-14Trend (getting worse), recency, cost if ignored
part_number NeedM667233660Groups parts into families, concentration
production_line Need49Concentration, and where to go look
defect_type NeedPlating pitsDefines the opportunity and the grouping
quantity_rejected Need12Volume and frequency
cost_impact_usd Need384.00The money: annualized savings and the ranking
supplier NeedApex CastingsSupplier signal (your sourcing lens)
part_description NicePac R tube 36in, frontSharper grouping by shape (front vs rear)
process NiceZinc platingProcess-level grouping and recommendations
disposition NiceReworkA more accurate cost number

Note: geometry and exact dimensions are not needed. The tool infers part family and shape from the description and part number. If the dollar cost is not already on the board, we only need a cost-per-part (or the rework and scrap cost) to turn quantities into dollars. The sample file below is about 1,200 records over 15 months, which rolls up into roughly 30 ranked opportunities (about $592K of cost of quality).

Open the live demo Download the sample data (CSV)
Opportunity Radar, prototype on synthetic data. Built to move into Azure for production with the real reject data.