Bidvest Plumblink · Distribution and procurement
Procurement intelligence for Bidvest Plumblink
An AI-powered procurement intelligence platform on Microsoft Fabric, Azure AI Foundry and Entra ID, hosted in Azure South Africa North.
Built with
- Microsoft Fabric
- OneLake
- Power BI semantic model
- Azure AI Foundry
- Azure OpenAI
- Entra ID
- React
- FastAPI
The challenge
Turning ERP data into daily procurement decisions: what to reorder and from which supplier, which stock is slowing down, and where the stock master needs attention.
A decision dashboard
Recommended reorders by supplier, slow-moving and dead stock, min and max coverage, demand and price anomalies, and stock redistribution, with ranked actions that export to Excel or a draft purchase order.
A daily AI briefing
A daily briefing with prioritised insight cards, so buyers start with the decisions that matter most.
An assistant over governed data
Buyers ask in plain language for supplier scorecards, branch comparisons, sales history or stock composition, and the assistant answers from the same data the dashboard uses.
A stock-master health check
Data-quality rules with drill-down show which records need fixing at the source.
How it is built
Three layers on one governed copy of the data.
- Data foundation
A medallion pipeline in Microsoft Fabric, from the ERP through bronze, silver and gold.
- Intelligence
A governed semantic model and machine learning models on the curated gold data.
- Delivery
A secured API serving a React dashboard and an Azure AI Foundry assistant, signed in with Entra ID and delivered as an Azure Managed Application.
- ERPSource dataDaily data from the ERP01
- BRZBronzeRaw data lands in Fabric02
- SLVSilverCleaned and conformed03
- GLDGoldModelled for the business04
- AISemantic model and MLGoverned definitions and predictions05
- AGTDashboard and assistantBriefing, ranked actions, answers06
- YOUThe buyer decidesDraft order exported, person approves07
AI principles
Built to be trusted.
Advisory by design: the assistant informs decisions and drafts actions, and people decide. It cannot place orders.
Grounded: numbers are checked against the source data, and an insight that cannot be grounded is not shown.
No training on customer data, and every AI tool call is logged and attributable.
Stack and services
What we used, and how we can help you.
- Microsoft Fabric
- OneLake
- Power BI semantic model
- Azure AI Foundry
- Azure OpenAI
- Entra ID
- React
- FastAPI
Data and Microsoft Fabric
One governed data foundation, from the ERP to the board report to the AI agent.
ExploreAgentic AI
Agents that answer from your governed data, draft the next step and hand real decisions to the right person.
ExploreMicrosoft Marketplace
Package your product, publish it on Microsoft Marketplace and deploy it into your customers' own Azure.
ExploreBook a discovery call
Let's build what's next.
Tell us about one decision you want AI and data to improve. We will come back with the smallest useful first step.