BRKZ has moved AI from answering questions to closing transactions: its agent Nusa reads photos of delivery notes sent via WhatsApp, matches them to orders, verifies quantities and closes deliveries. The agent now handles the majority of cases in the reconciliation workflow, with human staff left to review uncertain matches and exceptions. For a business built on bulk cement and fragmented supply chains, this matters because coordination work becomes executable by software.
From WhatsApp photos to closed deliveries
Bulk cement delivery outside Riyadh illustrates the change. A transporter unloads material and sends a photo of the delivery note, which previously triggered manual reading, order search, quantity checks and system updates. Nusa now extracts details from the image, compares them with internal systems and completes the reconciliation. Its share of closed deliveries keeps growing every month, while the team focuses on edge cases.
The result rests on three years of system building. BRKZ moved RFQs, pricing, supplier relations, orders, deliveries, credit and collections from phone calls, WhatsApp messages, spreadsheets, PDFs and personal knowledge onto shared systems. Every RFQ, quotation, supplier interaction, delivery and payment began leaving a digital trail, accumulating into tens of millions of structured data points across products, suppliers, transactions and payment behavior.
That history enabled a sequence the company describes as systemize, capture, understand, automate, agentize. Without proprietary context, a general model would work on public information only, so BRKZ first made operations observable. Then it built a pricing engine on historical RFQs and transactions, now live as Mizan, an AI procurement agent for first product categories that produces near-instant price recommendations for officers to review before quoting.
What agent-owned transactions mean for business
For companies working with physical goods, the practical effect is fewer handoffs between sales, procurement, operations, logistics, finance and credit. Each request no longer waits while someone copies data, seeks approval or forwards messages. A small firm can process more orders without adding support staff, while a large one can let one salesperson or procurement officer oversee far more transactions as agents handle preparation, analysis and routine coordination.
The limits remain clear and worth checking before adoption. Pricing still depends on quantity, location, delivery terms, timing and market conditions, so Mizan recommendations require human review. Agents operate within defined limits, with people retaining underwriting decisions, relationship ownership and exception handling. The approach also presumes digitized workflows: events outside systems, from late arrivals to quality failures, stay invisible to AI unless captured.
The marker to watch is whether agent-to-agent commerce moves beyond pilots. BRKZ outlines a five-year path where a contractor agent states a need for materials next Tuesday, its agent checks history, pricing and supplier availability, then coordinates supplier and logistics agents while humans set policy. Appearance of such linked transactions across companies would confirm that operating models, not just tasks, are changing.
