Supply’d, the Melbourne ERP platform built for food manufacturers, wholesalers and retailers, expanded its Supply’d Intelligence suite on October 28, 2026, with version 5.5, introducing an AI assistant that answers questions from a company’s own business data, an Expense Agent that investigates supplier invoice discrepancies, and read-only Model Context Protocol connectivity to compatible AI assistants, with approval controls governing any changes it proposes.
Ask Supply’d: Answers From Your Own Data
The headline feature is Ask Supply’d, an AI-powered assistant launching in beta that lets food manufacturers, wholesalers and retailers ask questions inside their ERP and receive answers based on authorised business data and Supply’d help documentation. The point, Supply’d says, is to bring AI into the workflows teams already use, rather than treating it as a separate tool bolted on the side.
Food businesses, the company argues, work with changing supplier prices, perishable inventory, complex purchasing and production requirements, and high volumes of documents. Information about a supplier, a price or a delivery is often buried across several reports, and finding it costs time. Ask Supply’d is meant to cut that search down to a question, answered from the data the user is already permitted to see.
Crucially, the assistant respects existing user permissions and is read-only in its initial release, so it cannot alter operational records. That constraint is deliberate: in a business where a wrong stock figure or a changed price can ripple into production and delivery, an assistant that only reads is a safer first step than one that writes.
Expense Agent: Chasing Invoice Discrepancies
The Expense Agent goes further, and it is launching as a private beta with selected customers. It reviews supplier invoices, investigates discrepancies and recommends corrective actions. Depending on the issue, it can update invoice lines, correct tax details, schedule price changes or request credits from a supplier.
Actions require approval by default, while businesses can selectively permit automatic approval as they gain confidence in the system. That approval gate is the product’s most important design choice: invoice reconciliation is exactly the kind of repetitive, detail-heavy work that AI should help with, but it is also work where a confident mistake costs real money. Keeping a human in the loop until trust is earned is the right trade-off, and the private-beta rollout suggests Supply’d knows it.
Supply’d founder Jason Stockton framed the problem in everyday terms. One of the biggest challenges in running a food business, he said, is the sheer number of small operational issues that need attention every day: an invoice that does not match what was ordered, a supplier that changes its pricing, or information buried across multiple reports. Individually they seem small, he said, but together they consume an enormous amount of time.
MCP Connectivity, Kept Read-Only For Now
Version 5.5 also adds Model Context Protocol connectivity in beta, allowing users to access authorised Supply’d ERP information through supported AI assistants. MCP connections are initially read-only, with future write capabilities planned behind a control plane designed to manage permissions and authorised actions.
That sequencing — read first, write later, behind controls — mirrors the rest of the release. Supply’d is opening its data to outside AI assistants, but only as far as its permission system allows, and only for reading until the control plane is ready. It is a cautious approach to the same agent-access question facing larger platforms, where the prize is an assistant that can act across systems, and the danger is an assistant that acts wrongly.
Beyond the AI features, version 5.5 improves expense reconciliation, adds a sell-by-unit and invoice-by-weight workflow for products such as meat, seafood and cheese, and introduces OAuth authentication for the Supply’d API. Those are unglamorous changes, but for a food ERP they matter: weight-based invoicing and sell-by-unit handling are daily realities for its customers, not edge cases.
Why It Matters
This matters because AI in business software is moving from generic chatbots to assistants that understand a specific company’s data, permissions and workflows. A food manufacturer does not need an AI that can write a poem; it needs one that can find why an invoice is wrong, without touching records it should not touch. Supply’d’s release is a small, practical version of that shift, and its approval gates and read-only defaults show a vendor thinking about trust, not just capability. The same pattern runs through agentic workforce platforms, AI working alongside accountants and super apps opening to outside agents: the tools that win will be the ones that act inside real operations, with controls users can see.
The test will be the Expense Agent’s accuracy in the wild. If it catches real discrepancies and its recommendations hold up under human review, selective automatic approval will follow naturally. If it generates plausible-looking corrections that need constant fixing, customers will keep it on manual approval, and the automation promise will stall.
Background
Supply’d, based in Melbourne, released Supply’d Intelligence version 5.5 on October 28, 2026, with Ask Supply’d and MCP connectivity in beta and the Expense Agent in private beta with selected customers, building on its existing purchasing, production planning, invoice processing and delivery optimisation tools.
Source: Supply’d announcement
0 Comments