AI Chat on a Legacy Produce ERP: Inventory, Sales & Accounts in Plain English

When a produce company has run on the same ERP for years, the data is usually there. The friction is getting to it—inventory in one screen, orders in another, accounts in a third—and the person who needs an answer mid-shift is not always the person who knows which report to run.

This case study covers a mid-market produce company that kept its legacy ERP as the system of record and added a secure analytics-and-chat layer beside it. Staff open a modern web app, scan dashboards for the big picture, and ask ordinary questions in English when they need a specific number. No forklift rewrite. No second system of record.

Operational AI on ERP-domain data—inventory, sales, and accounts—not our support-ticket AI MVP work with helpdesk tools.

The situation

Produce operations move on lots, pallets, and phone calls. Someone always needs a number now: what is on hand, what shipped, what is still open on the books.

On a typical legacy ERP, that meant:

  • Inventory buried in stock screens, location views, or aging reports only a few people could run confidently.
  • Sales and orders spread across order entry, history, and customer lookups—fast if you live in the system, slow if you do not.
  • Accounts—balances, invoices, credits—that meant hopping into AR modules or pinging whoever "owns" that part of the ERP.

The bottleneck was not missing data. It was that a normal question often required the right menu, export, or person. Warehouse, sales, and office staff needed the same facts without the same ERP muscle memory.

Replacing the ERP was off the table—it still ran receiving, order flow, and accounting. They needed a usable front door: dashboards for patterns, plain-English Q&A for lookups—without a multi-year rip-and-replace.

What we built

Interconnected Roots builds practical AI on the systems teams already run. Here we shipped a secure web app beside the existing ERP—analytics plus an in-browser AI assistant grounded in ERP-domain data for inventory, sales, and accounts.

  • Modern web app, not a rip-and-replace. Next.js and TypeScript frontend and API backend. The ERP stays the system of record for posting and transactions.
  • Sign-in that fits mid-market IT. Azure AD / Microsoft Entra ID SSO, plus password login where needed, with JWT-backed sessions.
  • Role-based access. Admin, management, and salesperson roles control what someone can see.
  • Dashboards for the overview. Charts for sales, stock/inventory, and orders—shared picture without report wizards.
  • An in-browser AI assistant. Plain-English questions about inventory, sales, and accounts inside the app—not Teams or Slack. An LLM (OpenAI) interprets the ask and returns answers from data the app is allowed to expose.
  • Validation and a clean UI. Zod validation and a modern React interface for non-power users.

Shipped product, honestly: a pattern of analytics plus AI chat over ERP-domain data. Dashboards and chat sit beside the ERP so people can read and reason over operational facts without full module training.

How it works day-to-day

Day-to-day is straightforward: sign in, check a dashboard for trends, open the assistant for a specific answer, keep moving.

Inventory:

  • "What's on hand for this SKU across locations?"
  • "Any lots close to expiration for this item?"
  • "What came in on today's receiving for this vendor?"
  • "Show me low stock on the items we move most."

Sales and orders:

  • "What did we ship to this customer last week?"
  • "Open orders for this customer still waiting on product?"
  • "Sales by product family month-to-date—just the totals."
  • "Which orders are stuck waiting on inventory?"

Accounts:

  • "What's the open balance for this customer?"
  • "Any past-due invoices over a certain number of days?"
  • "Credits issued this month for this customer?"
  • "Invoice status for this order number."

Trust is the product. If data is unavailable—or the role should not see it—the assistant should say so, not invent a number. Dashboards cover the scan; chat covers the interrupt.

Why this approach

  1. No dual entry.
  2. Faster lookups for non-power users.
  3. Roles that match the business.
  4. Incremental delivery.
  5. Lower change risk.

Practical shop, Oklahoma City home base, teams across the U.S.—useful AI on existing systems, not a buzzword stack.

Who this fits

Mid-market operators with a legacy ERP; ops-heavy warehouse/sales/accounting; leaders who won’t rip out the ERP; AI on real workflows; in-app assistant (not chat-tool bolt-ons). Weaker fit if data is unreliable or write-backs are day-one. Support-ticket AI is a different shape—this is operators asking the ERP questions in English.

Next step

Book a Strategy Call or AI MVP Planning Worksheet → Contact us about AI.