Case Study

Agentic AI in the Real World: UniPro Foodservice's Production Story

UniPro Foodservice is the world's largest foodservice distribution cooperative, whose members generate $168B in annual sales. They deploy Nanonets AI agents in production to manage their complex Sales Order confirmation process, processing $1.2B of orders through autonomous agents.

Louie Newton

As a cooperative, our strength comes from scale and shared efficiency. With AI agents developed by Nanonets, we're transforming how work gets done across UniPro — automating manual workflows, improving data visibility, and enabling faster growth into new offerings and customer segments. This isn't about replacing people; it's about elevating our operating model so our teams can focus on strategic value creation instead of administrative burden.

Louie Newton
CIO, UniPro Foodservice

UniPro processes $1.2B of orders across hundreds of trading partners

UniPro coordinates between 400+ member distributors and a global supplier network. Every order follows the same lifecycle:

Order lifecycle
Member submits PO
400+ distributors
UniPro places order
With supplier
Supplier confirms
PDF, email, spreadsheet
Reconciliation
Match against original order
Sales Order finalized
Sent back to member

Thousands of confirmations arrive monthly, each formatted differently. Every supplier-member relationship carries its own business logic for pricing, substitutions, and shipping. These rules change constantly as contracts are renegotiated. IDP tools broke whenever a supplier changed their document layout. RPA workflows required new code for every rule change. Neither could adapt without engineering effort.

Instead of a fixed workflow, Nanonets deployed an agent that reasons through each order

Traditional automation requires every step to be defined in advance. If a supplier sends a new document format or a policy changes, the workflow breaks and an engineer has to rebuild it. The Nanonets agent takes a fundamentally different approach: it receives a goal and an operating procedure written in plain English, then reasons through each transaction on its own, deciding what to do next based on the data it encounters. Because no sequence is hardcoded, the system adapts to new suppliers, new formats, and new business rules without any code changes.

Nanonets Agent
Decide next step
Perform action
Repeat until goal is met
Goal
Reconcile Sales Order confirmation
Instructions and custom rules to run the process
Context
Nanonets
Nanonets Document AI
Understand documents. Extract data.
Memory
Past learnings and resolved exceptions
Systems
Outlook
Microsoft Outlook
Read and send emails
Dynamics 365
Microsoft Dynamics 365
Read and modify Sales Orders.
Escalations
UniPro Team
UniPro team
Process experts for exception handling

Consider a real scenario: a PDF order confirmation arrives from a seafood supplier in a format the agent has never encountered. It extracts the line items, matches them against the corresponding Sales Order in Dynamics 365, identifies 10 unshipped cases as a backorder, and checks its operating procedure to auto-approve a $0.12/lb price variance within contractual tolerance. No template was configured for this supplier. No human was involved. The entire process took under two minutes.

When the agent encounters a genuinely novel situation, it escalates to a human reviewer with full context: the source document, the matching ERP record, and the exact point of uncertainty. Once resolved, the agent incorporates that decision and handles similar cases autonomously going forward. The more the system runs, the fewer exceptions require human attention. For an organization serving a cooperative with $168B in member sales, this is the difference between automation that creates technical debt and intelligence that compounds over time.

Orders requiring human review
% of total orders escalated to a human, by month post-deployment
93% touchless processing
100% 75% 50% 25% 0% Pre-deploy Month 1 Month 2 Month 3 Month 4 Month 5+

The agent learns from every resolved exception — each human decision reduces future escalations for similar scenarios.

The most significant shift was organizational

UniPro's team moved from manually processing orders to governing the AI: defining business rules in natural language, handling strategic exceptions, and focusing on supplier relationships. The people who once spent their days copying data between systems now spend their time on work that requires human judgment.

The impact was immediate and measurable

The deployment transformed UniPro's order reconciliation from a manual, error-prone process into an autonomous operation.

10,000
Annual hours freed from manual data entry and ERP updates.
>93%
Orders processed end to end without human intervention.
15x faster
average turnaround per order
Key takeaways

See how AI agents can work for your supply chain

Talk to our team about deploying AI agents for your specific reconciliation workflows, ERP, and compliance requirements.

Book a Demo