Challenge

ACR (AmerCareRoyal) is a high-growth B2B distributor of essential packaging and preparation products serving foodservice, healthcare, hospitality, and industrial sectors across North America. Following multiple acquisitions, the company operates across a fragmented digital landscape with diverse systems, multiple shipping points, and a heterogeneous customer base with varying order formats and commercial conditions.

A significant share of customer purchase orders arrived as unstructured PDF attachments via email. Before automation, each of these required manual re-keying into ACR's IBM AS/400 ERP environment, a process that averaged approximately 8 minutes per order. Across roughly 2,000 monthly PDF and email POs, this translated to approximately 267 hours of manual effort per month, consumed almost entirely by customer service staff.

The problem was not simply one of speed. Manual entry introduced data entry errors requiring downstream corrections, ERP re-submissions, and delivery adjustments. It created variable processing times leading to inconsistent order confirmation and customer experience. It generated operational overhead in IT through ERP error code triage and support escalations. And it imposed scalability constraints, as growth through acquisition multiplied order volume without a proportionate increase in processing capacity.

Rules-based automation was insufficient because ACR's incoming purchase orders are inherently unstructured and inconsistent. Different customers use different PDF templates, field labels, and data formats. A rules engine cannot reliably interpret intent across this variety. The system needed to understand order intent across diverse, unformatted documents, validate pricing and customer-specific commercial conditions in real time, detect anomalies and route exceptions intelligently, and make autonomous submission decisions without requiring manual sign-off on every transaction.

This demanded genuine AI-driven decision intelligence, not process choreography. The autonomous approach also aligned directly with ACR's strategic imperative: to scale order operations without scaling headcount, and to build a composable, acquisition-ready commerce foundation that can absorb new entities without replatforming.

"We saw an opportunity to remove manual friction from the process so our teams could focus on work that truly drives value. Automation was not just about speed, it was about elevating how we operate." Thai Vong, CIO, ACR

Strategy

In February 2026, ACR launched its Autonomous Order Intake capability across its email-based purchase order inboxes in North America, processing approximately 2,000 PDF and email purchase orders per month.

The solution operates across multiple systems with minimal human intervention. An intake and triage layer handles incoming email-based purchase orders and routes them into the automated processing pipeline. An intelligent document extraction layer, powered by an LLM, parses PDF purchase orders, normalises fields, and detects anomalies. An orchestration and business rule validation layer checks pricing, customer-specific conditions, and routes exceptions. And an autonomous ERP submission layer delivers clean, validated order data into ACR's IBM AS/400 backend for fulfilment execution.

What makes this agentic, not just automated, is the system's intelligence in three critical areas. First, interpretation of ambiguous inputs: the document agent interprets unstructured, inconsistently formatted PDFs from hundreds of different customers, each using different templates, field labels, line item structures, and pricing conventions. This is not template matching; the agent understands order intent across document diversity. Second, confidence-based decision autonomy: the system evaluates its own extraction confidence before deciding whether to submit autonomously or route for human review. High-confidence orders proceed straight through, while low-confidence orders are flagged with contextual annotations explaining why, reducing reviewer effort by 60-70%. Third, cross-system orchestrated judgment: each order is validated against live business rules, customer-specific pricing, product catalogue, and commercial conditions, with autonomous go/no-go decisions made before ERP submission.

The initiative is part of ACR's enterprise AI Framework and Center of Excellence, established under the direction of CIO Thai Vong, and represents the first production deployment within that programme. It was presented to ACR's Board of Directors and communicated company-wide through internal town halls, confirming its status as a board-level strategic priority.

"The real milestone was not the technology going live, it was the moment we trusted the system to submit a purchase order to our ERP without anyone reviewing it first. That is a fundamental shift in how you think about operational control. We did not lose oversight, we redesigned it." Thai Vong, CIO, ACR

Impact

Even before full rollout, the Autonomous Order Intake capability is delivering measurable impact across operational efficiency, customer experience, and business performance.

PO processing time has been reduced from approximately 8 minutes to under 60 seconds per order, roughly 87% faster. The straight-through processing rate has gone from 0% (fully manual) to approximately 99% for structured POs. Around 267 hours of manual effort are freed per month, redirected to higher-value customer-facing work. Approximately 2,000 PDF and email POs per month are handled fully automatically with no headcount increase required. Error-driven corrections have been significantly reduced, with fewer delivery corrections and ERP escalations.

Based on the cost of a customer service representative, the approximately 267 hours of monthly capacity freed by automation represents an estimated $95,000 to $110,000 in annual redirected labor capacity, equivalent to roughly 1.7 FTEs. This capacity has been redirected to higher-value, customer-facing work rather than eliminated, compounding the business value beyond direct cost savings.

Customer experience has improved from inconsistent, variable confirmation times to accurate, fast, and transparent order processing. Faster processing and higher first-pass accuracy mean orders are confirmed sooner and delivery is less prone to errors caused by re-keying mistakes.

The capability also enables ACR to absorb volume growth from organic expansion and acquisition integration without increasing order processing headcount, a direct contribution to unit economics and scalability. As ACR continues to integrate newly acquired entities, Autonomous Order Intake provides a ready infrastructure for onboarding new order flows without custom-building intake processes from scratch.

The roadmap extends this foundation in three phases. The current phase focuses on full rollout of Autonomous Order Intake across all email-based PO inboxes. Phase two activates ACR's customer self-service portal at scale, enabling customers to place, track, and manage orders digitally. Phase three expands into additional value streams including returns management automation, cart and checkout, and account self-service.

Composable architecture in action

ACR's composable architecture was already partially in place prior to this initiative. What the Autonomous Order Intake implementation demonstrated was how that composable foundation could be extended upward: from process orchestration into genuine agentic decision intelligence.

An iPaaS layer enables seamless integration with existing systems without requiring replatforming. ACR connected its ERP backend and intake layer to the orchestration engine via APIs, not custom point-to-point integrations. An orchestration engine provides the structured workflow execution layer, with business rule validation, pricing checks, customer condition enforcement, and exception routing, giving the AI agent a governed, auditable decision framework to operate within. An API-first, headless architecture ensures all components communicate via APIs with clear separation of concerns. The document intelligence layer, the intake layer, the orchestration layer, and the ERP each play a distinct role, replaceable and extensible without disrupting the others.

The most important architectural insight from this implementation is the separation between orchestration and intelligence. The orchestration layer handles workflow structure, rule enforcement, and ERP connectivity. The document agent handles interpretation of unstructured inputs. Neither replaces the other; together they form the execution layer that makes autonomous commerce possible.

This architecture also directly supports ACR's acquisition integration strategy: as new entities are onboarded, their order flows can be connected to the same orchestration layer without rebuilding intake infrastructure.