More than 14,000 products updated in one run, and one of Snag's fastest launches, live in about two weeks.
Snag Product Operations Agent
Agentic Achievement Award
ACR Autonomous B2B Order Intake
Purchase order handling fell from about 8 minutes to under a minute, freeing roughly 267 hours a month.
Unstructured purchase orders arrive by email and reach the ERP in under 60 seconds, with the agent deciding when to ask.
What the agent does
Reads an unstructured purchase order, checks it against live pricing and commercial rules, scores its own confidence, and either submits or escalates.
- 2026 Agentic Achievement Award, MACH X: Toronto
01
The problem
Why this exists
ACR is a high-growth B2B distributor of packaging and preparation products serving foodservice, healthcare, hospitality, and industrial customers across North America. Growth by acquisition left it with a fragmented digital landscape: diverse systems, multiple shipping points, and a customer base whose order formats and commercial conditions vary by customer.
Read more
A significant share of purchase orders arrived as unstructured PDF attachments by email. Each one was re-keyed by hand into an IBM AS/400 ERP, averaging around 8 minutes. Across roughly 2,000 monthly PDF and email orders that came to about 267 hours of manual effort a month, consumed almost entirely by customer service staff.
Speed was not the whole problem. Manual entry introduced data entry errors that produced downstream corrections, ERP re-submissions, and delivery adjustments. Processing times varied, so order confirmation was inconsistent. IT absorbed the overhead in ERP error triage. And because acquisition multiplied order volume, the constraint got worse with every deal.
02
Outcomes
What changes when this is in place
- Processing time fell from approximately 8 minutes to under 60 seconds per order, roughly 87% faster.
- For structured purchase orders, straight-through processing went from 0%, fully manual, to approximately 99%.
- Around 267 hours of manual effort are freed per month and redirected to customer-facing work. That is an estimated $95,000 to $110,000 in annual redirected labor capacity, roughly 1.7 full-time equivalents, with no headcount reduction.
Read more
- Reviewer effort on low-confidence exceptions fell 60 to 70%, because the agent annotates why it is unsure rather than only flagging that it is.
- ACR can absorb volume growth from expansion and acquisition integration without adding order processing headcount, and can onboard a newly acquired entity's order flow without building an intake process from scratch.
03
How it works
Architecture
The system is four layers. An intake and triage layer receives email-based purchase orders and routes them into the pipeline. An LLM-powered document extraction layer parses PDF orders, normalizes fields, and detects anomalies. An orchestration and business rule validation layer checks pricing, customer-specific conditions, catalog data, and eligibility, and routes exceptions. An autonomous ERP submission layer delivers validated order data into the IBM AS/400 backend for fulfillment. Structured EDI flows continue in parallel.
Read more
Three properties make it agentic rather than automated. First, it interprets ambiguous input: hundreds of customers use different templates, field labels, line item structures, and pricing conventions, and the agent understands order intent across that diversity rather than matching templates. Second, it decides its own autonomy: the system evaluates its extraction confidence before choosing whether to submit or route for review, and a low-confidence order carries contextual annotations explaining why. Third, it exercises cross-system judgment, validating each order against live rules and making a go or no-go call before ERP submission.
The MACH Alliance overview places Zendesk, Emporix, and the AS/400 in the stack. The detailed case study describes the extraction layer only as LLM-powered and does not name the model version or the integration middleware.
CIO Thai Vong describes the milestone as trust rather than technology: the moment they let the system submit a purchase order to the ERP with nobody reviewing it first. Oversight was redesigned rather than removed.
Open standards used
REST APIsEDI
Tech stack
Emporix Autonomous Commerce ExecutionLLM document extractionIBM AS/400 ERPZendesk
Resources
Next step
Request an intro
Emporix is behind this solution. Ask about automating your order intake, and the MACH Alliance will put you in touch with the person who can answer.
Not ready to talk to anyone? MACH Insider is the Alliance mailing list. New solutions, program openings, and the research behind them, once a month.
Everyone credited on this solution
- EmporixAutonomous Commerce Execution platform
Every organization named here has a page listing the rest of its work on the Hub.
Related solutions
Content operations for more than 16,000 users across multiple agencies, with 90% of metadata creation automated.
General Motors AssetIQ Content Agents
Agentic Achievement Award
Sales teams answer quote requests in minutes instead of re-keying them, without a volume ceiling set by headcount.
commercetools B2B Intake
Verification
Published case study
The detail on this page is drawn from a case study published by the contributor or the MACH Alliance. Any figures on this page are the ones that case study states.
What this page does not establish
- The model version behind the extraction layer. The case study describes it only as LLM-powered.
- A straight-through processing rate for unstructured PDF orders. The case study's 99% figure covers structured purchase orders only.
Where this came from
Published from the winner's case study and the 2026 MACH Impact Awards announcement.
Status: Live in productionAdded Apr 29, 2026A listing is an editorial decision, not a MACH Alliance certification.
Built something worth showing?
Put your agentic solution in front of the enterprise leaders who use the hub to decide what to build next. Submissions are open to MACH Alliance Members, Collaborators, or Ambassadors.