Automating Purchase Orders in E-Commerce: How Agentic AI Handles Unstructured Input

How Agentic AI simplifies messy, manual order processes

In an era of connected e-commerce, it’s easy to assume that every order flows cleanly through APIs or online checkouts. The reality, however, may be very different, especially in industries where B2B and wholesale operations still rely on unstructured, offline, or legacy input formats.

Across manufacturing, distribution, retail, and consumer goods, order requests may arrive as:

• Handwritten notes faxed or photographed and sent as PDFs
• Filled-out catalog pages with items circled or marked
• Scanned purchase orders or postcards from trade shows
• Free-form emails listing SKUs, quantities, and delivery instructions

For many teams, these inputs still require manual interpretation and re-entry into ERP or e-commerce systems - a process that’s slow, error-prone, and costly. Even businesses running modern e-commerce stacks find themselves blocked by the same bottleneck: data that doesn’t start structured.

How AI can make sense of the mess

This is exactly where AI excels. Instead of hard-coded parsing or template-based OCR, modern Agentic AI systems combine language understanding, document reasoning, and contextual matching to structure even the most inconsistent input.

An AI agent can:

• Read handwritten or scanned content
• Identify product names, SKUs, and quantities
• Match them against product catalog
• Build a draft order or order automatically in e-commerce backend

The result: teams no longer spend hours retyping or verifying orders. They simply review, approve, and move on - freeing up capacity for higher-value work like client support and sales enablement.

The Agentic AI backbone for e-commerce operations

At Upside, we built Enthusiast to give engineering teams a foundation for deploying such AI workflows directly within their stack. It’s an open-source, developer-first toolkit that lets the team build, test, and orchestrate agents that reason, decide, and act — with full control over infrastructure and data.

To make adoption faster, Enthusiast includes a growing library of pre-built agents for common commerce challenges. One of them is the Purchase Order OCR Agent, released in version 1.4.

Purchase Order OCR: from handwritten notes to structured orders

The Purchase Order OCR Agent automates order creation from any unstructured input. The user simply upload or forward a scan, photo, or document — and the agent:

1. Reads and interprets the content
2. Extracts item details, quantities, prices etc.
3. Matches them to product catalog (with integrations to Medusa, Shopware, Shopify, Solidus)
4. Builds a ready-to-review (draft) order

Unlike traditional OCR tools, this agent combines text recognition with catalog awareness, ensuring higher accuracy across different file types. It effectively connects legacy order channels — from paper to PDF — with your modern commerce systems.

Beyond OCR: a foundation for endless workflows

What makes Enthusiast powerful is its open-ended architecture. The same backbone that processes purchase orders can be extended to handle countless other unstructured workflows — from supplier onboarding and catalog enrichment to invoice validation and claims processing.

Engineering teams can design and deploy custom agents using the same modular framework, leveraging shared memory, reasoning, and integrations with tools like Medusa, Shopware, or custom APIs.

In short:
Managing unstructured input is no longer a data-entry problem — it’s an opportunity for intelligent automation. With Enthusiast, AI becomes not just a feature, but an operational backbone that continuously learns, adapts, and streamlines how commerce actually gets done.

Get started

Enthusiast is open-source and built for engineering teams who want full control over their AI workflows:

GitHub: upsidelab/enthusiast

Docs: upsidelab.io/tools/enthusiast