Paste your worst PO. Watch SideQuest parse it.

Drop in a real customer purchase order email or upload a PDF. The same parser SideQuest uses on inbound mail runs right here in your browser. You'll see the customer, ship-to, line items, and confidence scores in real time. Nothing leaves this page.

How the parser thinks

SideQuest reads PO emails the way a person does: scan for the customer name, find the ship-to address, locate the line items, separate the quantities from the unit prices. The parser is deterministic. It looks for anchors first ("Customer:", "Ship to:", "PO #", "Net 30"), then tries several line-item shapes until one matches cleanly: 3 × PN-123 — desc, tabular QTY PN DESC PRICE columns, bulleted lists, free-form mentions of part numbers next to a quantity.

Every line carries four named confidence subscores. Customer-format recognition goes up when standard PO anchors show up — the more "Ship To" / "Bill To" / "Net 30" / "P.O.#" markers we find, the more confident we are about the rest. Table structure measures how consistently the line items match a single shape. Qty / price disambiguation rewards POs that put dollar signs on prices and bare integers on quantities (the disasters are the ones where qty looks like a price). Text legibility is a default 1.0 for typed email text and drops as soon as OCR enters the picture.

A score above 0.65 across the board means the connector treats the PO as clean. SideQuest builds the draft Estimate in QuickBooks and queues it for you to review and submit. The default is human-in-the-loop — you click submit. Two opt-in cuts are available for teams that want more automation: Cut 1 auto-submits clean drafts without a review step (set SIDEQUEST_AUTOSUBMIT=true in your .env); Cut 2 drafts an auto-reply to the buyer confirming receipt. Below the 0.65 threshold, the connector still builds the draft but flags the low-scoring lines and routes the email to a human reviewer with a specific reason. No guessing.

If your sample scored low

The fastest way to get a clean parse on every customer is to publish a templated request format. The field-order template is a good starting point — eight named anchors and a clear line shape. You can also adapt the heuristics to your customer's existing format by adding cross-references in the connector (every time you hand-match a part number, SideQuest learns it for next time, per customer).

If you'd rather not change anything on the customer side, that's fine too. The parser plus the cross-reference learning layer catches most quirky formats inside a week of real-world traffic. The five-PO-formats post walks through the harder cases and what the connector does about each one.