Mail2Order Configuration Guide
Table of Contents
1. Configure Extraction Instructions
This guide explains how Hyperfox decides which data to look for in an incoming order, and how you can fine-tune that behaviour through Extraction Instructions.
What are Extraction Instructions?
Extraction instructions tell Hyperfox's AI exactly which information to find in a received order document.
When an order arrives (an email, a PDF, an Excel file, or any other supported source), the AI reads through it and looks for the data you care about: the order lines, the delivery date, the customer reference, and so on. Each extraction prompt is a short, plain-language description of one piece of information and where or how it typically appears in your orders.
Think of an extraction prompt as a clear instruction to a new colleague: the better you describe what to look for, the more reliably they'll find it.
AI mapping: linking fields to instructions
Every field in Hyperfox's data model is linked to its own extraction instruction. We call this link the AI mapping.
The data model is the structured set of fields that make up an order in Hyperfox — for example:
- Order line (Item, Quantity, Unit code)
- Notes
- Issue date
- Delivery date
- Customer reference
- Buyer customer party
- Accounting customer party
- Split instruction (see documentation)
For each of these fields, the AI mapping holds an extraction instruction that guides the AI on what to capture. For example, the Order line field might carry the instruction:
An order line containing a product, quantity and optionally the price ordered.
From extraction to a ready-to-validate order
Capturing the raw data is only the first step. Once the AI has extracted a value, it tries to match it against your company's Reference Data (your master data of customers, products etc..).
For example, when the AI extracts an item from an order line, it looks for the matching product in your Reference Data so the order line is linked to the correct article in your catalogue. The same applies to customers, units and other fields.
The result is an order that is not only captured, but also enriched and aligned with your own data — ready for you to review in the To Validate screen.
Editing the AI mapping
You can review and adjust the extraction instructions yourself through the AI mapping interface.
To edit an extraction instruction:
- Navigate to the platform settings via the gear icon in top right corner.
- Select Mappings in the side menu.
- Choose the mapping you want to edit.
- In the list on the left, select the field you want to configure (for example, Order line or Delivery date).
- In the Extraction instruction box, describe — in clear, plain language — what the AI should look for in the order document.
- Save your changes via the Save button in the top right corner.
Changes to the AI mapping apply to orders processed after the change is saved. If an order was processed before, use Retry on that order to reprocess it with the updated instructions. Orders which where already in processing still use the AI mapping prior to your changes.
How to split a single order
Sometimes a single incoming document actually contains more than one order. Hyperfox can automatically split such a document into several separate orders, so each one can be validated and processed individually.
Splitting works just like any other extraction: it is driven by an instruction. The AI mapping includes a dedicated Split instruction field where you describe the rule the AI should follow when deciding whether — and how — to split an order.
To configure splitting:
- Open the AI mapping screen.
- Select the Split instruction field from the list.
- In the Extraction instruction box, describe in plain language when the document should be split into multiple orders (for example, per delivery address, per delivery date, or per purchase order reference). If the Split instruction is left empty, no splitting takes place — the document is always processed as a single order.
Tips for writing good instructions
- Be specific and concrete. Describe what the data represents and how it usually appears in your orders.
- Use the language of your orders. If your customers refer to a "your reference" or "PO number," mention those terms.
- Keep it focused. Each instruction should describe one field, not the whole order.
- Write the instruction as you would explain it to a co-worker on his first without any prior knownlegde about your business.
Examples and the JSON view
Two additional elements are available on the AI mapping screen:
- Examples — For each field you can add sample values that should match it. These help the AI recognise the kind of data you expect for that field.
- Fields / JSON toggle — At the top right you can switch between the Fields view (the guided, field-by-field editor described above) and a JSON view, which shows the underlying mapping as structured data for advanced configuration. Detailed guidance on Examples and the JSON view will follow in a future update of this guide.
Good to know
- Extraction Instructions shape what the AI looks for, while the Codex captures how previous orders were validated. Together they help Hyperfox process your orders with increasing accuracy and consistency over time.
- Changes to the AI mapping apply to orders processed after the change is saved. If an order was processed before, use Retry on that order to reprocess it with the updated instructions.