AI Fields: extract custom data from your documents

What are AI Fields?

AI Fields let you teach SparkReceipt to extract specific data points from your documents that the standard scan doesn't already capture. When SparkReceipt processes a receipt or invoice, it automatically reads the standard details — vendor, date, total, tax, and so on. AI Fields sit on top of that: you define an extra field (for example Supplier VAT ID or Project code), describe what to look for, and SparkReceipt extracts that value from the document text on its own.

This is most useful when a particular detail is required by your local accounting rules or by an internal workflow, but it isn't one of the fields SparkReceipt collects by default. Instead of opening every document and copying the value by hand, you define the field once and it's filled in automatically going forward.

AI Fields is currently a Beta feature and requires an Elite subscription. If your plan doesn't include it, any field values that were already extracted stay visible, but you won't be able to add or change field definitions.

Creating an AI Field

Open AI Fields from your settings to manage your fields. From a document you can also use the Manage link in the document's AI Fields section to jump straight there.

On the AI Fields screen, click Add field. The Create AI Field dialog asks for:

• Field name — a short label for the data point, for example Supplier VAT ID, SKU, or Project Code. Names must be unique and can be up to 100 characters.
• Extraction instruction — a plain-language description of what the AI should look for in the document text, up to 500 characters. This is the most important part: the clearer your instruction, the more reliable the extraction.
• Allow multiple values per document — turn this on for fields that can appear more than once on a single document, such as a list of product codes or serial numbers. When enabled, a document can hold up to 50 values for that field.

You can define up to 20 AI Fields per account. The screen shows how many you've used, and you can enable or disable individual fields, reorder them, and choose whether each one is included in reports.

How extraction works

Extraction runs automatically. After a document finishes its normal processing (the same OCR step that reads the vendor, totals, and dates), SparkReceipt looks at your enabled AI Fields and tries to fill each one from the document's text. While this is happening, the document's AI Fields section shows Extracting AI Fields...; the values appear as soon as it finishes.

A few things worth knowing about how the extraction behaves:

• It only reads what's on the document. AI Fields extract values that actually appear in the document text — they don't invent or calculate data that isn't there. If a field's value can't be found, it's simply left empty for that document rather than guessed.
• It works from the scanned text. Because extraction relies on the text SparkReceipt reads from the document, a field can only be captured if the detail is legible in the original file.
• Re-processing refreshes values. If a document is processed again, its AI Field values are re-extracted and replaced with the new results.

Viewing and editing extracted values

Extracted values show up in the AI Fields section of each document. Values the AI extracted are marked with a check icon so you can tell them apart from ones you entered yourself.

You're always in control of the values:

• Edit a value to correct or refine what the AI found.
• Delete a value you don't want.
• Add value manually — useful when the detail wasn't on the document, or when you've turned on multiple values and want to add another. Each value can be up to 500 characters.

Including AI Fields in reports and exports

Each field has an Include in reports setting that controls whether it appears in your exports by default. When you build a report, you can also pick exactly which AI Fields to include for that export. Selected fields become their own columns in CSV and Excel exports (and appear in PDF exports), right alongside the standard document data — so a compliance column like a supplier VAT ID can travel straight into your accounting workflow.

Example use cases

The right AI Fields depend on your jurisdiction and how your team works. A few grounded examples:

Compliance details required by local VAT rules

In the EU, a valid VAT invoice must show the supplier's VAT identification number and a sequential number that uniquely identifies the invoice — these are mandatory under Article 226 of the EU VAT Directive (2006/112/EC), and a missing field can put the right to deduct input VAT at risk.¹ The UK has equivalent rules: a full VAT invoice must carry the supplier's VAT registration number and a unique, sequential invoice number.²

If your standard scan doesn't already break out the supplier VAT number as its own field, an AI Field is a clean way to capture it. For example:

• Field name: Supplier VAT ID
• Extraction instruction: The seller's or supplier's VAT identification number, usually labelled "VAT No", "VAT ID", or "VAT Reg No".

Organizational workflow fields

Plenty of useful fields have nothing to do with tax law and everything to do with how your organization tracks spend:

• Purchase order / PO number — The purchase order number referenced on the invoice, often labelled "PO" or "Order No".
• Project or cost-center code — The project code or cost center the expense should be billed to.
• Serial numbers (with multiple values enabled) — Every product serial number listed on the document.
• Payment reference or IBAN — The bank account number or payment reference the invoice should be paid to.

A good rule of thumb: if you find yourself manually copying the same kind of detail off documents over and over, that detail is a good candidate for an AI Field.

Tips for reliable extraction

• Be specific in the extraction instruction. Mention the labels that typically appear next to the value on a document (for example, "labelled 'VAT No' or 'Tax ID'"). The more concrete your description, the better the result.
• Use one field per data point. Don't try to capture several unrelated values in a single field — create separate fields so each one stays clean and reportable.
• Turn on multiple values only when needed. Use it for genuinely repeating details like serial numbers or line-item codes; leave it off for single values like a VAT ID.
• Disable fields you're not using. Only enabled fields are extracted, so switching off a field you no longer need keeps your documents tidy.
• Spot-check early results. When you add a new field, review the first few documents it runs on and refine the instruction if the values aren't what you expected.