Build a Custom AI Analyzer for Your Own Invoice Format in Power Automate: A 3-Step Dropbox Workflow

A Custom AI Analyzer Template is a named JSON schema you define in the PDF4me dev portal that tells the AI which fields to extract from a specific document type. This walkthrough builds an Analyzer for PDF4me's own SaaS subscription invoices (33 fields: vendor address block, Swiss MWST rate, subscription Id, line items, discount details), then calls it from a 3-step Power Automate flow that reads a PDF from Dropbox and returns structured JSON.

The built-in default_invoice_extraction covers generic invoices, but real AP teams have vendor-specific layouts, custom line-item columns, jurisdiction-specific tax fields, and subscription metadata that the generic schema misses. Building a custom Analyzer once gets you exact-match extraction forever, no per-field prompting required, no follow-up logic to clean up the response.

Why-Based Q&A​

What You'll Get​

Input: A PDF invoice in the format your Analyzer was designed for, stored in Dropbox. This walkthrough uses a PDF4me SaaS subscription invoice (Ynoox GmbH / Switzerland, MWST 8.1%, subscription metadata, Pdf4me Base 1000 plan). Output: A structured JSON response with all 33 fields from the schema, available as Power Automate dynamic content for the next action (write to SharePoint List, post to Slack, append to Excel Online, route to Approval, etc.).

Sample input invoice​

Sample output JSON (excerpt from the real run)​

{
  "documentTitle": "INVOICE",
  "invoiceNumber": "Pdf4me-202601-32324",
  "invoiceDate": "Jan 26, 2026",
  "currency": "USD",
  "paymentStatus": "PAID",
  "vendorName": "Ynoox GmbH",
  "vendorAddress": "Riedstrasse 2B, Hedingen, Zürich 8908, Switzerland",
  "vendorTaxRegNumber": "CHE-360.751.198",
  "billTo": "Chandra Arora",
  "subscriptionId": "2e46662a-32e0-4e79-ba06-b13fe2c113ac",
  "billingPeriodString": "Jan 26 to Feb 26, 2026",
  "nextBillingDate": "Feb 26, 2026",
  "mwstRate": 8.1,
  "discountCode": "F_yn_100!",
  "discountAmount": 35,
  "lineItems": [
    { "description": "Pdf4me Base 1000", "price": 35, "discount": 35, "amount": 0 }
  ]
}

The complete 33-field response is downloadable in the section below.

What You Need​

The schema: 33 fields the Analyzer extracts​

The Analyzer schema is the heart of the custom extraction. Each field declares the name Power Automate will see, the type Power Automate will receive, and a description that guides the AI on where to look for the value.

The full JSON schema (with the exact fieldDescription strings the AI consumes) is in the downloadable file above.

Setup A: How do you log in to the PDF4me dev portal?​

Where you are: before any flow exists.

1. Open dev.pdf4me.com in any browser.
2. Either type your email and password and click Login, or pick one of the SSO options on the right: Continue with Microsoft, Google, Facebook, Apple, GitHub.
3. New here? Click No Account? Register to create one.

The login screen​

Setup B: Open AI Document Parser in the dashboard​

Where you are: logged in, on dev.pdf4me.com/dashboard.

1. The dashboard left sidebar shows your account sections: Home, Profile, Subscription, Payment Info, Api Keys, Logs, Invoices, AI Document Parser, My Documents, Classify Document, Parse Document, Job History, Shared Documents, Refer a friend, Rewards.
2. Click AI Document Parser.

The dashboard sidebar​

Setup C: How do you add a new Analyzer Template?​

Where you are: AI Document Parser page.

1. The page lists every Analyzer Template you have already created with its Id, Type, and edit / delete actions.
2. Click + Add.
3. A blank row appears with two fields:
   • Analyzer Id: type a short, memorable name (e.g. Test Invoice). This is the exact string you will pass as Customisation Note in Power Automate.
   • Analyzer Type: dropdown with options Parse and Classify. Pick Parse for structured field extraction. Pick Classify when you want a routing decision instead.
4. Click Save.

Add Analyzer​

Setup D: Paste the JSON schema and save​

Where you are: editing the newly created Analyzer.

1. Click the pencil edit icon next to your new Analyzer (Test Invoice).
2. The Parse Info editor opens with the Analyzer Id at the top and the Document Schema field below.
3. Paste the 33-field JSON schema from analyzer-schema.json into the Document Schema text area. Each entry is { "fieldName": "...", "fieldType": "...", "fieldDescription": "..." }. The fieldDescription is the most important field, it tells the AI where to look and what to capture.
4. Pick a Sample Document Type preset (Invoice or Purchase Order) if you want a starting template; otherwise paste your schema directly.
5. Click Save Changes.

Parse Info schema editor​

Flow 1: How do you wire the Analyzer into Power Automate?​

Flow so far: trigger only.

The Power Automate side is short: 3 actions. Start with a manual trigger so you can iterate; swap for an HTTP / SharePoint / Outlook trigger when ready for production.

1. In Power Automate click Create, choose Instant cloud flow.
2. Name the flow (e.g. Custom Invoice Extraction), pick Manually trigger a flow, click Create.

The complete 3-action flow​

Three actions. AI - Document Parser is the only step that calls PDF4me, and it takes most of the runtime (around 21 seconds for a one-page invoice).

Flow 2: Get the source PDF from Dropbox​

Flow so far: trigger plus Get file content.

1. Click + New step, search Dropbox, pick Get file content using path.
2. Connect your Dropbox account if you have not already.
3. Configure:
   • File Path: /pdf4me.pdf
   • Infer Content Type: Yes (under Advanced parameters)

Get file content configuration​

Flow 3: Why does the Customisation Note field carry your Analyzer name?​

Flow so far: trigger plus Get file content plus AI - Document Parser.

This is the action that calls your custom Analyzer. The trick is the Customisation Note field: paste the exact Analyzer Id name from Setup C (Test Invoice in this walkthrough). Without it the action falls back to the default invoice extraction.

1. Click + New step, search PDF4me, pick AI - Document Parser.
2. Connect PDF4me AI with your API key (first time only).
3. Configure:
   • File Content: pick the File Content token from step 2 (Get file content)
   • File Name: Test.pdf (label only, not a path)
   • Customisation Note (under Advanced parameters): Test Invoice (the exact Analyzer Id you set in Setup C)
4. Save and Test.

AI - Document Parser action​

Run the flow and verify​

1. Save the flow at the top right.
2. Click Test, choose Manually, click Test then Run flow.
3. Open the run details. All three actions should be green. The AI - Document Parser step shows a JSON body in its outputs.
4. The body of the response carries every field your schema declared, populated from the PDF.

Sample response from a real run​

Common Variations You Can Add Without Rebuilding​

Custom Analyzer vs default_invoice_extraction​

Common questions​

How long does it take to design the schema?​

For a familiar invoice format (your own SaaS billing, your top three vendors), 15-30 minutes for the first draft. Read through one invoice end-to-end, list every field your AP team manually transcribes, write a one-sentence fieldDescription for each that names the label as printed and gives an example. Then test against 3-5 sample invoices and iterate on descriptions that come back wrong.

What happens if a field is missing from the document?​

The Analyzer returns the field with a null or empty value. It does not error. Your downstream Power Automate logic should handle missing values with coalesce() or if(empty(...)) expressions, especially for required fields like invoiceNumber. This is the same behaviour as default_invoice_extraction.

Can I version my Analyzer schema?​

Not natively in the dev portal yet. The recommended pattern is to create a new Analyzer Id when you make a breaking schema change (Test Invoice v2) and update Power Automate's Customisation Note to point at the new version. Keep the old Analyzer active until every flow has migrated.

How does the AI know which fields to extract for fieldMethod: generate?​

The AI reads the fieldDescription and the names of sibling fields in the same level. For vendorAddress defined as "Complete vendor address (street, city, state, postal code, country). Use comma(,) to separate lines.", the AI assembles the parts from the already-extracted vendorStreet, vendorCity, vendorState, vendorPostalCode, vendorCountry fields. No regex, no string templates, no follow-up action required.

Does the Analyzer work across Power Automate, Make, Zapier, and n8n?​

Yes. Once an Analyzer Id is published in the dev portal, every PDF4me AI - Document Parser action across every platform can reference it by name. Build the schema once, use it from any automation tool, the same JSON response shape comes back.

Troubleshooting​

Next Steps​

The same setup-once / call-many pattern (define Analyzer, paste schema, pass the Id as Customisation Note) works for any document type: purchase orders, contracts, bank statements, medical claims, customs declarations. Define one Analyzer per format, then mix and match from any Power Automate, Make, Zapier, or n8n flow.