Power Automate Invoice Processing: How to Set It Up

Power Automate can read an invoice and pull out the vendor, invoice number, date, and total without anyone retyping it, using the AI Builder invoice processing model inside a cloud flow. It is a good fit for teams already on Microsoft 365 that want hands-off automation. This guide walks through how to set it up step by step, what the model extracts, how accurate it is, what it actually costs, and where it falls short. If you just need clean Excel or CSV rows today without building a flow or buying credits, drop your PDF or image invoices into the converter at the top of this page.

How do you process invoices in Power Automate?

You process invoices in Power Automate by building a cloud flow that triggers when an invoice arrives, runs the AI Builder "Extract information from invoices" action on the file, and writes the extracted fields wherever you need them. The flow does the reading and typing so the AP team does not have to.

A typical flow looks like this. Create an automated cloud flow with a trigger such as "When a new email arrives" with an attachment, or "When a file is created" in a SharePoint or OneDrive folder where invoices land. Add a new step, choose AI Builder, and select Extract information from invoices. Point the action's Invoice file input at the attachment or file content from the trigger. In the steps that follow, use the model's output fields (invoice ID, vendor, total, and the rest) to add a row to an Excel table, create a record in Dataverse or SharePoint, or start an approval. Save, run a test with a real invoice, then turn the flow on.

What is the invoice processing model in Power Automate?

The invoice processing model is a prebuilt AI Builder model that recognizes common invoice fields out of the box, with no training required. Microsoft optimized it for standard invoice layouts, so you can drop it into a flow and start extracting data on day one rather than labeling sample documents first.

Because it is prebuilt, it handles the everyday case well: a typed invoice from a vendor with a normal header, a line-item table, and a totals block. If your invoices are unusual or you need fields the prebuilt model does not capture, AI Builder lets you build a custom invoice model and train it on your own samples, which takes more setup but can lift accuracy on tricky layouts.

How does AI Builder extract invoice data from an invoice?

AI Builder reads the invoice file you pass it, locates the fields it was trained to recognize, and returns each one as a value your flow can use, along with a confidence score. You feed it the file, it gives back structured data such as invoice number, date, and total that the rest of the flow can act on.

For long documents you can narrow the work by entering a page or page range in the Pages parameter so the model only scans the pages that matter. The confidence score on each field is the useful part for automation: you can branch the flow so high-confidence invoices post automatically while anything below a threshold routes to a person for a quick check.

Can Power Automate read PDF invoices?

Yes. Power Automate, through the AI Builder invoice model, reads PDF invoices directly, including both digital PDFs and scanned images saved as PDF. It also handles common image formats, so an invoice that arrived as an email attachment or a phone photo can be processed by the same flow.

Quality still matters. A clean, typed PDF reads far more reliably than a crooked phone photo or a faint scan. For the best results, capture scans at around 300 DPI and keep the page flat and well lit. Poor images do not fail outright, but they push more fields below the confidence line and into manual review.

What does the Power Automate invoice model extract?

The prebuilt invoice model extracts the standard header and totals fields most AP teams need: vendor name and address, customer details, invoice ID, invoice date, due date, purchase order number, subtotal, tax, and total amount due. These come back without any training on your part.

It also returns line-item detail (description, quantity, unit price, and amount per row) for invoices where the table is clearly laid out. Line items are the hardest part of any invoice to read accurately, so review those closely on complex or multi-page invoices. For a deeper look at why the line-item table trips up automation, see how invoice line item extraction works.

How accurate is Power Automate invoice processing?

Power Automate invoice processing is accurate enough for production on clean documents, typically returning confidence above 0.95 for key fields like invoice number, date, and total on typed PDFs. Scanned documents at 300 DPI usually score 0.90 or higher, while phone photos and handwritten invoices drop to roughly 0.70 to 0.85.

In published comparisons the prebuilt model landed around 91% accuracy versus about 94% for a custom-trained model, so the prebuilt option is reliable but a custom model can do better on your specific layouts. The practical takeaway is to use the confidence scores: auto-post the high-confidence invoices and send the rest to a review queue. To see how OCR and AI methods differ on accuracy, read about invoice OCR software.

How much does Power Automate invoice processing cost?

Invoice processing in Power Automate runs on AI Builder credits, and the most common path is a Power Automate Premium license at about $15 per user per month, which includes a pool of AI Builder credits. Document processing consumes 10 credits per page, so a Premium seat's roughly 5,000 credits cover a few hundred pages a month before you need more.

If your volume is higher, you buy AI Builder capacity add-ons on top of the license. Microsoft has also said the seeded credits bundled with Premium are being removed on November 1, 2026, after which organizations will need to purchase AI Builder capacity separately. Budget for the license plus per-page credit consumption, and remember a maker has to build and maintain each flow.

What are the limits of Power Automate for invoice processing?

The main limits are setup effort, licensing, and credit cost. You need someone comfortable building and maintaining flows, a Power Automate Premium or AI Builder licensed environment, and enough credits to cover your page volume. None of that is hard for a Microsoft 365 shop, but it is real overhead for an occasional or one-time job.

It is also less suited to a quick backlog. If you have a folder of a few hundred invoices to clear this week, standing up a flow, sorting licensing, and managing credits is slower than just uploading the files to a tool and downloading a spreadsheet. Power Automate shines for ongoing, hands-off automation, not for ad-hoc cleanups. If you also need to read non-invoice paperwork in the same pipeline, a general document OCR tool can cover contracts, forms, and receipts alongside it.

Is there a faster way to extract invoice data than Power Automate?

Yes. If you do not already run Power Automate or you just need invoice data now, uploading invoices to a browser tool is faster: there is no flow to build, no license to provision, and no credits to manage. You drop in PDF or image invoices and download clean Excel or CSV rows in seconds.

That is exactly the gap our tool fills. It uses AI to read the same fields the AI Builder model does (vendor, invoice number, dates, line items, and totals) and hands you a spreadsheet without any setup. For ongoing Microsoft-centric automation, Power Automate is a fine choice, and you can read more on the trade-offs on our Power Automate invoice processing page. For backlogs, ad-hoc batches, or teams without a maker, extract invoice data to Excel by uploading your files, or use AI invoice data extraction to handle varied vendor layouts with no templates.