Priced Per Page, Checked

Power Automate PDF to Excel: Convert PDF to Excel in a Flow, and What Each Route Costs Per Page

There is no free PDF to Excel action. Every route out of a flow and into a spreadsheet is metered, and the meter is almost always the page. AI Builder custom document processing charges 100 credits for every page it reads. The generic HTTP action is a premium connector before it sends a single byte. This page prices each route from Microsoft own documentation, then converts your file so you can see the output before you build anything.

Figures checked against Microsoft Learn and the Power Automate pricing page in September 2026. Nothing here is affiliated with or endorsed by Microsoft.

Drop your PDF here or click to browse

PDF files up to 50MB

Convert a sample document here before you spend a credit in a flow.

The short answer

Power Automate cannot convert a PDF to Excel out of the box. The three working routes are an AI Builder document processing model, a third-party PDF connector, or the generic HTTP action pointed at a conversion API, and all three are premium connectors that require Power Automate Premium at $15.00 per user per month paid yearly. On top of the license, AI Builder custom document processing bills 100 credits per page, which is $0.05 a page on the old capacity add-on and $0.08 a page for a new customer who has to buy Copilot Credits instead. Because that meter counts pages rather than documents, a single 40 page report costs the same as forty one-page invoices. A conversion API on a flat monthly plan bills per file, so the comparison turns entirely on how many pages your average document has.

Last updated September 2026

100 credits

What AI Builder custom document processing charges per page

$15.00

Power Automate Premium, per user per month, paid yearly

Per file

How this converter counts, so page count stops mattering

OCR included

Scans are read on every plan, not metered as a separate model

There is no free PDF to Excel action, and the meter is the page

People arrive at this problem expecting a step in the action picker, the way there is a step for sending mail or writing a row to a spreadsheet. There is not one. Power Automate can move a PDF around all day, and it can hand the bytes to something else, but reading the tables inside is a separate, licensed capability every time.

That matters more than it sounds, because the pricing model that comes with each route is different from the one your workload has. Almost every document workflow is described in documents: 500 invoices a month, 40 statements a quarter. Almost every PDF meter counts pages. The gap between those two units is where flow budgets go wrong.

Route one

AI Builder

Train a document processing model on sample layouts, then call it from the flow and get named fields back. It is the most Microsoft-native answer and the most expensive per page. It also returns fields you defined rather than whole tables, which is a different shape of output than a spreadsheet of every row.

Route two

A third-party connector

Encodian, Adobe PDF Services, PDF4me and ConvertAPI all publish Power Automate connectors with a PDF to Excel action. You get a tidy action in the designer, and you pay the vendor separately from Microsoft. The connector itself is premium, so the Power Automate license is still required underneath it.

Route three

The HTTP action

Point the generic HTTP action at any conversion API and you are not limited to whoever built a connector. It is more work to wire up the first time and it is the only route where you choose the billing model rather than inheriting it. Our PDF to Excel API is designed for exactly this call.

What each PDF route inside Power Automate actually costs

AI Builder rates are from the Microsoft Learn capability rate table, checked September 2026. Dollar figures are Microsoft own conversions: the AI Builder column assumes the Tier 1 add-on at one million credits for $500 a month, and the Copilot Credit column assumes pay-as-you-go at one cent per credit.

Capability	Unit	AI Builder credits	Cost on the add-on	Cost on Copilot Credits
Custom document processing	1 page	100	$0.05	$0.08
Receipt, invoice or identity document analysis	1 page	32	$0.016	$0.08
Contract processing	1 image	32	$0.016	$0.08
Text recognition, plain OCR	1 page	3	$0.0015	$0.001
PDFxlsx over the HTTP action	1 file, any page count	Not applicable	Flat monthly plan from $9, unlimited pages per file

Add the Power Automate license underneath all of it. Premium is listed at $15.00 per user per month paid yearly, Process at $150.00 per bot per month, and Hosted Process at $215.00 per bot per month.

Three things in that table that change what you should build

All three are checkable in Microsoft own documentation in about a minute, and none of them appear in the tutorials that walk you through building the flow.

Text recognition is 33 times cheaper than custom document processing

Three credits a page against one hundred. If what you need is the characters off a scanned page, and the structure work is happening somewhere else, reaching for the custom document processing model because it sounds like the document one multiplies the bill by thirty three for the same underlying job. The rule of thumb: custom document processing earns its price when you want specific named fields out of a layout you trained it on. It does not earn its price when you want the whole table.

New customers cannot buy the cheaper meter at all

Microsoft documentation states plainly that AI Builder capacity add-ons can only be purchased as a renewal or true-up by existing customers, and that new customers must purchase Copilot Credits. That moves custom document processing from $0.05 a page to $0.08. It gets tighter still: from November 1, 2026, no new or renewed Power Platform or Dynamics 365 license will carry seeded AI Builder credits, and existing customers keep theirs only to the end of the current contract term. AI Builder trials have been discontinued. If you are costing this today for a renewal next year, cost it at the Copilot Credit rate.

Credits do not roll over, and running out stops the flow

Consumption resets on the first of the month and unused capacity is gone. Exceed it and the runs are blocked outright until someone buys or reassigns more. For a workload that arrives evenly that is a budgeting annoyance. For anything seasonal, a quarter-end close, an annual audit pull, a year-end statement run, it means the month you most need the flow is the month it stops. You have to size capacity for your peak and pay for that ceiling in the eleven quiet months too.

The same workload, priced two ways

One accounts payable team, one flow, one month. The only thing that changes between the two columns is what a unit means. This is the arithmetic worth doing before you pick a route, and it takes two numbers you already know: how many documents a month, and how many pages each one runs to.

Monthly workload	Pages processed	AI Builder at $0.08 a page	Flat per-file plan
300 one-page invoices	300	$24.00	Covered by a mid tier
300 invoices averaging 3 pages	900	$72.00	Unchanged, still 300 files
40 statements averaging 30 pages	1,200	$96.00	40 files, the smallest tier
One 400 page annual report	400	$32.00	One file

Add $15.00 per user per month for the Premium license to every figure in the AI Builder column. The pattern is consistent and it is the whole point: page pricing punishes long documents and rewards short ones, so a team whose documents are single-page invoices does reasonably well on it, and a team that handles statements, reports or contracts does badly. Our own pricing page lists what a file costs across sixteen tools if you want the wider comparison, and converting large PDFs to Excel covers the long-document case in detail.

How to convert a PDF to Excel in a flow with the HTTP action

Four steps, and the shape is the same whichever conversion API you point it at. The full endpoint reference lives on the PDF to Excel API page.

Trigger and fetch

Start the flow on the file arriving: a new item in a SharePoint library, an attachment in a shared mailbox, a file dropped in OneDrive. Then use Get file content so the actual bytes are in the run, not just a link to them.

POST to the API

An HTTP action, method POST, a bearer token in the Authorization header, and the file content as multipart form data. The response comes back with an identifier and a status. Store the identifier in a variable, because the next two steps need it.

Wait for done

Conversion is asynchronous, so put a Do until loop around a GET on the status endpoint with a short delay inside. Give the loop a sensible count limit rather than leaving it unbounded, and branch on a failed status so a bad file raises something a human sees.

Write the result

Download the XLSX and save it with Create file, or take the rows as JSON and loop them into Excel Online with Add a row into a table. The JSON path is the one you want if the data is going into a system rather than into a workbook someone reads.

One detail that saves a support ticket

Check the row count in the response before you write anything downstream. A conversion that returns a file but almost no rows usually means the source was a scan that arrived without an OCR pass, or a layout the extractor read as free text rather than a table. Failing the flow loudly at that point is far cheaper than discovering a half empty ledger import three weeks later. There is a fuller list of checks in how to check a PDF to Excel conversion for errors.

Which route fits which job

Stay with AI Builder when

You want a handful of named fields, not whole tables: the vendor, the invoice number, the total.
Documents are short and the layouts are stable enough to train a model against.
The output has to land in Dataverse and stay inside the Microsoft estate for governance reasons.
Volume is low enough that per-page pricing never becomes the line item anyone questions.

Call a conversion API when

You want every row of every table, not fields you defined in advance.
Documents run long, so per-page billing is the thing making the flow expensive. See batch PDF to Excel conversion.
Layouts vary between senders and training a model per layout is not realistic.
Scans are common and you want OCR included rather than metered as a second model. That is what OCR PDF to Excel handles.

In practice a lot of teams end up running both, and that is a reasonable answer rather than a failure to decide. AI Builder reads the header block of a purchase order into Dataverse fields, and a conversion API turns the line-item grid underneath it into rows. The finance side of this is covered on PDF to Excel for finance teams, the document-specific behavior on converting invoice PDFs to Excel and purchase orders to Excel, and if the destination is a plain feed rather than a workbook, PDF to CSV is usually the cleaner target.

Power Automate and PDF to Excel: common questions

Not with a built-in standard action. Power Automate has no native PDF to Excel step, so every working flow reaches for one of three things: an AI Builder document processing model, a third-party connector such as Encodian, Adobe PDF Services or PDF4me, or the generic HTTP action pointed at a conversion API. All three are premium, so the flow needs a premium license before any of them will run.

No. The actions that read a PDF are premium connectors, and Microsoft documentation is explicit that a flow using any AI Builder action is a premium flow requiring premium licensing. The Microsoft 365 seeded use rights cover standard connectors and Microsoft 365 data only, which is why a flow that works in the designer stops running once someone checks the license.

Yes, for every practical PDF to Excel route. Power Automate Premium is listed at $15.00 per user per month paid yearly, and Power Automate Process at $150.00 per bot per month paid yearly. The generic HTTP action and all AI Builder actions sit behind that license. The service-specific actions such as Send an HTTP request to SharePoint are standard, but they only call Microsoft 365 endpoints.

Microsoft rate table puts custom document processing at 100 AI Builder credits per page, which is $0.05 a page against the Tier 1 add-on of one million credits for $500 a month. Prebuilt receipt, invoice and identity document models cost 32 credits a page, or $0.016. Plain text recognition is 3 credits a page. Those are per page, not per document.

Custom document processing consumes 100 credits for every page it reads. A 40 page report is 4,000 credits in a single run. Prebuilt invoice and receipt analysis is far cheaper at 32 credits a page, and pure text recognition is 3. The 33x gap between text recognition and custom document processing is the single most expensive thing people get wrong when they build these flows.

No. Microsoft documentation states that AI Builder capacity add-ons can only be purchased as a renewal or true-up by existing customers, and that new customers must purchase Copilot Credits instead. At the pay-as-you-go rate of one cent per Copilot Credit, custom document processing costs 8 Copilot Credits a page, which is $0.08. A new customer therefore pays roughly 60 percent more per page than the old add-on rate.

Trigger on the file arriving, get the file content, then hand those bytes to whichever premium action you licensed. AI Builder returns named fields you mapped when you trained the model. A conversion API called through the HTTP action returns the tables as structured rows or a finished XLSX. From there you write to Excel Online or drop the file in SharePoint.

Only through an action that runs OCR. A scan has no text layer, so anything that just reads file content gets nothing. AI Builder document processing and text recognition both handle images, and they are metered per page as above. If OCR is the whole job rather than field extraction, text recognition at 3 credits a page is the far cheaper of the two.

Because one of its actions is licensed separately from Microsoft 365. The HTTP action and every AI Builder action carry that flag. The flow will usually build and even test under a trial, then fail or get suspended once the trial lapses. Check the connector list on the flow details page before you promote anything to production, not after.

Compare the meters, not the sticker prices. AI Builder bills per page, so a long report is expensive even when it is one file. A conversion API on a flat monthly plan bills per file, so page count stops mattering. Work out your monthly page volume and your average pages per document, then multiply. The answer flips depending on which of those two numbers is larger.

Related comparisons

Alternatives

Convert one document before you build the flow

Drop a real file into the converter at the top of this page and look at the output. If the rows come back the way you need them, the PDF to Excel API is the same engine behind an HTTP action. For everyday work start with the PDF to Excel converter, pull only the grids with the PDF table extractor, and see the tiers on the pricing page.

Convert a PDF View Pricing