Textract vs Bedrock Data Automation

Short answer: Amazon Textract AnalyzeExpense and Amazon Bedrock Data Automation both turn invoices into structured data inside the same AWS account, but they are priced very differently for that job. AnalyzeExpense costs $0.01 per page, or $10 per 1,000 pages, and returns invoice fields against a fixed schema with no setup. Bedrock Data Automation costs $0.01 per page for Standard Output and $0.04 per page for Custom Output, so getting your own field schema out of BDA costs $40 per 1,000 pages, four times the AnalyzeExpense rate.

Last updated August 2026. Every rate below was taken from the official AWS Textract and Amazon Bedrock pricing pages on the date shown, including the worked pricing examples AWS publishes there. Confirm current rates before you commit budget, since AWS pricing changes and varies by region.

The pricing difference in one table

The line that surprises most teams is the fourth one. BDA Standard Output costs exactly the same per page as AnalyzeExpense, $0.01, but it does not give you invoice fields. It gives you a document analysis: text, layout, tables, figures. If you want vendor name, invoice number, due date and a totals breakdown out of BDA, you need Custom Output and a blueprint, and that is where the price jumps to $0.04.

How BDA Custom Output pricing scales with field count

Custom Output has a pricing rule worth understanding before you design your blueprint. The $0.04 per page rate covers any blueprint with 30 fields or fewer. Above that, AWS adds $0.0005 per page for each additional field.

AWS publishes the formula in its own pricing example: a blueprint with 40 fields is priced at $0.040 plus 10 additional fields multiplied by $0.0005, which comes to $0.045 per page. So a detailed invoice blueprint capturing header fields plus a wide set of line item attributes can drift well past $0.045 without anyone noticing, because the field count grows during development and the price is never shown per blueprint in the console.

A practical consequence: adding twenty extra fields to a blueprint costs $0.01 per page, which is the entire price of an AnalyzeExpense call. Field count is a budget decision on BDA in a way it simply is not on Textract.

What each service actually returns

AnalyzeExpense is purpose-built for invoices and receipts. It returns summary fields such as vendor name, invoice date, invoice receipt ID, due date, subtotal, tax and total, plus line item groups, each with a normalized field type and the raw text it came from. OCR is included in the output at no extra charge, so you are not paying for Detect Document Text on top. The schema is fixed, which is the point: you get invoice semantics without defining anything.

Bedrock Data Automation is a general multimodal service. It handles documents, images, video and audio through one API, generating Standard Output using modality-specific defaults, or Custom Output shaped by a blueprint you write against your own data schema. For documents that means you control exactly which fields come back and what they are called, which matters if the output has to land in an existing warehouse table without a mapping layer.

That flexibility is real, and for mixed content estates it is the reason to pick BDA. Processing invoices, contracts, scanned photos and recorded calls through one service with one billing line is genuinely simpler than stitching together four APIs. BDA also integrates with Bedrock Knowledge Bases as a parser for retrieval pipelines, billed at the Standard Output rate of $0.01 per page.

Cost at realistic invoice volumes

AWS confirms the 100,000 page figure in its own example: extracting data from 100,000 invoices with AnalyzeExpense at $0.01 per page in US West (Oregon) totals $1,000. Past a million pages the rate drops to $0.008, so 1,500,000 pages costs $14,000 rather than $15,000.

Neither service is expensive in isolation. The gap only matters at volume, and it matters most when a team picks Custom Output for a job the fixed schema already covered. Paying $4,000 a month for blueprint-defined invoice fields when $1,000 a month buys the same fields from a purpose-built API is the single most common way this decision costs money.

Free tiers are not comparable

Textract offers a free tier of 100 AnalyzeExpense pages per month, and critically it lasts three months from account creation rather than renewing indefinitely. That is enough to validate output quality on your own documents and nothing more. Plan to be paying by month four.

Bedrock Data Automation has no equivalent standing free allowance for document processing, so evaluation costs real money from the first page. At $0.04 per page for Custom Output, testing a blueprint across a representative 500 page sample costs $20, which is trivial, but it is worth knowing the meter runs during the proof of concept.

If you are comparing against services outside AWS, the free tier picture changes a lot. Adobe, for instance, gives 500 free Document Transactions every month with no expiry, though its paid tier has no published price at all. The full breakdown is in our guide to Adobe PDF Extract API pricing, which covers the Document Transaction metering rules and why Extract is billed per 5 pages while most other operations get 50.

Which one should you pick for invoices?

Pick AnalyzeExpense when your documents are invoices, receipts or bills and you want vendor, dates, totals, tax and line items without designing anything. It is the cheapest path to invoice semantics on AWS, it includes OCR, and the fixed schema is a feature rather than a limitation for accounts payable work. Our page on AWS Textract pricing breaks down every API in the family, including the async limits and the English-only constraint on AnalyzeExpense.

Pick BDA Custom Output when the field schema genuinely has to be yours, when you are processing several content types through one pipeline, or when the documents are not invoices at all and no prebuilt schema exists. Paying 4x for control you actually need is a reasonable trade. Paying 4x for control you do not need is not.

Pick BDA Standard Output when you want document analysis for a retrieval or search pipeline rather than typed business fields. At $0.01 per page through the Knowledge Bases integration it is priced sensibly for that job.

The costs that do not appear on either pricing page

Both services return data. Neither returns a finished accounts payable workflow, and that is where budgets usually break. Somebody still has to validate that the extracted total matches the sum of line items, handle the invoices where the vendor name comes back as the remit-to address, map vendors to your supplier master, route exceptions to a human, and keep the whole thing running when a supplier redesigns their template.

Async Textract and BDA jobs are queue-based, so a pipeline can stop producing results without throwing a visible error anywhere your team will notice, which is why it pays to watch the processing endpoint for silent failures rather than discovering a three-day gap during month end close. Engineering time for the surrounding logic routinely exceeds the per-page API spend by an order of magnitude at mid volumes, so compare total cost of ownership rather than rate cards.

That calculation is the reason many finance teams skip the raw API layer entirely and use a finished extraction tool, where header fields and line items arrive validated at a flat monthly price with no per-page meter to model and no blueprint to maintain.

Frequently asked questions

Is Bedrock Data Automation cheaper than Textract?

Only in Standard Output mode, where both cost $0.01 per page and neither is cheaper. For structured invoice fields, BDA Custom Output costs $0.04 per page against $0.01 for Textract AnalyzeExpense, making BDA four times more expensive for that specific job. Above 30 blueprint fields, BDA costs more still.

Can Bedrock Data Automation replace Textract?

For many document jobs yes, and BDA additionally handles video, audio and images that Textract cannot. But for invoices and receipts specifically, AnalyzeExpense returns a purpose-built schema at a quarter of the Custom Output price, so replacing it with BDA usually raises cost without improving output quality.

Does Bedrock Data Automation include OCR?

Yes. BDA performs document analysis including text and layout as part of Standard Output at $0.01 per page, with no separate OCR charge. Textract AnalyzeExpense also includes OCR in its output, so neither service bills Detect Document Text on top when you use its higher-level features.

How much does a BDA blueprint with more than 30 fields cost?

AWS charges $0.040 per page for up to 30 fields, then adds $0.0005 per page for every field above 30. A 40 field blueprint costs $0.045 per page and a 50 field blueprint costs $0.050 per page. The same rule applies to images, starting from $0.005 per image for 30 fields or fewer.

Which AWS service is best for invoice data extraction?

Amazon Textract AnalyzeExpense is the best-value AWS choice for invoices, at $10 per 1,000 pages with a prebuilt invoice schema, included OCR and no blueprint work. Choose Bedrock Data Automation Custom Output instead only when you need a bespoke field schema or are processing multiple content types through one pipeline.