Microsoft’s AI Builder credit shift puts document-processing budgets on notice

From 1 November 2026, new and renewed Power Platform and Dynamics 365 licences will lose seeded AI Builder credits, pushing organisations towards Copilot Credits and a fresh look at costs.

A recent guide to intelligent document processing may read like a buyer’s checklist, but for Microsoft administrators it highlights a more immediate issue: the economics of extracting data from documents are changing. The article points readers towards two Microsoft routes – Azure AI Document Intelligence and Power Automate with AI Builder – and its broad advice is sensible enough: start with the documents you actually handle, define where the extracted data must go, decide what level of error is acceptable and plan for human review where confidence is low.

The most useful advice in the guide is also the simplest: ignore vendor marketing claims and test the tools on your own files. That matters because document-processing products are only as good as the paperwork they are fed. In practice, the difference between a promising demo and a working deployment often comes down to scan quality, layout consistency, handwriting, tables that span pages and the tolerance of the downstream system for imperfect extraction.

For organisations already using Microsoft’s low-code stack, the larger concern is licensing. Microsoft’s licensing overview says AI Builder actions in apps and flows use AI Builder credits first, then Copilot Credits, and that model runs stop if neither is available. It also says monthly consumption resets and unused credits do not roll over. Just as importantly, a Power Automate flow that uses an AI Builder action is treated as a premium flow, with its run charges billed separately from credit consumption.

Microsoft is also changing the way capacity is supplied. According to its licensing documentation, from 1 November 2026, new and renewed Power Platform and Dynamics 365 licences will no longer receive seeded AI Builder credits. Licences bought or renewed before that date keep their seeded credits for the rest of the contract term, but new customers must rely on Copilot Credits to run AI Builder features. Microsoft says the functionality is staying in place; what is changing is the billing model.

That has real implications for cost planning. Microsoft’s rate table lists page-based consumption for invoice, receipt and identity-document analysis, as well as custom-document processing. Under pay-as-you-go, Copilot Credits are priced at $0.01 each, which works out to roughly 8 cents a page for those common document actions. Because the same licensing page says capacity is consumed per page or per token depending on the action, organisations should model usage against peak monthly volumes rather than average demand.

For teams that want more control, Azure AI Document Intelligence remains the more developer-oriented option. Microsoft’s documentation describes three main areas: Read and Layout analysis, prebuilt models and custom models. Read is designed for printed and handwritten text, while Layout can pull out text, tables, selection marks and document structure such as headings and footers. Prebuilt models cover documents such as invoices, receipts, identity documents, contracts, bank statements and mortgage paperwork.

The service is built for scale, but it has its own limits. Microsoft says paid-tier PDFs and TIFFs can run to 2,000 pages, while the free tier processes only the first two pages. File size, image dimensions and minimum text height also matter, and the service is asynchronous, so latency can vary. For custom extraction, Microsoft says five sample documents can be enough to begin training, though that is a starting point rather than a guarantee of production-grade accuracy.

For less technical users, Microsoft offers the same capability through a Power Automate connector. The connector supports prebuilt and custom models, including the newer document-analysis operations, and lets users choose specific pages from a PDF or TIFF rather than processing an entire file. That makes it easier to build selective workflows around large documents. Microsoft’s guidance also points users to prebuilt models such as invoice or layout for an initial test rather than starting with a complex custom build.

AI Builder itself remains useful for low-code teams, particularly where extraction needs to be combined with review and approval. Microsoft’s form-processing guidance says the prebuilt invoice model can return fields such as invoice identifiers, supplier details, totals, tax and line items. For custom models, Microsoft separates fixed-template documents from more varied general documents and from invoice-specific training. It also says a model can hold up to 200 collections, and that newer v4.0 models add overlapping-field support, signature detection and confidence scores for tables.

The strongest argument for staying within Microsoft’s ecosystem is not the extractor itself, but the surrounding workflow. Microsoft’s document automation toolkit ties together Power Automate, AI Builder, Power Apps and Dataverse so that extraction, human review, queueing and storage sit in one architecture. That matters because document processing rarely ends when the data is captured; low-confidence results still need a person, an audit trail and a place for exceptions to live.

In the end, the practical lesson is the same whether an organisation prefers Azure APIs or low-code tools: document processing should be designed around real files, not vendor promises. The guide’s advice to test on your own material is sound, but Microsoft administrators should focus on a more urgent point than the product comparison alone. Seeded AI Builder credits are being phased out for new and renewed licences, and the cost structure is shifting towards Copilot Credits. The capability remains, but anyone budgeting for it should recalculate before the next renewal.

Source Reference Map

Inspired by headline at: ^[1]

Sources by paragraph:

Source: Fuse Wire Services