Software Integrations

D365 F&O Batch Vendor Invoice Import via DMF

Batch import vendor invoices into D365 F&O via DMF: destination-artifact choice, header/line file shape, staging-table fixes, and workflow handoff.

Published
Updated
Reading time
30 min
Topics:Software IntegrationsDynamics 365Finance & OperationsDMFvendor invoice import

Dynamics 365 Finance & Operations imports vendor invoices in batch through the Data Management Framework (DMF), using a paired Vendor invoice header entity and Vendor invoice line entity that link through a header reference: each line row's HeaderReference points at its header row, and many integrations set that reference to the invoice number. Before that import can run cleanly, the batch needs a destination: one of three F&O artifacts, each with its own posting behaviour and its own downstream consequences for approval and matching.

The three destinations are the vendor invoice journal (lightweight, journal-style entries useful for cleanup batches and historical migrations), the vendor invoice register plus invoice approval journal (the classic two-step capture-then-approve flow with separate posting), and pending vendor invoices (the destination where imported invoices can be submitted to the vendor invoice workflow and matched to product receipts automatically, when the Accounts payable automation parameters are switched on). The destination decision drives both the entity selection and the shape of the import file.

That single decision, destination first and entities and file shape second, is what separates a d365 f&o batch vendor invoice import that posts cleanly from one that stalls in staging or lands invoices in the wrong place. Pick the wrong destination and a migration of already-approved historical invoices can end up in approval queues it should never have reached; pick the right one and the same file follows the approval and matching path the organization has configured for it.

This article is the DMF batch path specifically. It takes a structured file (commonly from an upstream extraction step, a migration source, or a third-party capture service), runs it through DMF data entities into the destination artifact you chose, and hands the result off to the workflow and matching layers configured on your environment.

Choosing the destination: vendor invoice journal, register, or pending invoices

The three destination artifacts solve different problems, and the right choice depends entirely on what AP wants to happen after the invoice lands.

Vendor invoice journal is the lightweight option: journal-style entry and direct posting, outside the vendor invoice workflow. Once the journal posts, the invoice transaction posts with it. This is the right destination for cleanup batches and for migrations of historical invoices that were already approved in the legacy system, where re-running them through F&O's vendor invoice workflow would be both pointless and noisy. It is also the right destination for adjusting entries and corrections that AP wants to land as journal entries rather than as workflow-tracked vendor invoices. The trade-off is that nothing about this path engages F&O's matching policies or the vendor invoice workflow; if you need either, this is the wrong destination.

Vendor invoice register plus invoice approval journal is the two-step capture-then-approve flow. The register stage captures the invoice and posts a credit to the vendor account against a holding account, recognising the liability the moment the invoice arrives, even before expense coding or approval has happened. Some time later (typically days or weeks, depending on the organization's close cadence), a second step picks up the registered invoice, debits the expense account or the PO accrual, clears the holding account, and posts the invoice into its final state. Microsoft documents two routes for that second step: the invoice approval journal, and the invoice pool, where the registered invoice becomes a pending vendor invoice that goes through receipt matching and workflow like any other. A DMF import reaches the register by the second route. With the Vendor invoice automation with tax registration feature, setting the VENDORINVOICETYPE field on the Vendor invoice header entity to InvoicePool makes the import create and post an invoice register journal and create the pending vendor invoice in the invoice pool (Microsoft Learn's invoice pool guidance). This fits organizations where AP wants liability recognition on receipt of the invoice but the expense coding and approval work belongs to a different person, often working on a different timeline, and organizations whose country's tax registration rules want the accrual recorded when the invoice arrives. A d365 f&o vendor invoice register batch lands in this two-step flow when AP needs the liability visible immediately but the rest of the invoice can wait for the approver. The closest parallel on the Dynamics SMB side is Business Central's invoice registration journal as the SMB counterpart to F&O batch intake, which fills a comparable role for BC organizations.

Pending vendor invoices is the destination that connects the imported invoice to the vendor invoice workflow and to product receipt matching. Once a row lands here, the invoice is the same kind of record as one captured interactively through the vendor invoice center or surfaced through Intelligent OCR: matching policies evaluate it against the purchase order lines it references, and once it is submitted to workflow, approval routing resolves against the invoice's amount, dimensions, and vendor. Whether submission and receipt matching happen without anyone touching the invoice is set by two Accounts payable parameters, covered in the workflow section below. An f&o pending vendor invoices import is the right choice when the batch needs to follow the same governance path as in-flow invoices: it is the natural destination for invoices coming from an external capture service, an extraction pipeline upstream of F&O, or any source where the batch is operational rather than migratory.

The downstream consequences of these three choices diverge sharply. The vendor invoice journal posts on its own and never touches the vendor invoice workflow or matching policies. The register flow splits the posting in two: through the invoice approval journal, approval happens in that journal's own flow, separate from the vendor invoice workflow; through the invoice pool, which is where a DMF import of the register lands, the second step is a pending vendor invoice. Pending vendor invoices, whether imported directly or created from the pool, are the only records here that go through the vendor invoice workflow and product receipt matching, and the only ones that show up in the vendor invoice workspace alongside invoices captured through the other intake paths.

The decision mistakes are predictable. The most common is dumping migration data (old, already-approved invoices being brought across from a legacy system) into pending vendor invoices. With automatic workflow submission switched on, every one of those historical records is submitted to the vendor invoice workflow, and approvers are flooded with thousands of items that should have skipped it entirely; with it off, the same records sit in pending invoices until someone submits or deletes them. The opposite mistake is importing live operational invoices into the vendor invoice journal because it is the simplest path, only to discover later that the matching policies the organization relies on were never applied and that there is no workflow audit trail on any of these invoices. The destination has to match the use case; choosing it for convenience rather than for fit is where most multi-week clean-up projects start.

The DMF data entity model for vendor invoices

The DMF data entity model for vendor invoices is built on a header-and-line pair: a Vendor invoice header entity and a Vendor invoice line entity. Each line row is keyed by its HeaderReference, a foreign key to the header row, plus an InvoiceLineNumber (Microsoft's Vendor invoice line entity reference). Both entities have to be mapped together in the same DMF data project, with the import order honouring the parent-child relationship so the header row lands before any of its line rows. A DMF data project that includes only the header entity will import invoice headers with no lines attached; a project that includes both but runs the line entity first will fail every line row at validation, because the header it references does not yet exist.

The entity you map has to correspond to the destination you chose in the previous section. The Vendor invoice header and Vendor invoice line entities create pending vendor invoices; with VENDORINVOICETYPE set to InvoicePool, the same pair posts an invoice register journal and leaves the pending invoice in the invoice pool. The vendor invoice journal sits behind a pair of its own, the Invoice journal header and Invoice journal line entities. This is the detail that catches teams who treat the entities as interchangeable: the journal entities will not land data in pending vendor invoices, and the pending-invoice entities will not produce invoice journal entries. The entity choice is part of the destination choice, not a separate decision to make later.

The header reference is where most entity-level failures originate. It is an identifier from the source system, not the invoice ID, though many integrations set it to the invoice number. Whatever the source uses, the line row's reference has to carry exactly the same value as its header row's, character for character. A header exported as INV-00123 and lines exported as 00123 do not join; the lines fail to attach to their header. This pattern often survives all the way through a source system unnoticed, because header and line exports are frequently built separately, and one applies a prefix, zero-padding, or a stray leading space that the other does not.

Architecturally this is one place where Dynamics 365 Finance & Operations and Business Central diverge sharply. BC does not use this entity model at all: BC's import path runs through Configuration Packages or an import extension, with field-mapping happening against a CSV or Excel template rather than against parent-child data entities. Teams with experience importing invoices into Dynamics 365 Business Central via Configuration Packages and CSV sometimes assume the F&O path is a similar field-mapping exercise; it is not. F&O's DMF model is staging-table-based with explicit entity validation, and the BC mental model does not transfer to it.

Shaping the import file so DMF accepts it

Once the destination and entity choice are settled, the work shifts to the import file. F&O takes Excel, CSV, and several other DMF-supported formats; what matters is not the file type but the field-level shape against the entities you mapped.

The header row needs, in practice: the header reference its lines will point at, the invoice number, a vendor account that exists in the target legal entity (not just somewhere in the tenant), an invoice date, due date or terms of payment, and a currency code. Each of these has to be valid against the destination company's setup, not against any company in the environment: a vendor master that is set up only in USMF will not satisfy an invoice imported into DEMF, even when the vendor name and contact details are notionally the same.

The line row needs the HeaderReference that ties it to its header and an InvoiceLineNumber, a currency, an item number for stocked items or a procurement category otherwise, quantity and unit, unit price or line amount, and any financial dimensions the legal entity marks as mandatory on vendor invoice postings. Lines that bill against a purchase order also carry the PurchaseOrder number, and the PurchLineNumber where the line maps to a specific PO line. The line entity is where most of the substantive data lives, and where most of the validation effort lands.

Several optional fields change posting behaviour rather than just shaping the data. A posting profile on the header determines which ledger accounts the invoice posts against and can override the vendor's default profile. Sales tax codes and item sales tax groups resolve against the destination legal entity's tax setup: a tax code defined in one company will not resolve in another, and there is no tenant-wide tax code list to fall back on. Financial dimension values come in as a single display string, DimensionDISPLAYVALUE, in the format set up under Financial dimension configuration for integrating applications. Freight, customs, and similar amounts come in through entities of their own, Vendor invoice charges for the header and Vendor invoice line charges for the line, mapped in the same project after the rows they depend on. Document date versus posting date matters when the organization runs period control or accrues across period boundaries. Payment reference fields populate the data the bank file generation step needs later.

Tax code mapping deserves its own attention because it fails so reliably. Tax codes have to exist in the destination legal entity and resolve cleanly against that legal entity's tax setup. The same code name in two different companies can refer to different rates or different reporting structures, and a code that resolves in one will not necessarily resolve in another. Pre-validate every tax code against the destination company's tax setup before submitting the import, ideally by querying the destination company's tax code table from a small validation step in whatever produces the source file.

Legal-entity scoping runs through every other validation. An imported invoice belongs to the company the DMF job runs in, unless the integration sending the data package names a company ID in its HTTP request, which overrides the job's company (Microsoft Learn on vendor invoice automation). That company's setup is what every row validates against: vendor account, tax codes, financial dimensions, posting profile, exchange rates, the lot. Microsoft's documentation describes no per-row company scoping for these entities, so a batch that spans companies is split per company before it is imported, as the multi-legal-entity section below covers.

Currency handling adds another failure surface. An exchange rate for the invoice currency on the invoice date has to exist in the destination legal entity's currency exchange rate setup; otherwise the import fails on rate resolution before the invoice is ever posted. Triangulation currencies, where the invoice currency differs from both the company's accounting currency and the reporting currency, compound this, because rates have to be present for each leg of the conversion. Loading currency rates is a separate maintenance task that has to be in place before the import runs, not something the DMF job will handle for you.

Document attachments come in through their own data entity. The Vendor invoice document attachment entity sits in the same DMF data project as the header and line and depends on the header: it attaches PDFs or images of the source invoice to the invoice header, and it does not support line attachments. Where an invoice carries several attachments, one can be flagged Is default attachment through the entity, and that document opens automatically in the side-by-side attachment viewer when the pending invoice is opened, which is where reviewers and approvers see the source document beside the imported data.

Clearing the staging table: why imports halt and how to fix them

DMF stages every imported row in an entity-specific staging table before writing anything to the target. Validation happens on the way from staging to the target, and a row that fails validation stays in staging with its error. The target record (the pending vendor invoice, the invoice register journal, or the invoice journal entry) is never created until that row passes. This is the safety net: a half-validated batch leaves staging rows behind but leaves the target tables untouched for the rows that failed.

The error patterns that show up in staging are predictable enough to enumerate, and each has a specific cause and a specific fix. These are the d365 f&o vendor invoice staging table errors most teams see week after week.

  • Vendor not found in the target legal entity. The vendor account on the row exists somewhere in the tenant, usually in another legal entity that was the source for setting it up, but not in the legal entity the invoice is imported into. F&O does not automatically share vendor masters across companies; each company maintains its own vendor table unless the organization explicitly configures shared vendor master data. Fix: confirm the vendor is set up in the destination legal entity (and create it if not), or move the invoice to the batch for a company where the vendor does exist.
  • Tax code mismatch. The tax code or item sales tax group on the row does not exist in the destination legal entity's tax setup, or it exists but does not resolve cleanly against the tax group on the vendor or item. The same code name in different companies can mean different things, and F&O validates against the destination company's setup specifically. Fix: validate the source file's tax codes against the destination legal entity's tax code table before submitting, or update the source mapping to use codes that exist in the destination.
  • Currency exchange rate missing. No exchange rate is defined for the invoice currency on the invoice date in the destination legal entity's currency exchange rate setup. This happens most often for back-dated invoices where the historical rate was never loaded, or for currencies the company has only recently started transacting in. Fix: load the missing rate (or rates, for triangulation cases) into the exchange rate setup, then re-run the affected rows.
  • Header-line key break. Line rows carry a HeaderReference that has no matching header row in the same batch. Almost always a data-quality issue at the source: a prefix or zero-padding applied to one export and not the other, a stray leading space, or a header row that was filtered out by an export step while its lines survived. Fix: re-export the source so header and line rows carry exactly the same reference value.
  • Mandatory dimension missing. The legal entity requires a financial dimension on vendor invoice postings (cost centre, department, project, or similar) and the source row does not supply a value for it. F&O will not default a mandatory dimension; the row fails validation. Fix: add the dimension value to the source file, or reconfigure the legal entity's dimension setup if the requirement was set up in error.
  • Posting profile mismatch. The posting profile name on the row does not exist in the destination legal entity, or has been renamed. Fix: align the profile name on the source row with what is configured on the destination.

AP does not have to work these from the Data management workspace. Accounts payable > Invoices > Import failures > Vendor invoices that failed to import lists the header staging records of the Vendor invoice header entity with the error message DMF issued for each. Edit opens a record to fix it, and Create pending invoice then creates the invoice in the background; an attachment on a failed invoice does not display there and has to be reattached. Security set on the import job, by user role, by user, or by legal entity, carries over to the page, so exceptions can go to one shared-services team or to each company's own AP staff.

The cleanup pattern for a failed staging batch matters as much as the diagnosis. Deleting staging rows for a failed run is generally safe: nothing has been posted to the target table, so removing staging rows for the failed portion does not leave orphaned records in the actual vendor invoice tables. The case to be careful about is partial success, where some of the batch posted and some failed. In that situation, the staging rows for the failed portion are the audit trail of which rows landed where; do not delete them until the posted portion has been verified against the target tables and you know exactly what state the data is in.

Re-running corrected rows beats re-importing the whole batch. DMF does not deduplicate imports. With the Check the invoice number used parameter set to Reject duplicate, F&O catches an invoice number that is already on a posted invoice, but the check compares against posted invoices, so re-running a full file after a partial success can still leave two unposted pending invoices with the same number to find and delete. The discipline is: fix the source file for the failed rows specifically, import just those rows, leave the posted rows alone.

There is a deeper pattern worth naming when the same staging errors recur batch after batch. When mismatched header references, inconsistent tax code naming, missing financial dimensions, or vendor-not-found errors show up across every weekly load, the problem is not in F&O at all: it is in the upstream extraction or capture step that is producing inconsistent data. Fixing it in the staging table over and over is the most expensive way to handle it. The cheaper fix is upstream: making the extraction step produce clean, normalized, DMF-ready data the first time, every time.

Workflow approval and PO matching on DMF-imported invoices

Once an invoice lands cleanly in its target table, the next question is whether it reaches the vendor invoice workflow and whether matching runs. That depends first on which destination the import targeted, and for pending vendor invoices, on how Accounts payable is configured.

Invoices landed in pending vendor invoices reach the vendor invoice workflow in one of two ways, set on the Vendor invoice automation tab of Accounts payable parameters (Microsoft Learn on vendor invoice automation options). With Automatically submit imported invoices to workflow set to Yes, which needs an active vendor invoice workflow in the legal entity, a background task submits imported invoices once they have completed the product receipt matching job. With it set to No, someone has to submit each imported invoice to workflow. When either automation parameter is on, imported invoices arrive with Include in automated processing switched on, which also keeps them from being edited while automation has them. The background tasks run hourly or daily, as set under System administration > Setup > Process automations, so an imported invoice reaches an approver's queue at the next run, not the moment the import finishes. For invoices to go from import to posting with no one touching them, the workflow itself needs an automated posting task. Once submitted, the workflow engine treats a pending invoice from a DMF batch the same as one an AP clerk typed in.

Invoices landed in the vendor invoice journal never enter the vendor invoice workflow; approval, if any, happens on the journal itself, through a vendor invoice journal workflow where one is configured or through whoever posts it. Register imports through the invoice pool post the register journal on import and then follow the pending-invoice path above for matching and workflow. The invoice approval journal route has its own journal approval workflow, configured independently of the vendor invoice workflow.

Approver routing on a submitted pending vendor invoice resolves the same way whether it was imported or typed in: by the fields on the invoice itself. Amount, legal entity, financial dimensions, vendor, and project (where applicable) are what the workflow evaluates. An imported invoice for $250,000 routes to the higher-tier approver just as an in-product capture for the same amount would. If those fields are missing or wrong, the routing falls through to whatever default the workflow is configured with: sometimes a catch-all approver, sometimes an outright failure that parks the invoice in an unassigned state requiring manual intervention. The fields that drive routing are part of the import contract, not nice-to-haves.

The record an import leaves is worth planning for. The originating user for an imported invoice is the account that ran the DMF job, not whoever produced the source file, and the header entity's IMPORTEDAMOUNT field keeps a snapshot of the source total, so the imported invoice can be checked against what the source said it was. Organizations with strict provenance requirements often route DMF imports through a dedicated service account so the trail is unambiguous, rather than letting the submitting administrator's account become the recorded originator across thousands of historical invoices.

PO and product receipt matching on imported invoices depends on what each line carries. Matching policies are configured per legal entity, vendor, item, or procurement category, and they evaluate invoice lines against the PO lines they reference. That reference comes from the import file: PurchaseOrder on each line that bills against a purchase order, and PurchLineNumber where the line maps to a specific PO line. The line entity has no product receipt field. For lines under a three-way matching policy, the Automatically match product receipts to invoice lines parameter (available once invoice matching validation is enabled) has a background task match posted product receipts to the invoice lines, retrying up to the limit set in Number of times to attempt automatic matching; when that limit is reached, the invoice is removed from automation and left for AP to match. With the parameter off, an AP user matches receipts on the pending invoice. When the PO fields are not populated at all, matching has nothing to work against, and the invoice posts as a non-PO invoice. That is rarely what AP actually wants for an operational batch; non-PO posting bypasses the accrual the receipt created, and the open PO line stays open with no matching invoice against it.

The contrast with the vendor invoice center and Intelligent OCR paths is in when matching happens. Those paths capture invoices interactively, with PO and receipt suggestions surfaced at the point of capture, and the AP user resolves matching exceptions inline before the invoice ever enters the workflow. A clerk capturing a vendor invoice in the vendor invoice center will see immediately if the PO has no matching receipt, or if the amount is outside tolerance, and can resolve it before submitting. A DMF-imported invoice arrives with whatever PO references the import file carried: if the PO is wrong, or no posted receipt matches the line, the issue surfaces later, in the automatic matching attempts, in the workflow, or in the pending invoice list, rather than at point of capture. The choice between paths is partly a choice about where in the process the exception-handling work happens.

Many F&O environments run with several legal entities (operating companies, regional subsidiaries, shared services centres), and any AP batch of meaningful size will touch more than one. Because each import lands in one company, there are two viable patterns for handling this, and the right one depends on how the source data arrives and what does the splitting. An f&o multi-legal-entity vendor invoice import sits in one of these two shapes.

Pattern A: one DMF job per legal entity. Each job runs in a single company, and the source file is split or filtered per company before the job runs. Operationally this is the simpler pattern: each job's failures only affect one company's staging rows, and a failure in one company does not block any of the others. It fits naturally when the source data is already segmented by company, which is often the case when the upstream extraction step is configured to produce one file per receiving legal entity, or when separate AP teams own each company's batch.

Pattern B: one combined feed, split per company by the integration. The source data arrives as a single batch, typically because the extraction or routing step upstream determines each invoice's receiving company from the bill-to address, a vendor-to-entity map, or a PO lookup. The integration then sends one data package per company, each carrying that company's ID in its HTTP request, so one integration feeds every company from the same feed. What it cannot do is put several companies' invoices in one package: the company named in the request, or else the job's company, is the company every invoice in that package lands in. Every company's setup (vendor account, tax codes, posting profile, mandatory dimensions) still has to be valid for the invoices sent to it, and when one company's setup is incomplete, its package fails on the same vendor-not-found or tax-code-mismatch errors covered earlier while the other companies' packages go through.

Vendor mapping across companies is the recurring complication. A vendor doing business with multiple group companies will typically have a different vendor account ID in each company unless the organization has explicitly configured shared vendor master data. The vendor name "Acme Corp" can map to vendor account V-1042 in one company and ACME001 in another, and the import file has to use the right account ID for the destination company on every row. Relying on vendor name to do the mapping at import time is not an option: DMF validates against vendor account, not against name.

Inter-company posting is the other complication. When an invoice received in one legal entity needs to be posted across to another (a common pattern when a shared services centre handles AP on behalf of operating companies, or when a parent company pays a vendor and rebills the subsidiary), F&O's inter-company accounting setup governs how the cross-entity entries generate. The inter-company trading relationships, posting profiles, and dimension defaults all have to be configured in advance. DMF imports do not bypass the inter-company setup; they trigger it the same way an interactively entered invoice would. The cross-postings happen automatically when the setup is in place, and they fail when it is not.

Cross-company allocations and charges (rebill arrangements, shared-cost allocations, inter-company management fees that need to land on multiple companies) usually do not survive a naive single-import approach. The two workable options are breaking them out in the source data into separate lines for each receiving company, imported with that company's batch (where the source extraction step is sophisticated enough to know about allocations), or importing the invoice into the company that received it and relying on F&O's inter-company functionality to generate the cross-postings. Which is right depends on whether the allocation logic lives in the source system or in F&O itself.


When to extract upstream vs use F&O's native invoice automation

DMF batch import is one of three viable intake paths for vendor invoices in F&O, and it is not always the right one. F&O ships with its own native intake (the Intelligent OCR capability inside the invoice capture solution, and the vendor invoice center for interactive entry), and that path is well-suited to a specific set of conditions: invoice volumes that are moderate rather than overwhelming, a document mix that is reasonably uniform, and an AP team that wants point-of-capture control over matching exceptions and dimension assignment. When those conditions hold, the native path keeps the entire workflow inside F&O, with no external dependencies and no separate audit trail to reconcile.

The DMF batch path is the right choice when the conditions are different: invoices arrive in batches rather than one at a time, the source data is already structured (or will be, after an upstream extraction step), and the organization wants explicit control over field-level transforms, multi-entity coding, and data quality before any of it touches F&O. The two paths are complements, not substitutes; many F&O environments run both: Intelligent OCR for in-flow invoices that arrive by email or upload, DMF batch for periodic loads from external systems, migration runs, or third-party capture services.

Where upstream extraction belongs in this picture is specifically when the source documents are unstructured and varied: PDFs, scans, vendor portal exports, the mixed stack that real AP teams actually receive. Producing a clean, structured, line-level dataset shaped to the F&O Vendor invoice header and Vendor invoice line entities before DMF runs converts a heterogeneous problem (every invoice looks different) into a uniform one (every batch lands the same way). The staging-table error patterns covered earlier (vendor not found, tax code mismatch, header-line key break, mandatory dimension missing) become rare rather than routine, because the upstream step is enforcing data quality at the entity boundary instead of leaving it for DMF to discover and reject.

The decision factors come into focus when you compare the paths against the work they have to do:

  • Volume and variety. High volume with high document variety pushes hard toward upstream extraction; the same prompt or extraction logic processes the whole stack and produces the same DMF-ready output every time. Moderate volume with uniform documents fits native intake, where Intelligent OCR's defaults are well-matched to the input.
  • Multi-entity coding control. When each invoice's legal entity, dimensions, and tax coding are determined by upstream logic (a routing service, a vendor-to-entity map, an invoice-to-PO lookup), upstream extraction can supply those fields deterministically before DMF runs. Intelligent OCR uses its own defaults and learned mappings, which suit some organizations and not others.
  • Audit trail. Native intake keeps the audit trail inside F&O end to end; upstream extraction adds an external step with its own logging that needs to reconcile against the F&O trail. Some organizations strongly prefer the single-system trail; others explicitly want the external step for control reasons (a separate party producing the extraction record, a different system enforcing the data-quality contract).
  • Field-level transforms. When invoices need normalizing or enriching before they hit F&O (date format standardization, tax-code re-mapping from supplier conventions to internal codes, vendor name resolution against the master list, currency conversion at a specific rate), upstream extraction is the natural place to do it. F&O's native intake assumes the data is already in the shape it expects.

This is where Invoice Data Extraction enters as the upstream step for the DMF path. The product converts a mixed batch of invoice PDFs, scans, and portal exports into a structured spreadsheet: natural-language prompt in, structured Excel, CSV, or JSON out. A batch can hold up to 6,000 files and a single PDF up to 5,000 pages, with every page and every file read and checked the same way as the first, so the output keeps the same column structure and field-level rules on every invoice regardless of supplier format. A prompt configured to produce the F&O header and line shape (header reference, invoice number, vendor account, receiving legal entity, currency, line number, item or procurement category, quantity, unit price, line amount, PO number and PO line where the invoice bills against an order, and financial dimensions where the organization makes them mandatory) turns the upstream extraction into the entity boundary that DMF then consumes cleanly. That is the practical role: extract vendor invoice PDFs into a DMF-ready dataset once, on a stable prompt, so the staging table stops being the place where data-quality problems are discovered.

For readers whose answer is to stay on F&O's native intake, the complementary coverage is Dynamics 365 Finance's native Intelligent OCR and vendor invoice center workflow, which goes into the native automation path in depth. For BC organizations that landed here on the wrong product, the parallel native intake on the BC side is Business Central Payables Agent for native PDF invoice intake, which fills a similar role to F&O's Intelligent OCR for AP teams on the SMB platform.

The strategic point underneath all of this is that the intake quality determines the ceiling on what AP automation can achieve. The Hackett Group's 2025 Digital World Class Matrix for accounts payable found that top accounts payable platforms deliver 60% touchless, straight-through invoice processing and 3.5X higher AP productivity than peers, with 59% faster cycle times. Those numbers compound from clean upstream data: invoices that arrive at the workflow already coded correctly, matched against the right PO, landed in the right legal entity. Whether that data quality comes from F&O's native intake doing its job well or from an upstream extraction step producing DMF-ready output, it is the lever, not a footnote.

Extract invoice data to Excel with AI

Upload your invoices, say what you need in plain words, and download the spreadsheet.

  • A panel of AI agents has to agree on every value
  • What the panel cannot agree on is flagged as Review Needed
  • 50 free pages every month, no subscription
  • Any layout, any language, scans and phone photos
  • Numbers come through as numbers and dates as dates
  • Files encrypted and deleted within 48 hours