In Document Intelligence the prebuilt invoice models claim to have support for Swedish in one place, and not in another.

Question

In Document Intelligence the prebuilt invoice models claim to have support for Swedish in one place, and not in another.

Anonymous

Hello!

We have some questions about your prebuild document-intelligence invoice model (2024-11-30 GA).

On this page:

language-support

We understand that you have support for Swedish invoices.

However on this page:

schema

We can't find any references to specific Swedish elements of an invoice, e.g OCR or

payment details regarding Bankgiro (a very common Swedish payment method).

Other than parsing the resulting document response and extracting the data ourselves from the parts of the tokenized response does it exist a way for us to get this data?

Here's a list of the Swedish parts that we're missing:

Invoice type: Credit, Debit or Reminder
PaymentType: Bankgiro and PlusGiro
OCR: Used to identify an invoice, the payee enters this information during payment.
Reference Number: A reference number, similar to P.O number

Can we expect these fields to be part of the prebuilt models in the future?

How should we understand the language support, do you mean that the models can understand the Swedish language but that they don't have an understanding of Swedish invoices?

Anonymous

2025-05-27T07:24:57.9066667+00:00

Hi @Daniel Figueroa

The prebuilt invoice models do indeed claim to support the Swedish language, which means they can understand and process text in Swedish invoices. However, it appears that while the model recognizes Swedish text, it may not have the specific logic or fields you mentioned (like OCR for Bankgiro payment details, Invoice type, and Reference Numbers). This can be the reason why you aren't seeing those particular Swedish elements in the documentation or the output.Currently, if the specific fields are not recognized or structured into the model's output, your best option is to parse the response data manually, as you mentioned. You would need to extract these details from the tokenized response.As for whether we can expect these fields to be part of the prebuilt models in the future—there's typically no official word on added features until they are announced. You’re correct that the models can understand the language; however, that doesn’t necessarily mean they’ll completely comprehend all the nuances of particular document formats or locales unless specified in the model training. It might be helpful to keep an eye on updates from Microsoft or to provide feedback through official channels requesting these specific features for Swedish invoices. What's new in Azure AI Document Intelligence

Hope this helps. Do let us know if you any further queries.

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.
Manas R Mohanty 17,270 Reputation points Moderator

2025-05-28T08:36:31.6+00:00

Hi @Daniel Figueroa

Checking to see if you are getting betting result with OCR models or pointers shared from Vinodh247 (post processing, custom model training and Hybrid extraction strategy).

Attached relevant documentation on compose model operation for same.

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/train/composed-models?view=doc-intel-4.0.0 .

Please let us know if you want to share any challenges or observations.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2025-05-29T08:47:09.79+00:00

Hi @Daniel Figueroa

We have not heard from you. Hope the pointers shared were useful to you.

Thank you.
Anonymous

2025-06-02T09:05:12.7766667+00:00

Hi Manas Mohatny and Kothapally Snigdha, I've just gotten the time to review the answers here. We're getting some acceptable results I believe by using the key-value response alternative. As for now composed models is probably not what we want, but it may very well be so in the future.

Thank you for your answers and comments.

Answer accepted by question author

Vinodh247-1375 43,911 Volunteer Moderator

Hi ,

Thanks for reaching out to Microsoft Q&A.

Based on the schema documentation you referenced, the prebuilt invoice model appears to be primarily designed around common international invoice fields, which explains why you're missing:

Bankgiro/PlusGiro payment methods - These are specifically Swedish/Nordic payment systems
OCR reference numbers - The swedish OCR payment reference system
Invoice type classifications (Credit/Debit/Reminder) in Swedish context
Swedish specific reference number formats

To obtain these specific fields, you have a couple of options:

Post-Processing: After the model extracts the general fields, you can implement custom logic to parse the raw OCR output and identify the swedish specific fields. This approach requires additional development effort but allows for tailored extraction.
Custom Model Training: Azure Doc Intelligence allows you to train custom models using your own labeled data. By providing examples of Swedish invoices and annotating the desired fields, you can create a model that accurately extracts the specific information you need.
Hybrid Extraction Strategy: Use the prebuilt model for standard fields (amounts, dates, vendor info) and implement custom extraction logic for swedish specific elements from the raw OCR output.

As of now, there's no official roadmap indicating when or if the prebuilt invoice model will include Swedish-specific fields. Microsoft continuously updates its services based on user feedback and demand. If support for these fields is crucial for your operations, consider submitting feedback through Azure's official channels to highlight the need. https://feedback.azure.com/

Understanding "Language Support"

In this context, "language support" means the model can recognize and process text written in Swedish, enabling OCR and extraction of standard invoice fields. It does not imply comprehensive understanding or extraction of all locale specific elements unique to swedish invoices.

Please feel free to click the 'Upvote' (Thumbs-up) button and 'Accept as Answer'. This helps the community by allowing others with similar queries to easily find the solution.

0 comments

1 additional answer

Your answer

Anonymous

2025-05-27T07:24:57.9066667+00:00

Hi @Daniel Figueroa

The prebuilt invoice models do indeed claim to support the Swedish language, which means they can understand and process text in Swedish invoices. However, it appears that while the model recognizes Swedish text, it may not have the specific logic or fields you mentioned (like OCR for Bankgiro payment details, Invoice type, and Reference Numbers). This can be the reason why you aren't seeing those particular Swedish elements in the documentation or the output.Currently, if the specific fields are not recognized or structured into the model's output, your best option is to parse the response data manually, as you mentioned. You would need to extract these details from the tokenized response.As for whether we can expect these fields to be part of the prebuilt models in the future—there's typically no official word on added features until they are announced. You’re correct that the models can understand the language; however, that doesn’t necessarily mean they’ll completely comprehend all the nuances of particular document formats or locales unless specified in the model training. It might be helpful to keep an eye on updates from Microsoft or to provide feedback through official channels requesting these specific features for Swedish invoices. What's new in Azure AI Document Intelligence

Hope this helps. Do let us know if you any further queries.

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.
Manas R Mohanty 17,270 Reputation points Moderator

2025-05-28T08:36:31.6+00:00

Hi @Daniel Figueroa

Checking to see if you are getting betting result with OCR models or pointers shared from Vinodh247 (post processing, custom model training and Hybrid extraction strategy).

Attached relevant documentation on compose model operation for same.

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/train/composed-models?view=doc-intel-4.0.0 .

Please let us know if you want to share any challenges or observations.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2025-05-29T08:47:09.79+00:00

Hi @Daniel Figueroa

We have not heard from you. Hope the pointers shared were useful to you.

Thank you.
Anonymous

2025-06-02T09:05:12.7766667+00:00

Hi Manas Mohatny and Kothapally Snigdha, I've just gotten the time to review the answers here. We're getting some acceptable results I believe by using the key-value response alternative. As for now composed models is probably not what we want, but it may very well be so in the future.

Thank you for your answers and comments.

Answer 1

Hi Daniel Figueroa

As part of my repro

I saw change in fieldname value from "Payment method" to "Payment term" for swedish invoice as you mentioned.

Screenshot (174)

But you can extract those value with given key field and post process them with another key field value as Vinodh247 suggested

Additionally, you can also go with V3.1 or V4.0 OCR models instead of prebuilt invoice models to extract the exact value, but please make sure you have checked the language options in "Analyze option" Setting.

User's image

Screenshot (175)

As you can see, "Payment method ": Paypal is extracted as it is with OCR model.

Hope it helps.

Thank you.