Finance | Project Operations, Human Resources, ...

Invoice capture and OCR in D365 F&O

(2) Share

Report

Posted on by Benjamin Cleaver CM...

We are in the infant stages of reviewing and attempting to set up (in a test environment) the out of the box invoice capture and OCR for our accounts payable.

We are a US-based manufacturing company with 2 locations and have a small AP department.

Has anyone recently implemented this? Any implementation tips/tricks? What were some of the biggest struggles? What were some of the biggest wins?

Thanks,

Benjamin Cleaver

Controller

Australian Gold, LLC

Categories:

Dynamics 365 Finance

I have the same question (0)

All responses (3)

Answers (0)

Sort by

Suggested answer

Giorgio Bonacorsi 2,232 on at

Like (2)

Report
Copy link

Link copied!
Hello Giorgio,

Yes, I did it. You can find the articles here regarding the configuration and the overall process and configuration: https://dax365fo.org/index/.
My observation is that invoice capture works well for simple cases, such as:

Service invoices without a PO;

Invoices with a single PO, where the PO number generated by D365FO gets properly inserted into the layout. It's this case you can have an auto-match, otherwise is manual

There is already a generic AI model available to use as the OCR engine. However, if you need to create a new custom one, it’s not particularly complicated: https://learn.microsoft.com/en-us/ai-builder/form-processing-train

The main requirement is that you need to provide a collection of your vendor invoice layouts (samples) to submit and train your model.

Thank you,
Giorgio

Was this reply helpful? Yes No
Suggested answer

Syed Haris Shah 1,613 Super User 2026 Season 1 on at

Like (1)

Report
Copy link

Link copied!
Hi @Benjamin Cleaver CMA MBA MSA,

M biggest recommendation is to treat it as a process improvement initiative not just a feature toggle.

For a small AP team like yours, it can deliver solid value, but success depends on setup and governance:

Start with a pilot (a few high-volume, consistent vendors).

Clean up vendor master and PO discipline first — most early issues are matching/data quality, not OCR.

Enable continuous learning and make sure corrections are consistently reviewed.

Plan for exception handling ownership — automation shifts effort from keying to managing exceptions.

Biggest struggles: PO matching tolerances, receiving timing issues, and exception backlogs.
Biggest wins: reduced manual entry, faster cycle times, and better visibility into AP performance.

If your vendor base is fairly standardized and your PO process is strong, you should see meaningful efficiency gains.

If you find this answer helpful, please consider verifying the answer. 👍

Regards,

Syed Haris Shah

Was this reply helpful? Yes No
Devanshi shah 21 on at

Like (1)

Report
Copy link

Link copied!
Hi @Benjamin Cleaver CMA MBA MSA,

Yes, I’ve implemented this for a customer using US Based customer.
My recommendation is to start with a thorough requirement‑gathering exercise—specifically around:

Types of vendor invoices

Invoice layout complexity

Invoice charges

Taxation structure implemented

Any variations across vendors

Microsoft provides some helpful documentation on the Invoice Capture application, and since the product is still evolving, there are several useful features planned in the roadmap.

Useful links:

Invoice capture solution overview - Finance | Dynamics 365 | Microsoft Learn

What's new or changed in Invoice capture - Finance | Dynamics 365 | Microsoft Learn

My observations so far:

Works well for service/cost invoices, as it mainly captures header‑level details.

Performs accurately on standard invoices containing clear fields such as invoice number, PO number, dates, vendor and item numbers, quantity, and amounts.

Also performs well for procurement category invoices.

Recently, AI traces started capturing invoice charges, and the model continues to improve based on user selections.

For complex invoice structures, some degree of human review is still needed.

Taxes and charges may still require manual validation.

There are promising enhancements in the future roadmap that will further improve performance and coverage.
We also experimented with a custom AI model, which requires substantial training effort. At least five invoices per vendor format are needed to train it effectively and achieve a reliable score, helping reduce manual intervention.

My suggestion:
Try it out first in a UAT environment—as you get 100 free invoices per tenant and based on this decision between standard AI model / custom AI model will be better taken. If the solution satisfactorily processes the majority of your simpler vendor invoices, you can start adopting it for those vendors and then gradually expand usage as new features roll out.

Regards

Devanshi shah

Was this reply helpful? Yes No