Inconsistent AI Builder Table results

(0) Share

Report

Posted on by TYL

1,091 Super User 2024 Season 2

Hi all,

Been having a lot of fun with AI Builder recently. However, I found something very weird which I hope to find a solution for.

I have a Power Automate flow that processes invoices with AI Builder. When feeding the flow with one document (let's say document A), the AI Builder mode extracts tables perfectly. However, when I feed the flow with multiple documents (including document A) , the extracted tables from document A suddenly contain mistakes. When quick testing in the AI Model overview, the results are also fine. It's only different when processing multiple documents via a Power Automate flow (for each).

The documents are all single-page.

Have someone else encountered this problem? If yes, what would be the fix?

Thanks in advance!

Categories:

AI Builder

I have the same question (0)

All responses (3)

Answers (0)

Sort by

Antrod Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Hi @Tjan ,

Can you share screenshots of your flow so we understand how you extract and store the data?
Thanks!

Was this reply helpful? Yes No
TYL 1,091 Super User 2024 Season 2 on at

Like (0)

Report
Copy link

Link copied!

Hi @Antrod,

Thanks for your swift reaction. My flow is in Dutch so I don't think it will show u what u need. However I'll try to explain:
1. My flow fetches a sharepoint list containing documents (invoices)
2. Each document is processed by a 'for each' component
3. Because each document has multiple pages, I use the 'page rage' function in the AI Model component. Hereby I use the function: @{add(iterationIndexes('Do_until'), 1)}. By doing so, it takes one page at a time, thus feeding the AI model page by page.
4. The extracted data (table rows) is written to a SQL database.

Additional information: I looked back at the run history and it seems like the AI model is recognising the wrong collection (only when feeding multiple documents). When I feed only a single document, it seems like it's recognising the collection it belongs to perfectly well.

Was this reply helpful? Yes No
Antrod Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Thanks for the details!
That's weird, model should behave the same way no matter if there are multiple documents to process or not. For the documents you suspect the collection predicted is not correct, can you try doing a Quik test in the model detail page in AI Builder?

If values detected are correct, there's perhaps an issue with the action sequence in your flow. Try perhaps redoing a very simple flow that just gets processes all the documents in your SharePoint and just look at the run history to see if everything seems fine.

Was this reply helpful? Yes No