Unanswered

AI-builder - get extracted item on separate rows

(0) Share

Report

Posted on by reichkupong

I want to extract line item data from invoices and save these in a sharepoint-list. I've begun by setting up a primitive flow that does this, but I'm encountering some challenges.

For the supplier used for testing, their invoices consist of several pages. I'm trying to work around the problem of AI-Builder only scanning the first page (apparently it's not possible to tag multiple similar data fields across multiple pages) by splitting the PDF's and training AI-Builder on these separate pages.

What happens though, as illustrated in my attachment, is that the output doesn't come out line by line. I want Ref-01 on one row, Ref-02 on another etc. Also, since the pages are split, data is spread out over different rows. For example, the first cost on the second row belong to Ref-02 in the first post, but maybe I'll have to sort that out in PA.

(Another problem is that AI-Builder is not reading the content properly on the 3rd post but maybe that calso can be taken care of in PA).

Any advice would be appreciated!

p.s. The output in this case is the result of two invoices, indicated by the Event Date-field

Categories:

AI Builder

I have the same question (0)

All responses (7)

Answers (0)

Sort by

CedrickB Microsoft Employee on at

Like
a
(0)

Report
Copy link

Link copied!

Hi,

First of all extracting tables spaning accross pages should be working presuming you are using the "This table continues on next page" feature. See https://learn.microsoft.com/ai-builder/tag-form-processing-model#multipage-tables

Make sure to provide samples with 1, 2 and 3 pages (2 sample of each one)

Regarding empty rows for some cell.

You should record data in an object variable.

{"Col1":"MyValue1", "Col2":"MyValue2", ..., "Coln":"MyValueN"}

Inside the Apply to each, check if Col1 is empty,

- If so , just append Col3 to "Col3" property

- If not, your previous row is complete. You can consume it and create a new object for current row.

Was this reply helpful? Yes No
reichkupong 8 on at

Like
a
(0)

Report
Copy link

Link copied!

Hi Cedric, thank you for your reply!
I will try your advice concerning the empty rows.
Regarding the "This table continues on next page"-function. This only works when the desired fields are in an actual table right? On the invoices I'm using for testing, only some fields are inside a table.

Was this reply helpful? Yes No
reichkupong 8 on at

Like
a
(0)

Report
Copy link

Link copied!

I'm struggling with understanding how to proceed with the object variable. I'm not permitted to initialize a variable inside the Apply for each.

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

Like
a
(0)

Report
Copy link

Link copied!

Initialize the variable at the beginning of you flow.

Then use "Set variable" inside the "Apply to each".

And to your earlier question, yes "continue on next page" only works when the desired fields are in the table.

Was this reply helpful? Yes No
reichkupong 8 on at

Like
a
(0)

Report
Copy link

Link copied!

I see. Ok, I initialized the variable at the beginning.
However, I still get the first items lumped together (ref01 & ref02 for example from my sharepoint screenshot). I need to separate these.

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

Like
a
(0)

Report
Copy link

Link copied!

Ah ok, I didn't get this at the beginning.
In this case, you probably better to try this:

Condition split(Reference, ' ')[0] Equals Reference

THEN // There is only one information in the column

consume

Reference

Supplier

Event date

Unit status

Unit number

Cost

ID

ELSE

Apply to Each split(MYCOL, ' ')

consume

split(Reference, ' ')[loopid]

Supplier

Event date

split(Unit status, ' ')[loopid]

split(Unit number, ' ')[loopid]

split(Cost, ' ')[loopid]

ID

Was this reply helpful? Yes No
reichkupong 8 on at

Like
a
(0)

Report
Copy link

Link copied!

Thank you for all your support.
Unfortunately I don't have the know-how to translate your recommendations into PA-actions.

Was this reply helpful? Yes No