Extract Information from Invoice assumes date formats incorrectly

[jonillsley]

Hi all. 

I am trying to use the OCR step within a flow, when a document is created I run the extract information step to then add the information to a spreadsheet. I have read a lot about people struggling with date formatting but no one has an exact solution for this issue. 

The issue appears to be that when you extract the information, it assumes the received document is a US format. So if your date is 03/05/2023 it assumes this is March 3rd 2023. However, if the date is 13/05/2023 it assumes this is UK format 13th May 2023. What this means is that the date is being entered incorrectly. The problem here is that you can't reformat the dates here using a split and concat to re-arrange 03/05/2023 to 05/03/2023, because if you do then you rearrange 13/05/2023 to become 05/13/2023 which is invalid. 

Has anyone else experienced this issue and managed to resolve it / work around?

Thanks

[v-qiaqi-msft]

Hi @jonillsley,

Actually, Flow confuses with dd/MM format( Specially if month number belongs to date number just like your example) and always convert to standard format.

Please use yyyy-MM-dd format in flow, or you could use the Conver time zone as below:

Change the parameter in your scenario, like select the correct detination time zone and source time zone for your scenario.

[jonillsley]

Hi @v-qiaqi-msft , I think the issue here is though that the step which gives my time, is giving me different results. So, for example: 

Here, the invoice will display, UK Format 3rd May 2023 as 03/05/2023, however, the extract information step reads this sometimes as 5th March 2023, and sometimes 3rd March 2023. Since the system is not consistent, and I cannot tell it which to use, I am unable to use the convert function. When the date becomes 13th May 2023 for example, this will always assume UK format of 13/05/2023, but up until the 12th it is not consistent. 

[Anonymous]

Hey mate, I'm having the same issue and people don't seem to understand the AI invoice processor assumes the date format and from what I can find there is no way to force it to read those dates in a particular format. Doing conversations to the date afterwards or using con-cat doesn't work as the format to begin with changes back and forth. Super frustrating that our flow to extract invoice data can mix up the invoice date. 

[jonillsley]

Yes I am still having the issue too. When I get a date 13th or above in a month its fine. However, anything below that it gets mixed up. I've tried splitting it up, but then when you get say 4th May, I can't reliably get the result to be 04/05 or 05/04. It's whatever the AI feels like at the time. 

[Anonymous]

I've been using a custom model for a few weeks and I've just gone back in to add to my sample collection and can see that a date and format option has been added for new fields, including Day Month Year. Hopefully this solves the issue mostly. My flow still defaults back to the standard invoice processor when the confidence score is low on the custom model, indicating invoices that haven't been trained with the model. 

[Prasad_Shetty]

You can use below Expression to Convert date form Extract of default AI Models

To be updated as per your default model name:

1. "outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueText']"

2. "outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueDate']"

Expression:

if(and(equals(slice(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueText'], -4),slice(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueDate'],0, 4)),lessOrEquals(dayOfMonth(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueDate']),12)),formatDateTime(concat(slice(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueText'], -4), '-', slice(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueText'], 3, 5), '-', slice(outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueText'], 0, 2)),'yyyy-MM-dd'),outputs('Extract_information_from_invoices')?['body/responsev2/predictionOutput/result/fields/invoiceDate/valueDate'])