We have been working with a client from the States, where we are supposed extract information from 100 page SPD pdf documents (from around 900 PDF documents).
Below is the AI Builder license calculation for the Document processing custom model as per our understanding:
AI builder add-on plans:
Considering the above calculations, roughly we need 10+ packs of Add-on to cover our scope which is 900 SPDs, and this will roughly cost around 4000$ - 6000$, considering some testing iterarions etc.
I wanted to know if my calculations and understanding is correct?
You are right, 90,000 x 100 = 9,000,000 = 9M credits, so 9 units, sorry!
Each model consumes 100 credits per processed page. So if within a flow run your document is processed 4 times by different models, then your flow will consume 400 credits per page during a run.
You can also consider to build a single model which supports multiple layouts.
If this feature fits your need, then you will use a single model so only consume 100 credits per page per flow run.
@Antoine2F Thanks for the reply. Yes, my calculation was wrong. However, the required unit is only 9, and not 90.
Here is the updated calculation:
I also have another question, If we create multiple document processing models and use it for 1 document extraction, and use all of those models in 1 Power Automate flow, the flow runs for 1 document uploaded to a Library, each model counts the credits or only 1 document's total pages are counted?
I forgot to ask an important question about the timeline.
The above calculation is to perform the analysis of these 900 SPD documents every month.
If you can spread the analysis over 10 month, and analyze 90 SPD documents every month, then you only need 1/10th of the capacity, and 10 units (=$3750) will be enough (plus buffer if needed).
Hello,
Your understanding is correct, but there is a mistake in your calculation:
900 documents of each 100 pages makes 90,000 pages (not 9,000 pages), so 90M credits.
It means 90 units so T3 so 22,500$.
You can then add some buffer.
Antoine2F