Intelligent Document Processing ideal and non ideal project scenarios

Few limitations I can think of:

1. Appian IDP supports only PDF docs as inputs as of now, if there is any doc of any other type, we might want to convert that manually or use some smart services to convert them ourselves.
2. AI may not be 100 % accurate always, so if it our app is data sensitive/critical, then we should always route it to reconciliation step or manual data fix updates post extraction phase.
3. Process workflow cannot be synchronous, as in we can't expect to view a form or see any immediate results once we submit the document for processing initially, this needs to be later viewed as a task or in a record later post processing of extraction phase which usually takes around 1-2 minutes on an average.
4. Google AI/ML is comparatively better than Appian AI, for data extraction/parsing. I recommend using Google AI/ML for more accuracy/confidence.
5. I believe the ML training is not carry forwarded while we deploy the app to higher environments, and we have to re-train it again from scratch, which adds and extra overhead. Someone please correct me if I'm wrong here.

Few limitations I can think of:

1. Appian IDP supports only PDF docs as inputs as of now, if there is any doc of any other type, we might want to convert that manually or use some smart services to convert them ourselves.
2. AI may not be 100 % accurate always, so if it our app is data sensitive/critical, then we should always route it to reconciliation step or manual data fix updates post extraction phase.
3. Process workflow cannot be synchronous, as in we can't expect to view a form or see any immediate results once we submit the document for processing initially, this needs to be later viewed as a task or in a record later post processing of extraction phase which usually takes around 1-2 minutes on an average.
4. Google AI/ML is comparatively better than Appian AI, for data extraction/parsing. I recommend using Google AI/ML for more accuracy/confidence.
5. I believe the ML training is not carry forwarded while we deploy the app to higher environments, and we have to re-train it again from scratch, which adds and extra overhead. Someone please correct me if I'm wrong here.

Thanks Vinay it helps. I also went through the courses on academy and some limitations mentioned are

• .It only support PDF up to 15 pages or no more than 7 MB document
• It does not work well if document has large text area or footers
• And not ideal for highly unstructured documents where it is difficult to find name/value pairs