In this Developer Meetup, Arjun Meda reviewed and demonstrated the new Document Automation + Generative AI capability, while Micah Smith and Max Cassidy offered additional insights and answered live audience questions.

You can watch the full recording of the meeting below.

Let’s jump right into a recap of the power of Document Automation + Generative AI and how you can use it!

What is Document Automation?

Document Automation is part of an end-to-end solution fully integrated into the Automation Success Platform, which means no separate installations are required, and it works seamlessly across Automation Anywhere products.

But what makes Document Automation + Generative AI different? Document Automation + Generative AI is an essential tool for processing any form of unstructured document, such as PDFs or images. Whether you have one document or hundreds of documents, Document Automation + Generative AI can ingest them all at once, then classify, split, extract, and validate the non-tabular, unstructured data contained within your documents using generative AI—a capability that has never been harnessed before!

Unstructured Document Use Case Summary

Let’s review the real-world use case Arjun demonstrated to extract data from a Risk Assessment Report in an unstructured format. The LLM used in Document Automation is Azure Open AI, and the OCR used in this demo is Google Vision.

1. Start by logging into Control Room and, under ‘Manage’ on the left side, select ‘Create a Learning Instance.’ Name your learning instance and select your document type from the list. For this demo, Arjun showcased the new ‘Unstructured Document’ option available in Document Automation + Generative AI.
2. It will show the language*, locals, and provider, which is Automation Anywhere, as well as OCR provider, which is Google Vision. Then click ‘Next.’ *Currently, Document Automation + Generative AI is only available for English-language documents.
3. For this demo, Arjun is utilizing a two-page Final Risk Assessment Report. From this document, we want to extract the property address, report date, overall risk score, and any risks that are rated medium or high. To do this, click on ‘Add a Field’ on the right side and you’ll see field properties open up. This is where you define what you want to extract.
4. First, we want to extract the property address, so write “property address” and copy and paste that to ‘Field Label.’ You can select whether this is a required or optional field and specify the data type, such as text, number, date, or address. Because we’re using generative AI, there’s no specific format or keywords to use; it’s all about how well your prompts are written. Once you see the response returned, you can adjust your prompt query. For example, for the purposes of extracting the property address, we can specify “include the city, state, country, and zip code” in the prompt if necessary. The same goes for a date—you can specify “give a date in MM/DD/YYYY.”
5. Next, we’ll add the rest of the fields we want to extract: report date, overall risk score, and risks that do not include low risk. Again, you can always come back and tweak your queries if the responses returned aren’t correct.
6. Now, we’ve created our learning instance. Click ‘Process’ and upload the document, specify the folder where you want the output to download, and click ‘Process Document.’
7. Once the extraction is complete, you will see the validation screen (if any field validation fails). On the right side, you can see anything it was not able to identify. In this case, it could not identify if a certain field met higher risks, so we need to adjust our prompt. To do this, go back to the learning instance and update the prompt to say, “Are there any medium or high risk findings? Don’t include low risk findings.” Then, reprocess the document.
8. Now, the process has executed successfully and saved to our defined folder. We can open the CSV file and see it extracted the property address, report date, overall risk score, and any medium or high risks.

Who Can Access Document Automation + Generative AI?

Document Automation + Generative AI is only available in the Enterprise Edition and requires an Automation Co-Pilot license (formerly AARI). However, our traditional Document Automation tool is available in the Community Edition, but it cannot process data within unstructured documents with generative AI. Please contact your Account Executive for pricing details or to arrange a POC.

What Features Are Included?

Out of the box, Document Automation + Generative AI supports document types like invoices, arrival notices, bills of lading, mapping list receipts, standard forms, and unstructured documents. There’s no model training required. Each document type simply requires that you at least specify the fields you want extracted, as demonstrated in the use case above. Keep your eyes peeled for a Document Automation package update coming out between .31 and .32 to help with advanced table extraction using generative AI.

Generative AI is not limited to document processing and extraction—it is integrated in the entire Automation Success Platform! There’s a Gen AI package in the taskbot builder that enables you to invoke most of the major LLMs from Google, Microsoft, OpenAI, AWS, Anthropic, etc. With generative AI integration, the possibilities for what you can accomplish are virtually limitless!

Security Considerations

We never use customer data for model training. We put security measures in place that prevent customer data from being stored externally and provide guardrails, redaction, or masking in certain products. The Automation Success Platform ensures that customer data is always protected using industry-standard encryption for data at rest and in transit. The systems storing customer data are monitored 24x7 and access-controlled to ensure safe operations in compliance with SOC 2, ISO 27001, and HITRUST.

For full details on Automation Anywhere’s policies and practices regarding data security for generative AI, read our security documentation and disclosures here.

Language Support with Document Automation + Generative AI

Currently, only English language forms are supported for unstructured documents in Document Automation + Generative AI. For the latest updates on the languages supported for various document types in Document Automation, refer to this page.