Companies often process documents manually. Manually extracting data is time-consuming and requires a lot of manual effort. With the invention of OCR, the conversion from physical to digital happens quickly and with greater speed. But OCR works on rules and templates, which requires an expensive setup and more human effort. With a combination of OCR and AI, the companies can adapt rapidly changing conditions and customize its process accordingly.
Google’s Document AI provides OCR (optical character recognition) capabilities by leveraging advanced deep-learning neural network algorithms. But while dealing with complex data, we often need human review to check the consistency and compliance of the extracted data. Google Human In The Loop Processors provides workflow and Labeler workbench to review, validate and correct the given data.
For more information, refer to link.
Human in the Loop Reference Workflow
The following diagram provides a high-level reference workflow built on HITL and human review is only needed when a document’s confidence level falls below a certain threshold.
- Receives document processing request from the client
- Documents stored in the GCS bucket
- HITL invoker service invokes the Parser API
- High confidence predictions will be sent straight-through processing
- In the case of any inaccuracies, low confidence scores, or exceptions, the specialist receives a notification and opens Validation UI to validate data and handle exceptions in the Labeler workbench as shown below in the figure. Here you can also review all the extracted fields if needed to make sure the data is processed accurately. The below Figure depicts the same.
- After the specialist correction results are stored in GCS bucket
- API is provided to Client applications to fetch the results
Availability
Human in the Loop Platform is in Private GA, along with the below processors.
General Processors
- Form Parser
Procurement Processors
- Expense Parser
- Invoice Parser
- Utility Parser
Lending Processors
- 1003 Parser
- 1040 Parser
- 1040 Schedule C Parser
- 1040 Schedule E Parser
- 1099-DIV Parser
- 1099-G Parser
- 1099-INT Parser
- 1099-MISC Parser
- 1120S Parser
- Bank Statement Parser
- HOA Statement Parser
- Mortgage Statement Parser
- Pay Slip Parser
- Retirement/Investment Statement Parser
- SSA-89 Parser
- W2 Parser
- W9 Parser
- Contract Processors
- Contract parser
Benefits
- Document AI/HITL solutions can be deployed in weeks versus months or years
- The Human in the Loop processors can be integrated with existing solutions with API
- With Human in the Loop processors no need to train AI models
- With Human in the Loop, companies can eliminate large teams, and hire, train, and oversee the quality of human labor. Only a few Specialist pool of experts are required to validate the documents
- Monitor the productive of workflow with a predefined workflow
- Ensure the quality of data for downstream business applications
Demo
The below video shows how to perform HITL that includes the following capabilities:
- Configure human review for a processor
- Create a human review user resource pool
- Create a test human review task
- Assign a human review task to a user
- Complete a human review of a document
Enterprises should prepare themselves for future successes without having to rethink twice about their security and safety, SpringML which is a trusted Google Cloud Partner becomes an obvious choice. Our dedicated team of certified experts with a customer-centric approach chalk out differentiated services with ease to assess and integrate the solutions seamlessly. With esteemed partnerships, SpringML enables quick delivery and access to a wide range of various technologies for consumption that help run the business with minimum operational hiccups. We offer management services to our customers and help unlock vital insights. The experts at SpringML have rich proficiency expertise in providing data-driven innovative solutions to companies around the world. The clientele is widespread and includes several enterprises from the Fortune 500 quadrant.
