At enterprises across industries, documents are at the center of core business processes. Documents store a treasure trove of valuable information whether it’s a company’s invoices, HR documents, tax forms and much more. However, the unstructured nature of documents make them difficult to work with as a data source. We call this “dark data” or unstructured data that businesses collect, process and store but do not utilize for purposes such as analytics, monetization, etc. These documents in pdf or image formats, often trigger complex processes that have historically relied on fragmented technology and manual steps. With compute solutions on Google Cloud and Document AI, you can create seamless integrations and easy to use applications for your users. Document AI is a platform and a family of solutions that help businesses to transform documents into structured data backed by machine learning. In this blog post we’ll walk you through how to use Serverless technology to process documents with Cloud Functions, and with workflows of business processes orchestrating microservices, API calls, and functions, thanks to Workflows.

At Cloud Next 2021, we presented how to build easy AI-powered applications with Google Cloud. We introduced a sample application for handling incoming expense reports, analyzing expense receipts with Procurement Document AI, a DocAI solution for automating procurement data capture from forms including invoices, utility statements and more. Then organizing the logic of a report approval process with Workflows, and used Cloud Functions as glue to invoke the workflow, and do analysis of the parsed document.

We also open sourced the code on this Github repository, if you’re interested in learning more about this application.

In the above diagram, there are two user journeys: the employee submitting an expense report where multiple receipts are processed at once, and the manager validating or rejecting the expense report. 

First, the employee goes to the website, powered by Vue.js for the frontend progressive JavaScript framework and Shoelace for the library of web components. The website is hosted via Firebase Hosting. The frontend invokes an HTTP function that triggers the execution of our business workflow, defined using the Workflows YAML syntax. 

Workflows is able to handle long-running operations without any additional code required, in our case we are asynchronously processing a set receipt files. Here, the Document AI connector directly calls the batch processing endpoint for service. This API returns a long-running operation: if you poll the API, the operation state will be “RUNNING” until it has reached a “SUCCEEDED” or “FAILED” state. You would have to wait for its completion. However, Workflows’ connectors handle such long-running operations, without you having to poll the API multiple times till the state changes. Here’s how we call the batch processing operation of the Document AI connector:

  - invoke_document_ai:
    call: googleapis.documentai.v1.projects.locations.processors.batchProcess
    args:
        name: ${"projects/" + project + "/locations/eu/processors/" + processorId}
        location: "eu"
        body:
            inputDocuments:
                gcsPrefix:
                    gcsUriPrefix: ${bucket_input + report_id}
            documentOutputConfig:
                gcsOutputConfig: 
                    gcsUri: ${bucket_output + report_id}
            skipHumanReview: true
    result: document_ai_response

Machine learning uses state of the art Vision and Natural Language Processing models to intelligently extract schematized data from documents with Document AI. As a developer, you don’t have to figure out how to fine tune or reframe the receipt pictures, or how to find the relevant field and information in the receipt. It’s Document AI’s job to help you here: it will return a JSON document whose fields are: line_item, currency, supplier_name, total_amount, etc. Document AI is capable of understanding standardized papers and forms, including invoices, lending documents, pay slips, driver licenses, and more.

A cloud function retrieves all the relevant fields of the receipts, and makes its own tallies, before submitting the expense report for approval to the manager. Another useful feature of Workflows is put to good use: Callbacks, that we introduced last year. In the workflow definition we create a callback endpoint, and the workflow execution will wait for the callback to be called to continue its flow, thanks to those two instructions:

  - create_callback:
    call: events.create_callback_endpoint
    args:
        http_callback_method: "POST"
    result: callback_details
...
- await_callback:
    try:
        call: events.await_callback
        args:
            callback: ${callback_details}
            timeout: 3600
        result: callback_request
    except:
        as: e
        steps:
            - update_status_to_error:
              ...

In this example application, we combined the intelligent capabilities of Document AI to transform complex image documents into usable structured data, with Cloud Functions for data transformation, process triggering, and callback handling logic, and Workflows enabled us to orchestrate the underlying business process and its service call logic.

Going further 

If you’re looking to make sense of your documents, turning dark data into structured information, be sure to check out what Document AI offers. You can also get your hands on a codelab to get started quickly, in which you’ll get a chance at processing handwritten forms. If you want to explore Workflows, quickstarts are available to guide you through your first steps, and likewise, another codelab explores the basics of Workflows. As mentioned earlier, for a concrete example, the source code of our smart expense application is available on Github. Don’t hesitate to reach out to us at @glaforge and @asrivas_dev to discuss smart scalable apps with us.

Editor’s note: In this post, I’ll be showing some amazing ways Document AI can help you extract meaning from your documents  – keep reading, or jump directly into a tutorial using the Cloud Console!

Documents are a crucial part of most businesses, used to store and communicate important information. The variety is vast: invoices, contracts, receipts, applications, plus documents unique within industries and geographies. Unfortunately, making the information contained in these documents accessible can be a time-consuming and manual process.

Document AI is a document understanding platform in Google Cloud that takes unstructured data from documents and transforms it into structured data, making them easier to understand, analyze, and consume. By using this technology, you can streamline your document processing workflows, reduce errors, and unlock insights that were previously buried in mountains of paperwork. 

Whether you’re a small business owner or an enterprise looking to bring efficiency to your operations, Document AI has something to offer. So let’s take a look and see what it can do!

Understanding documents with Document AI

When we say that Document AI can understand documents, we mean that it is able to analyze the content within documents and derive meaningful insights from it. This goes beyond simply recognizing the characters and words within a document (which is what traditional OCR technology does) – Document AI can actually comprehend the meaning behind the text.

For example, let’s say you have a contract that needs to be processed. Traditional OCR technology might be able to extract the text from the document, but it would not be able to understand the legal terms and clauses within it. Document AI, on the other hand, can actually interpret the meaning of the text and extract key information such as parties involved, terms and conditions, dates, and signatures.

Document AI offers several pre-built models and processors that are specifically designed to extract different types of data from various document types. Within the specific document types, the processors can perform several tasks such as Optical Character Recognition (OCR), form parsing, splitting, classification or entity extraction. These processors can be customized and combined to create powerful document processing workflows that are tailored to a business’s unique needs.

Let’s look closer at a few of the processors available in Document AI, including the Form Parser, Invoice Parser, Expense Parser, Identity Document Proofing Parser, and Intelligent Document Quality Processor.

Form Parser

This general processor is designed to extract structured data from forms such as application forms, surveys, and questionnaires. It automatically identifies and extracts data from form fields (key-value pairs), such as names, addresses, dates, and other types of structured data; even checkboxes and tables. This processor also leverages deep learning models to extract generic entities that are common in various document types, meaning it can identify if something is an email address, phone number, datetime, organization, quantity, price, person, and more.

In this visualization of the API response, you can see that Document AI has identified several key value pairs that correspond to the form’s fields and the responder’s answers.

Also of interest is that the form parser recognized certain generic entities including: several dates, an address, phone numbers, email, and two people (the responder and their listed emergency contact).

Invoice Parser

This parser is designed to identify and extract relevant information from invoices including a large number of typical invoice fields, but can also be customized (uptrained) to recognize different invoice layouts, languages, and data fields.  Invoices are a critical part of the accounts payable process, making this functionality valuable across industries and companies building A/P features into their products.

In this visualization of the API response, you can see that Document AI has extracted a large number of key-value pairs and even provided normalized values for several of the fields.

Expense Parser

This specialized processor is designed to extract data from receipts and invoices, such as vendor name, date, and total amount paid. It can also identify line items within an invoice and categorize them based on the type of expense (e.g. meals, travel, office supplies). The expense parser makes it easier for you to process expense reports and other financial documents, and it can integrate with other tools and systems to completely automate the entire expense reporting process.

In this visualization of the API response, you can see that Document AI has extracted the text from the receipt and identified several typical entities such as purchase date and time, payment type, and total amount.

Identity Document Proofing Parser

This processor is designed to help predict the validity of ID documents with four different signals.

1. is_identity_document detection: Predicts whether an image contains a recognized identity document.
2. suspicious_words detection: Predicts whether words are present that aren’t typical on IDs.
3. image_manipulation detection: Predicts whether the image was altered or tampered via an image editing tool.
4. online_duplicate detection: Predicts whether the image can be found online.

If suspicious words are detected or the image can be found online, additional information is provided to explain these signals. 

This can be particularly useful for businesses that need to verify the identity of customers or employees as part of their operations. This processor could be used in conjunction with other processors that extract key information such as name, date of birth, ID number, and expiration date from specific identity documents (US Driver License Parser, US Passport, France National ID Parser, etc.).

In this visualization of the API response, you can see that Document AI has passed the document on the first detection point (is_identity_document), but failed the document for the other three items and provided additional information in the evidence fields.

Intelligent Document Quality Processor

This general use processor is designed to detect a variety of document quality issues, such as missing pages, blurry images, low contrast, inconsistent formatting, and incorrect data, and flag these potential issues which could affect their usability, accuracy, or compliance. It can also identify sensitive information that has not been redacted or missing information required by regulatory standards.

This quality assessment is returned as a quality score from 0 to 1, where 1 means perfect quality. If the quality score detected is lower than 0.5, a list of negative quality reasons (sorted by the likelihood) is also returned.

In this visualization of the API response, you can see that Document AI has determined a quality_score of 0.006 and also provided a list of several reasons including document and text cutoff, blurriness, and glare (among others).

Next steps

These are just a few examples of the types of processors that Document AI offers. Perusing the current documentation, you can find it has more than 40 processors, with each providing several functions, to explore. Document AI also offers you the ability to uptrain certain processors and supports the option to build your own custom processor. Across the board, if a document contains structured or unstructured text, Document AI has the capability to extract valuable data from it. 

Get started and learn more by heading to our tutorials in the Cloud Console:

• Get started with Document AI in the Google Cloud console
• Build a document processing pipeline with Workflows
• Uptrain a specialized processor
• Create a Custom Document Extractor in the Google Cloud console
• Create a Custom Document Classifier in the Google Cloud console
• Get started with Human-in-the-Loop

With unemployment application submissions reaching record numbers over the past year, state and local agencies in the United States have faced the challenge of processing unprecedented numbers of claims per week. The digital infrastructure most agencies have in place is unable to handle this volume, resulting in constituents waiting longer, and bad actors taking advantage of vulnerable systems. The Department of Labor Inspector General estimates that $63 billion in claims distributed is either an improper payment or fraud. 

Validating claims also requires secure data sharing with other agencies for document and identity verification. Government leaders need a way to allow case adjudicators to quickly and confidently release backlogged claims, integrate with existing systems, and segment legitimate claims from potentially fraudulent ones — all within limited government budgets — securely and at scale.

Implementing a fraud detection solution on Google Cloud

States were under pressure to release payments, while also filtering out potentially fraudulent claims. SpringML and Google Cloud developed a framework to give adjudicators a reliable verification process that quickly filters potentially fraudulent claims, while processing the remaining claims so benefits reach citizens in a timely manner. SpringML and Google Cloud, applied AI-powered machine learning models to detect anomalous patterns in large datasets. Using Google Cloud tools, SpringML implemented a solution to streamline workflows, improve efficiencies, automate processes and identify potentially fraudulent claims.

SpringML used a variety of Google Cloud products to deliver a fraud detection solution, including:

• Google Cloud Storage to store and manage data
• BigQuery to store tabular data and BigQuery Machine Learning (BQML) to conduct machine learning on that data
• AutoML solutions to build predictive models and risk scoring
• Visualization tools such as Looker and Data Studio to present data and help government leaders make informed decisions.

Implementing machine learning to detect improper payments allows agencies to classify claims as “fraud” or “not fraud” based on the number of flags, as well as prioritize the most urgent claims. Deploying intelligent virtual agents to handle frequently asked questions meant that live agents could focus their time on more challenging cases. 

Even once the pandemic is behind us, there will be bad actors trying to take advantage of overwhelmed or legacy systems. We’ve identified a few best practices for agencies managing enormous case loads and looking to improve improper payment analytics: 

• Move your systems to the cloud. Many on-premises legacy systems can’t update their applications and scale to meet the volume of claims. Moving to a cloud environment enables rapid solution deployment and ingestion of large amounts of data without fear of overloading the system. The cloud scales with you–cost-effectively and securely. 
• Understand patterns in the data. The answer is always in the data — we used deep analysis to help uncover suspicious patterns in large data sets. We implemented unsupervised machine learning to learn behaviors and create configurable rules that adjust to new information that comes into the system. We can uncover patterns that are likely associated with fraud – ones that a human might have missed. 
• Use AI/ML tools to automate your existing systems and teams. These tools enable humans to work smarter and more efficiently. We automate anomaly detection and create dashboards for adjudicators to rapidly process claims. We are enabling the Wisconsin Department of Workforce Development by implementing automatic calculations and processing of recharge amounts, resulting in faster processing times and fewer human errors. Proactive fraud detection and timely calculation of recharge payment allowed DWD to ensure the benefits reached the right individuals.
• Build flexibility into your systems. We discovered that fraud patterns change over time. For instance,flags for fraud during March-May 2020 were vastly different from those we found in June-July 2020. Google Cloud tools make it easy to continually update algorithms to detect patterns and integrate external data sources.

Using Google Cloud tools, we can update digital infrastructure and incorporate machine learning best practices to help organizations efficiently process large volumes of claims and identify high probability fraudulent ones. SpringML provides consulting and implementation services and industry-specific analytics solutions that deliver high-impact business value to accelerate data-driven digital transformation. Learn more about fraud detection and how to improve improper payments analytics by watching our webinar. 

Business Problem
UKG Ready primarily operates in the Small and Medium Business (SMB) space, so inherently many customers are forced to operate and make key business decisions with less Workforce Management (WFM) / Human Capital Management (HCM) data. In addition to volume, SMB lacks the variety of data needed to create a dynamic and agile organization. This puts SMB at a major disadvantage compared to larger segments.

Project Goals
People Insights module is committed to surfacing insights to customers in the context of their day-to-day duties and aid in decision making. With the SMB customer data limitations mentioned above, the goal of this project was to create a global dataset that augments individual customer data to bring light to less obvious, yet important information.

Challenges
UKG Ready is a highly configurable application that gives customers the opportunity to build solutions on a platform that meets their specific business needs. High configurability gives high flexibility to customers in their usage of the software. However, it becomes nearly impossible to create a global dataset for machine learning and data insights. UKG Ready manages just under 4 million of the US workforce and some 30,000+ customers. Despite the large employee dataset size, machine learning models that are specific to customers are starved for data because the individual customers have a relatively small employee population. Does that mean we cannot support our SMB customers’ decision making with ML?

Result
Partnering with Google, we were able to develop an approach that allowed us to standardize various domain entities (pay categories, time off codes, job titles, etc.) so that we could build a global dataset to augment SMB customer data. Using machine learning we were able to build a common vocabulary across our customer base. This common vocabulary encapsulates the nuances of how our customers manage their business and yet is generalized and standardized such that the data can be aggregated over the variety of customer configurations. This allows us to serve up practical insights to customers through various use cases. Our partnership allowed us to leverage Google Cloud Services to meet the needs of our complex machine learning models, distributed data sets and CI/CD processes.

How
UKG Ready decided to partner with Google for an end-to-end solution for the analytics offering. This allowed us to focus on our core business logic without having to worry about the platform, environment configurations, performance and scalability of the entire solution. We make use of various Google Cloud services such as Cloud Triggers, Cloud Storage, Cloud Functions, Cloud Composer, Cloud Dataflow, Big Query, Vertex AI, Cloud Pub/Sub… to host our analytics solution. Jenkins manages the entire CI/CD pipelines and cloud environments are configured and deployed using Terraform.

The standardization of business entities problem was solved in three distinct steps:

Step 1: Collecting aggregated data
We needed an approach to collect aggregated data from our highly distributed, sharded, multi-tenant data sources. We developed a custom solution that allows us to extract data aggregated at source for PII and GDPR considerations and transfer to Google Cloud Storage in the fastest manner possible. Data is then transformed and stored in Big Query. Services used: GCS, Cloud Functions, DataFlow, Cloud Composer and Big Query. All processes are orchestrated using Cloud Composer and detailed logging is available in Cloud Logging (Stackdriver).

Step 2: Applying NLP (Natural Language Processing)
Once we had the variety of customer configurations or the business entities available, we then applied NLP algorithms to categorize and standardize these in buckets. This approach assumes that customers use natural language for configurations like job titles, pay codes etc.

String Preparation
The input data for string preparation process is an entity string or several strings, that describe one entity object (like name-description pair or code-name pair). The output represents set of tokens that may be used to run a classification/clustering model. The process of string preparation tokenizes strings, replaces shortcuts, handles abbreviations, translates tokens, handles grammatical errors and mistypes

ML Models

Statistical
The idea of the model is to use defined target classes (clusters) and assign several tokens (anchors) to each of them an entity that has any of those tokens would be “attracted” to appropriate class. All other tokens are weighted according to frequencies of usage of theses tokens in the entities with anchor tokens:

Using anchor tokens, we are building kind-of Word2Vec - dimensionality of vector is equal to number of target classes. The higher the specific dimension (cluster) value, the higher the probability of entity to be included in appropriate cluster. Final prediction for entity tokens list for specific class is sum of weights of all the tokens included. Predicted cluster is a cluster that has maximal prediction score.

Lexical Model
We managed to generate reasonable amount of labeled data during statistical model implementation and testing. That opens a possibility to build “classical” NLP model that uses labeled data to train classification neural network using pretrained layers to produce token embeddings or even string embeddings. We started experimentation with pre-trained models like GloVe and got good results with single words and bi-grams but started getting issues in handling of n-grams. Our Google account team came to our rescue and recommended some white papers that helped formulate our strategy. We now use Tensorflow nnlm-en-dim128 model to produce string embeddings – it was trained on 200B records English Google News corpus and produces for each input string 128-dimensional vector. After that we use several Dense and Dropout layers to build a classification model.

Ensembling
To perform ensembling all the model results for each class are cast to probabilities using softmax transformation with scale normalization. Final predicted probability is maximal average score of both models among all the classes scores – appropriate class is predicted class.

The machine learning models are deployed on Vertex AI and are used in batch predictions. Model performance is captured at every prediction boundary and monitored for quality in production.

Step 3: Making available common vocabulary
Having the standardized vocabulary, we then needed a mechanism to have the results be available in UKG Ready reports and customer specific models like Flight Risk and Fatigue. For this we again used Google Services for orchestration, data transformation and data storage.

Once the modeling is complete, we made the customer specific models leveraging the above architecture be available in Reports. We utilized our proven existing technology choices in GCP for orchestration, data transformation and data storage

Results
We are able to build a common vocabulary of our customers’ business entities with good confidence. And be an expert advisor to our SMB customers in their decision-making using machine learning. With the advice of our Google account team and using Google services we can add value to our product in a relatively short amount of time. And we are not done! We continue to use this platform for new use cases, complex business problems and innovative machine learning solutions.

Sample result:

Special thanks to Kanchana Patlolla , AI Specialist, Google for the collaboration in bringing this to light