Simplifying Document and Email Automation Using Artificial Intelligence

Putting the power of AI into the hands of business users and RPA developers

Challenges in time-consuming and expensive machine learning (ML) model building and a lack of talent with ML experience have restricted the potential of AI in the enterprise.

That’s the feedback we’ve received from customers as we continue to check in to find out how we can help them get more value from AI. In particular, to help them process data from documents and emails as part of broader end-to-end automations—quite a common request across different domains.

In response, we’ve created new simple yet powerful AI capabilities for document and email automation. Available out-of-the-box and easily customized, they empower more business users with AI tools.

New AI capabilities

UiPath 2021.10 release delivers new AI capabilities—Forms AI, Email AI, and new pre-trained ML models for document processing. These functionalities will help companies easily deploy AI-enabled automations, saving them a lot of time and effort. Let’s take a quick look at each new capability.

Forms AI: no code intelligent document processing

As part of UiPath Document Understanding, Forms AI is used to process forms such as tax forms, insurance forms, pay stubs, and other structured documents. It provides a point-and-click experience that does not require any coding or ML skills, making it accessible to citizen developers and business users.

The workflow for using Forms AI is straightforward. A user uploads two to five document samples, chooses the fields for extraction in one document, and then gets an automatically generated model that can help robots read and understand these type of documents. Once tested out on the rest of the samples, this becomes a working AI solution, and your center of excellence (CoE) team can use it in the end-to-end document processing automation. See how it works in the demo below:

This easy, low-touch approach to AI eliminates the need to create and maintain static document templates or expensive ML models for documents with similar formats. Even with little or no familiarity with ML tools and technology, you can easily leverage Forms AI to automatically generate lightweight models in just minutes.

And it gets better because these models continuously learn and become smarter over time. There is no need to choose between using templates with rules-based approaches and using ML for document processing. Forms AI combines the strengths of both approaches while removing the drawbacks of either that might prevent broader user adoption.

One of the largest accounting firms servicing real estate clients was looking for a solution to process lease documents. They were on a very tight timeline and heavy ML building was not an option due to the cost and time investments required for the custom ML model building.

The customer was considering investing two to three weeks in collecting and preparing data using their own resources and then creating an ML model—translating into a few weeks of development effort. Forms AI created a model for their lease documents in five minutes and is delivering over 95% accuracy.

Email AI: an AI solution template for smart email processing

UiPath AI Center helps enterprises add AI to their automation workflows easily. With more than 25 pre-built ML models, multiple deployment options, a drag-and-drop interface, and many more foundational MLOps features, AI Center empowers users to tackle a wide range of use cases, even if users are not data scientists.

To further simplify users’ journey to AI-enhanced automation, AI Center now includes the Email AI solution template, which provides recipes for creating AI solutions to automate email processing.

This new capability is aimed at a common business challenge—managing the sheer volume of emails that arrive in inboxes every day. Research from Statista reports that 306 billion emails were sent globally each day in 2020. Many UiPath customers process thousands, millions, or even up to a billion emails every year:

Many enterprises have a dedicated team to monitor and process emails sent by their customers, partners, suppliers, vendors, and more. The processing is costly, laborious, error prone, and time consuming. According to research of 1,000 companies, the average response time to customer emails is 12 hours and 10 minutes. However, customers expect companies to respond within one to four hours.

Email AI helps alleviate some of those issues. It provides ready-to-use components for users to build AI solutions to easily process emails in bulk. Components include ML models, sample datasets, RPA workflows, analytics templates, and human-in-the-loop technology.

With Email AI solution templates, AI-enhanced robots can automate email processing intelligently. They can detect intent in emails, identify urgent and high-value emails, extract important information, and help you understand the impact of AI-enhanced automations.

A few of the use cases from early Email AI adopters include:

• A global Fortune 500 company is using Email AI to automate data entry into relevant systems and prioritize emails sent by their business customers. The bank plans to automate about 75% of one billion customer emails.

• A large investment management organization is using Email AI to automatically process equity orders placed by their customers via email.

• A financial services company is using Email AI to extract income data from emails sent by their business customers.

A truly integrated platform becomes more and more visible. A lot of our clients are going to be excited to see capabilities like Email AI embedded in the UiPath Platform because they don’t have to go to a third-party solution vendor.”

Donald Sweeney, Co-Founder, Ashling Partners

Pre-trained ML models for document processing

At the core of intelligent document processing are pre-trained ML models that can be retrained based on custom documents. UiPath now provides a set of pre-trained ML models so you can easily start processing documents like invoices, receipts, purchase orders, utility bills, identification (ID) cards, passports, contracts, and more.

Out-of-the-box solutions like this make AI accessible to automation specialists with no experience in data science, while allowing them to retrain the models to improve their accuracy over time.

With the 2021.10 release, we are offering more ML models via UiPath Document Understanding. The new ML models help customers automate the business processes that require processing the following types of documents:

• W-2 forms report the taxes withheld from a United States (U.S.) employee’s wage/salary for the year. UiPath can help extract the employer identification number (EIN), social security number (SSN), employee name and address, employer name and address, wages and tips amounts, withheld tax amounts, and others.

• W-9 forms, the Request for Taxpayer Identification Number and Certification, which is a commonly used Internal Revenue Service (IRS) form for business owners and independent contractors in the U.S. The pre-trained ML model can extract all the key fields, including business name, address, account numbers, SSN, EIN, signature, and date.

• Remittance advice is a document sent by a customer to a supplier as a proof of invoice payment. Our Remittance Advice ML model can get data from the document fields including vendor name, invoice number and date, vendor name and address, currency, and others.

• Delivery notes are sent with shipments of goods, listing the goods being delivered. The ML model can capture fields such as purchase order number, sender name, date, product codes, quantity, weight, measurement, receiver name and address, shipping address, carrier name, and more.

• Invoices for China is a ML model built for local invoice documents. It can extract information about the seller and purchaser, invoice number and date, data about the commodities, amounts, taxes, and other fields.

What’s especially valuable is that these models are all retrainable. That means a business can use their own custom documents to retrain the model.

The power of retraining is strengthened by human-in-the-loop capabilities. Basically, every time a human validates some exceptions or makes corrections to the output, this validated data is saved for further model retraining. In this way, robots learn and improve their document processing skills over time.

Get a preview as a UiPath Insider

These powerful additions to UiPath AI capabilities are now available in preview. We urge customers to give them a try to better understand the benefits of having easy-to-use, scalable AI functionality applied to common enterprise tasks.

To access them, sign up for the UiPath Insider program.

This blog post was co-authored by Gwen Chen, an AI Product Marketing Manager at UiPath.