Introducing generative validation: Maximize efficiency with Generative AI

There’s no single tool that does every job well. AI is much bigger than GPTs, and by looking at the broader AI landscape, enterprises can find many tools to complement and maximize the value of their AI investments.

Specialized AI describes a broad range of AI models trained and ‘specialized’ on a specific, limited task. A Generative AI model like ChatGPT performs well across a wide number of different tasks and use cases, but Specialized AI models are more like a perfectionist for one task. That’s why it’s an ideal partner for ‘general-purpose’ generative models, adding accuracy, reliability, and strong controls to Gen AI’s power and flexibility.

A great example of this synthesis between Generative AI and Specialized AI is found in our market-leading intelligent document processing (IDP) solution. Combining UiPath Document Understanding and Communications Mining, UiPath IDP capabilities understand and process documents and communications (structured and unstructured) at speed and scale.

Our Specialized AI models accurately interpret documents and messages. The models work side by side with Generative AI models that accelerate model training and accelerate time to value in IDP.

Now, I'm happy to introduce our latest Gen AI capability in IDP...

Introducing generative validation

AI models make mistakes. That’s why data validation—manually reviewing model outputs for accuracy—is an important part of IDP.

When using UiPath Document Understanding it gives you a confidence score for each bit of information it pulls out. The higher the confidence score, the higher probability that Specialized AI has accurately interpreted the information in the document.

The UiPath Business Automation Platform makes it easy for businesses to set up business rules for straight-through processing and human validation. For example, a prediction with high confidence will zip on to the next automated steps. But if a prediction doesn’t meet the business’ confidence threshold, a human will review it before letting the robots do their thing. Or, in workflows where human review is a legal requirement, business rules can easily be set up to flag for a human check.

Even in workflows where human validation is a mandatory step, IDP still brings huge time savings of 80–90% based on average handling time (AHT).

But what if we could take AHT savings even further by adding Generative AI to the mix?

Because Generative AI can be used to extract data from documents, the UiPath AI research and development team started exploring how it can help validate data extracted by our Specialized AI models. Initially, we found that using Generative AI alone for document processing resulted in higher error rates than using our specialized Document Understanding models.

But we didn’t stop there. We decided to put generative extraction to the test in areas (or “fields”) where Specialized AI wasn’t too sure of itself (i.e. low confidence). Our hypothesis was that if both Specialized AI and Generative AI returned the same output, then there was a much higher chance that the field was extracted correctly.

As a result of this research, I’m happy to introduce the latest feature for UiPath Document Understanding: generative validation.

With generative validation enabled in the data extraction workflow, Generative AI comes in to double-check the Specialized AI model’s work whenever it's not feeling too confident. Now, if both the generative and specialized models end up with the same result, it's a safe bet that we can skip bringing in a human to check over things.

With generative validation, customers can boost the performance of their Document Understanding models by getting a ‘second opinion’ from Generative AI. In our tests, this led to a big decrease in AHT owing to up to 20% increase in straight-through processing rate for document-heavy processes.

In this way, generative validation is especially valuable for automating processes with high volumes of documents that usually need human validation due to model uncertainty.

Helping humans in the loop with AI and IDP

The combination of Specialized AI and Generative AI, working together helps to build a better AI ‘brain’ for business. When coupled with the ‘body’ and muscle of automation, you have a powerful booster for business productivity.

A testament to how effectively UiPath has combined AI and IDP be found in the latest analyst reports on the IDP market. UiPath was recently named a Leader in IDC MarketScape: Worldwide Intelligent Document Processing 2023-2024 Vendor Assessment (doc #US49988723, November 2023). IDC sees the AI-powered UiPath Business Automation Platform as “an IDP on-ramp and front door.” According to the report, “UiPath’s success and broad market leadership as an AI-powered automation platform puts it in a great position for capturing growth as new and existing customers look to integrate a document component into their broader automation workflows.”

However, even the smartest AI ‘brain’ is no substitute for the experience, conscience, and judgment of a human one.

Having a human in the loop is still critical to the success of AI-powered automation. Generative validation uses Gen AI to review borderline unattended automations in IDP, but human experts are still essential for evaluating cases where the Specialized AI and Generative AI models lack confidence. Employees enjoy a reduced case load, but the final decision rests with them—as it should be.

As with all our AI innovations, the goal of generative validation remains the same: helping people. With generative validation that means reducing handling times, improving customer experience, and taking pressure off the humans in the loop.

The generative validation feature will soon be available in preview. Join the Insider Program to get early access to new features like this one. We appreciate your feedback.

Insiders will also have access to all the latest announcements about generative validation.

IDC MarketScape vendor analysis model is designed to provide an overview of the competitive fitness of ICT suppliers in a given market. The research methodology utilizes a rigorous scoring methodology based on both qualitative and quantitative criteria that results in a single graphical illustration of each vendor’s position within a given market. The Capabilities score measures vendor product, go-to-market and business execution in the short-term. The Strategy score measures alignment of vendor strategies with customer requirements in a 3-5-year timeframe. Vendor market share is represented by the size of the icons.