ABBYY Ascend Showcases AI-Based IDP Capabilities

ABBYY recently initiated Ascend - a series of virtual events designed to showcase the Capture & IDP vendor’s latest innovations. The first Ascend Webinar was held earlier this month and focused on a number of recent innovations from the global firm, which recently announced it has relocated its global headquarters to Austin, TX, which is also the home base of Chief Product and Technology Officer Patrick Jean.

Two of the big announcements focused on ABBYY’s recent AI innovation. The first was the introduction of Phoenix 1.0, a new multi-modal AI model that is purpose built for document processing. Phoenix represents the latest AI technology from ABBYY, which has been utilizing AI in its products since the mid-to-late 2010s. Phoenix incorporates a combination of convolutional neural networks and the RoBERTa LLM model, which is fined tuned for document processing.

Phoenix is complementary to the technology ABBYY has utilized to create more than 80 document models that are currently available as part of its Vantage IDP platform. Phoenix’s initial application will be to enable zero-shot learning and auto-labeling in Vantage to aid in the onboarding process for new document types.

The second Ascend AI-focused announcement has to do with the introduction of low-code/no code integration of Vantage with LLMs. ABBYY customers are already leveraging LLMs to handle unstructured documents through their Vantage workflows. In these cases, the Vantage models are used to provide extraction from structured and unstructured forms, with the LLMs only being called upon when required. This controls the number tokens sent to and received from LLMs, lowering licensing costs. Vantage can also be used to check and verify data being extracted by the LLMs, through techniques like field-related rules and database look-ups, thereby reducing hallucinations. Previously, ABBYY had used custom scripting to enable LLM integration.

Other advancements announced by ABBYY include the introduction of a new API for its Timeline Process Intelligence application to improve access to log data, as well as improving the scalability of Timeline to enable customers to perform more complex data analysis.

ABBYY also got back to its roots a bit and announced improvements to its recognition technology, as well as a renewed focus on its developer network. The recognition improvements include enhanced OCR, new handwriting recognition for multiple languages and simplified API support, as well as updated support for several development languages, including .NET and Python.

Historically ABBYY has been a market leader of licensing recognition toolkits to ISV and developers to utilize in their applications. In recent years, however, this portion of ABBYY’s business has increasingly been encroached upon, primarily by hyperscalers. This includes AWS, MS Azure, and Google Cloud, which offer AI-powered OCR and document processing through their ubiquitous cloud services. ABBYY is looking to reestablish itself by promoting its technology as price competitive, as well as doing a better job than the hyperscalers.

Infosource take

ABBYY, which in recent years has rated as a Star in our annual Capture & IDP vendor rankings, has come under a lot of pressure. It was an early innovator in AI but failed to fully capitalize on this early mover status, partially because fully bringing the Vantage product to market was a long process. In the meantime, the IDP space was being flooded with a sea of AI-first IDP ISVs. In addition, as mentioned, the hyperscalers have impinged on ABBYY’s lucrative recognition toolkit business. Also, ABBYY, which was founded in Russia, has also gone through organizational challenges as it has moved away from those roots.

One purpose behind Ascend is to illustrate ABBYY’s position in the market, as an innovative ISV that leverages AI, but also deeply understands document driven processes and applications and is not trying to solve everything with the latest flavor of AI technology. It’s a step in the right direction for a vendor that still has strong brand recognition, a solid and deep technology base, and a global footprint of customers and partners. Of course, none of this means that the market around ABBYY isn’t also continuing to evolve and move forward, but it’s certainly good to see ABBYY plant its foot in the ground, make some statements about what it can do and where it’s headed, and then presumably move to execute on these plans.