Exclusive: ABBYY launches hybrid AI document models

ABBYY has launched Phoenix Plus, a collection of generative AI models designed to work alongside its established document-processing technology, as enterprises confront the reliability, governance and cost problems of moving AI projects into production.

Hybrid launch

Phoenix Plus is initially available through an application programming interface and forms the generative component of ABBYY's broader Phoenix portfolio. The company plans to integrate the models more deeply into its Vantage document AI platform, providing customers with additional controls and guidance for deploying generative AI.

The release reflects ABBYY's position that large language models should complement, rather than replace, deterministic technologies in business-critical document workflows. Its existing models handle functions including image enhancement, layout analysis, parsing, optical character recognition and intelligent character recognition.

Phoenix Plus is intended to add generative capabilities for tasks requiring more flexibility or advanced document understanding. ABBYY said the models have been selected, customised and fine-tuned for document processing, drawing on proprietary datasets, model research and experience gained from processing billions of documents.

"Phoenix is not just one model. It's really a portfolio of both deterministic and generative models that are integrated and optimised for document processing by us at ABBYY," said Slavena Hristova, Director of Product Marketing, ABBYY.

The company is also exploring smaller and domain-specific language models aimed at particular industries and use cases. These models are intended to deliver greater precision while reducing computational requirements and operating costs.

Phoenix Plus does not require customers to standardise on a single AI architecture. ABBYY's document-processing pipeline can apply rule-based extraction, machine learning, large language models or visual language models at different stages, depending on the document and the task involved.

"It allows you to plug, at each one of these steps, the technology that actually makes sense for this part of the process, or that actually makes sense for this use case and for this document type," added Hristova.

A structured invoice, for example, may only need OCR and rules-based extraction. Contracts and other highly unstructured documents may benefit more from generative models capable of interpreting variable language and layouts. The architecture also permits customers to connect models they have already selected or fine-tuned.

Production gap

Generative AI can appear effective when tested against a small number of documents, but scaling the same process to thousands or millions of files introduces additional requirements. Documents used in operational systems vary in quality, structure and complexity, while damaged scans, unusual layouts and incomplete information can create exceptions that models must handle safely.

Production systems therefore need capabilities beyond the initial model inference. These can include image preprocessing, document classification, validation rules, exception handling, workflow integration and human review. Organisations may also need to build monitoring systems and maintain service levels as models and source documents change.

ABBYY cited the experience of a global fund administration company that processes around one million financial documents containing complex tables each year. Its internal AI team spent months trying to automate the work but did not reach the reliability required by the business. The organisation subsequently deployed ABBYY's technology and was operating it within about a month, according to Hristova.

The case illustrates the build-versus-buy decision facing enterprise AI teams. Developing an internal system gives an organisation control over its technology, but it also leaves that organisation responsible for the surrounding infrastructure, integrations, validation mechanisms and long-term maintenance.

Generative models create a particular complication because their outputs are probabilistic. Running the same document through a model more than once may not always produce an identical result. A model can also generate information that does not appear in the source, creating an operational risk when extracted data informs decisions in insurance, banking, finance or other regulated fields.

Those risks become more consequential when automated workflows affect customers directly. Organisations may need to demonstrate where extracted information came from, reproduce processing results and maintain an audit trail covering both automated decisions and human intervention.

Model choice

ABBYY already offers a bring-your-own-model option through a prompt-based activity in its Document AI platform. It provides pre-engineered prompts, facilities for testing them against customers' documents and controls over the resulting output.

The company supports connections to external generative AI services, including models from OpenAI, Google and Mistral. Microsoft Foundry can provide access to a wider selection of models, while ABBYY intends to expand the range of supported connections.

Mistral has attracted interest among European organisations concerned about data sovereignty, Hristova said. Phoenix Plus, by comparison, is hosted by ABBYY and follows the same data privacy arrangements as the Vantage cloud service.

The platform can supply a model with OCR-derived structure rather than sending an unprocessed PDF for the model to interpret independently. ABBYY also plans to make its DocLang format available in Vantage. The format is designed to preserve more of a document's structure and context, reducing the work required from a generative model and helping constrain its output.

This orchestration layer is intended to route each part of a workflow to the most appropriate technology. Customers can use ABBYY's generative models, connect an external model or retain deterministic processing where it provides sufficient accuracy.

Generative AI can also shorten the initial configuration process. Traditional machine-learning deployments may require customers to collect, label and train models on large sets of sample documents. Generative systems can begin extracting information from far fewer examples, potentially reducing the time needed to bring a new document type into an automated workflow.

ABBYY's hybrid approach retains that flexibility while using deterministic tools for tasks where repeatability and precise extraction are more important. Generative models can then be applied to exceptions, unstructured content or steps that previously required human interpretation.

Cost controls

The economics of enterprise AI are becoming a larger part of deployment decisions as organisations impose limits on token consumption and per-employee use. A model that works economically during a proof of concept may become expensive when applied to high document volumes, particularly if it must repeatedly interpret entire files.

CPU-optimised deterministic models can be less expensive for routine functions such as OCR, classification and rules-based extraction. Restricting generative AI to the stages where it adds measurable capability can also make processing costs more predictable.

"The fact is that this approach of throwing AI and an LLM at every problem is not going to be tolerated for a really long time from now on by finance departments, because there is uncontrolled, unpredictable cost, and very often there is no need to use AI for something that can be solved with much more efficient, much more reliable, much more accurate technology, and something that is simply cheaper to run and operate," said Hristova.

Costs associated with an internally developed document AI system extend beyond model inference. Organisations must account for infrastructure, validation, monitoring, integration work, exception management and model drift. They also assume responsibility for service-level agreements and the engineering resources required to maintain the system.

For ABBYY, the production case rests on selecting technology at the level of the individual task rather than applying a single model across the full workflow.

"You end up optimising the process for reliability, for accuracy, for speed instead of trying to fit the use case to the technology," added Hristova.