EY, Deloitte, Digits announce agentic AI partnerships with Nvidia

Big Four firms EY and Deloitte, as well as accounting automation solutions provider Digits, all announced partnerships with technology company Nvidia, which has gone from a company known mainly for its video graphics cards to a major player in the AI space within just a few years. 

EY and Deloitte's respective announcements both centered around the launch of their own agentic AI platforms built on Nvidia's technology infrastructure, including its new Llama Nemotron family of open reasoning models, which is adapted from the Llama LLM initially developed by Meta but used widely since it leaked in 2023.

Nvidia said that, through training and refinement, Llama can more effectively perform multistep math, coding, reasoning and complex decision-making.

Deloitte's Zora AI

Deloitte announced the release of its Zora AI by Deloitte platform Monday, which offers a suite of ready-to-deploy agents that are said to perceive, reason and act, autonomously executing complex business functions, serving as a way for clients to augment their workforce as well as boost their effectiveness. The three agents offered by the platform are Zora AI for Finance, Zora AI for Procurement and Zora AI for sales and marketing. 

Deloitte said the agents can source, extract and interpret real-time, multimodal data from structured and unstructured sources; run analytical and mathematical models to define related insights and trends; translate insights into easily consumed formats; provide scenario analysis and recommendations on business-critical decisions; and coordinate and perform a set of actions—in collaboration with other agents—to execute complex, nuanced workflows, from beginning to end, including transaction processing, anomaly detection and resolution, and self healing. 

"We are entering the autonomous enterprise era where agents can transform work and business models, ushering in entirely new ways of working," said Deloitte US CEO Jason Girzadas in a statement. "Our vision with Zora AI is to assist our clients in their transition into this new era, where agents and employees interact to reinvent business processes and unlock new sources of business value, growth and innovation for their organizations."

Deloitte itself is using Zora AI for Finance internally to streamline and automate its finance processes, including expense management. The expense management agents monitor expenses across payroll, facilities, sales and marketing, and employee time and expenses, enabling finance leaders to identify expense outliers, compare expenses against industry and competitor trends, and drill down into specific budgets. Deloitte estimates Zora AI will reduce its costs by 25% and increase its productivity by 40%. Deloitte plans to implement Zora AI for thousands of users by the end of 2025.

EY.ai Agentic Platform

EY also announced Monday the deployment of its own EY.ai Agentic Platform on the full Nvidia AI stack to respond to real-time events, adapt to regulatory changes and drive smarter financial and risk decisions across global operations. The EY.ai Agentic Platform will run across client clouds, on-premises, at the edge, and the Nvidia Cloud Provider ecosystem. Nvidia is holding a conference in San Jose this week.

The platform is, for now, primarily for internal use as part of EY's "Client Zero" transformation, in which EY tests AI deployments to guide clients as an example of effective and responsible use of the technology. This initial deployment will integrate 150 AI agents supporting 80,000 EY professionals across data collection, document analysis and review, and income and indirect tax compliance. EY.ai risk agents will also work with risk professionals to deliver new AI-native services. The third-party risk management agent will enable clients to manage risk more comprehensively and increase productivity.

The platform overall supports EY's Responsible AI Frameworks as well as Nvidia NeMo Guardrails and EY SafePrompt software to more effectively mitigate AI risk at the agent level. It also supports a framework for agent creation and orchestration, which will use Nvidia Blueprints, including AI-Q Blueprints, to operate across third-party agent platforms. These frameworks will, in turn, support a collection of "fit for purpose" models chosen and designed for agentic solutions, such as indirect tax, income tax compliance, financial crimes, regulatory compliance and financial reporting, and targeted sector solutions for finance, marketing, cyber resiliency and supply chain, all powered by client-specific reasoning models. The platform also has a Model Development Suite that lets users create custom, enterprise-ready AI reasoning models using NVIDIA AI Foundry.

"With the EY.ai Agentic Platform, we are moving fast to help the world's largest organizations transform their enterprises and streamline increasingly complex compliance requirements, while enhancing productivity and operational excellence across our own businesses," said EY global chair and CEO Janet Truncale. "In collaboration with NVIDIA, we're harnessing the collective knowledge of 400,000 skilled professionals, and the broad spectrum of EY services, to help shape the future with confidence in a fast-moving, highly competitive global economy."

Digits AGL on NVIDIA Triton

Finally, accounting automation and solutions provider Digits, on the same day, also announced that its own complete solution, centered around the recently-released Autonomous General Ledger, successfully developed and deployed vertical-specific large language models (LLMs), achieved using NVIDIA accelerated computing and NVIDIA Triton Inference Server, which is optimized for AI models.

Digits said that, through using this optimized inference server, they were able to increase the number of requests it can process (a metric generally referred to as "LLM throughput") tenfold to create its verticalized application of Accounting AI. By "vertical," Digits means the models, rather than having generalized training like ChatGPT or Claude, have been trained only on relevant, domain-specific information, which focuses outputs and reduces the possibility of inaccurate information. This reflects an overall move in the industry away from generic one-size-fits-all models, of which there are now many, toward specialized applications that deeply understand specific business domains. 

"You can think of LLMs as very generic," said Digits CEO Jeff Seibert in an email. "They train on substantially the entire internet, and they have a broad base of horizontal knowledge, but they are not specialized in any specific field. We have combined the power of LLMs with over a dozen custom-trained models that specialize in double-entry accounting and the related workflows [specific to the accounting industry]."

When asked about the development process, he said Digits both fine-tuned publicly available LLMs to be more accurate for given tasks and spent five years training its own predictive models from the ground up on a proprietary data set of $825 billion worth of transactions. 

"You can think of Digits AGL as a symphony: we orchestrate over a dozen models together in production, most of which are completely unique and created from scratch in-house," said Seibert. 

While Digits has been providing solutions since 2018, Seibert said this will be the first time the company is launching the full set of products, including the Automated General Ledger. 

"Previously, only pieces of Digits have been available (Reporting, Dashboards, Bill Pay and Invoicing), and this is the first time we are launching the full ledger — the AGL — to actually automate the bookkeeping," he said. "After a year of intensive testing with hundreds of businesses via our full-service accounting offering, we've now launched it self-serve for small business owners and startup founders to automate their finances."