Embracing AI Agents: KPMG’s Enterprise-Level New Productivity Solutions

There are three important stages for the implementation of enterprise-level new productivity: the automation stage centered on RPA, the intelligent stage centered on RPA+AI integration, and the human-machine collaboration stage centered on AI Agents. However, most enterprises’ new productivity remains in the automation and intelligent stages, where digital technologies act as passive productivity tools, executing tasks only under clear human directives and rules, with a highly limited level of autonomy and business processing capabilities.

The performance of AI Agents relies on the maturity of large models (LLM), while the diversity and complexity of enterprises’ data present significant bottlenecks affecting performance. Without a flexible “brain,” planning, decision-making, and action capabilities are difficult to ascertain. Moreover, the implementation of AI Agents lacks overall capability planning, and the integration of API ecosystems and external tool capabilities is insufficient, limiting the operability of enterprise-level scenarios.

As the potential of AI Agents continues to be explored, viewing them solely from the perspective of individual application scenarios cannot support the disruption and transformation of enterprise-level production relations. It is necessary to redefine the positioning, responsibilities, and capability boundaries of AI Agents within the business value chain, establishing a closed-loop, transparent management mechanism centered around new productivity to reshape the division of labor and collaborative relationships between humans and digital technologies, driving the large-scale implementation and iterative evolution of enterprise operational models.

Therefore, KPMG’s Technology Empowerment (TE) team adheres to a human-centered, digitally intelligent integrated management philosophy, providing enterprise-level new productivity solutions centered on AI Agent technology. The focus is on advancing the establishment of productivity systems from two aspects: integrated platform capabilities and horizontal and vertical collaborative operational mechanisms, leading the landing and large-scale promotion of enterprise-level star applications with profound business insights, ultimately achieving a comprehensive innovation of productivity and production relations.

Data quality is the foundation for training enterprise large models. Establishing a set of data management mechanisms and work paradigms to achieve efficient storage, processing, and analysis of big data is critical, focusing on breakthroughs in data labeling automation, synthetic data generation, and enhancing data privacy and protection in important sub-fields to support the construction and application of enterprise-level large models.

Starting from general large models and vertical large models, establish enterprise-level large model capabilities. General large models possess broad applicability and high flexibility, suitable for information collection, intelligent Q&A, and other general scenarios; vertical large models, through deep knowledge accumulation in specific industry verticals and pre-training on enterprise data, are more suitable for focused business scenarios and services, supporting personalized “expert” industry solution outputs.

Fully adapt to the existing digital architecture of enterprises, integrating enterprise-level business system APIs and various tool capabilities, embedding AI Agents into the enterprise digital business environment. Under the planning and task orchestration of the “knowledge brain,” flexibly invoke various operational capabilities to achieve the execution and realization of established goals.

From a technical security perspective, new productivity in different forms requires ensuring transparency in processes and decisions, managing its entire lifecycle uniformly, including demand management, development management, operational management, and evaluation management, achieving real-time monitoring and comprehensive insights into new productivity, enhancing operational management efficiency through data-driven methods.

Horizontal collaboration of business and digital organizations accelerates the creation of enterprise star applications. The implementation of new productivity lies not only in capability construction but also in the joint development of digital intelligent scenario solutions. Driven by business consulting and relying on enterprise-level large models, it targets more complex business scenarios and demands, injecting enterprise-level business experience and knowledge assets into customized AI Agents, expanding capability boundaries, extending scenario ranges, and accelerating the creation of enterprise star applications. From point to surface, promoting usage, gradually achieving application upgrades from general Agents to specialized Agents, delivering new productivity to the most urgent and efficient value “frontlines.”

Vertical integration across levels promotes enterprise-scale applications. Establish vertical integration channels that can be accessed from top to bottom, fully encouraging various departments to leverage frontline business insights, exploring and constructing various scenario applications, and promoting reuse at the company level, achieving localized innovation and universal sharing.

For example, a large state-owned enterprise client’s practice of utilizing digital technology to empower the group in creating new productivity. In the era of digital economy, the client faces a rapidly changing, competitive external market environment and urgently hopes to explore the best practices suitable for its situation through innovative practices and value release of new digital technologies in data assets.

The KPMG team built a digital solution tailored to the industry characteristics and enterprise features based on the client’s digital transformation strategy, focusing on the integration of digital intelligence as the key line to create innovative tools for new productivity. The KPMG team centered on full-domain data integration, AI LLM large model applications, RAG retrieval-enhanced production practices, and industry and enterprise professional knowledge bases as the core empowerment handles, forming practical support for diverse scenarios such as data analysis, business insights, leadership decision-making, and frontline empowerment, constructing leading new productivity for the enterprise.

The solution follows the basic process of insight-driven decision-making in enterprise operations, focusing on four basic steps: understanding the current situation, identifying root causes, predicting the future, and making decisions. Through chat-based human-machine voice or text interactions, the “Digital Assistant” is activated in real-time, achieving natural language conversion, precise semantic representation, LLM large model collaboration, and professional knowledge base retrieval learning, completing data analysis and prediction, insights and decisions, as well as tracking and follow-up of tasks. This helps frontline business personnel, business department heads, and company leaders better understand business status and development dynamics in a flexible and timely manner, providing valuable reference information and business suggestions, driving efficient and precise business operations, and further enhancing the lean level of enterprise resource allocation.