The following partner insight was authored by José Márquez, AI Director, Advisory, PwC US.
Executive Summary
In today’s fast-paced and tech-disrupted business landscape, enterprises are turning to automation to streamline operations and stay competitive. Enterprise Resource Planning (ERP) systems are the digital nervous system of modern enterprises, orchestrating core business functions across finance, supply chain, human resources, and operations. The past decade has seen the rise of deterministic process automation, also traditionally known as RPA, as a key enabler of efficiency within ERP environments. This technology has excelled at automating repetitive, rules-based tasks – delivering speed, accuracy, and cost savings. From invoice processing to system integrations, this approach has digitized many labor-intensive workflows. At the same time, it was a key element to integrating processes and tasks via different interfaces that span across various enterprise systems, as ERP itself is no longer a monolithic system.
However, the traditional automation techniques are reaching limits. Their deterministic logic makes it brittle when faced with variability in data or context. Complex decisions, unstructured inputs, or evolving high-touch processes require a new approach for automation. This is where generative AI (GenAI), also understood as agentic AI while being materialized into modularized lifecycle managed automations enters the scene, not as a replacement, but as a complement that extends the reach of automation into the (so far human-driven) non-deterministic domain of understanding, reasoning, and creativity.
Individually, each technology offers substantial benefits; combined, they create a powerful synergy that addresses each other’s limitations and maximizes automation outcomes.
Research shows that organizations adopting agentic enhancements expect performance improvements of 35% on average. By blending deterministic with non-deterministic agentic automations, companies can automate end-to-end workflows—from routine transaction processing to intelligent decision support—with greater accuracy, flexibility, and scale.
From Traditional Automation to Agentic AI: Complementary Forces
Traditional automation and agentic AI are fundamentally different, yet highly synergistic. RPA excels at consistency; agentic AI thrives on complexity. RPA follows pre-defined rules; agentic AI learns and adapts. At the same time, while agentic AI offers complex decision-making elements, RPA adds system integration and a framework to make insights actionable. Alone, each has limits — together, they overcome them. The real breakthrough comes from integrating the former with the latter. For example, RPA may flawlessly process thousands of transactions daily but cannot interpret a freeform customer request. Agentic AI can understand such ambiguity, propose a course of action, and then pass it back/forward to RPA for execution in the ERP backend system.
In this hybrid model, agentic AI performs perception, synthesis, supports decision-making (e.g., extracting intent from e-mails, leveraging ERP and multi-source knowledge, recommending actions); while RPA enacts the decisions within ERP systems. This division of labor brings automation to processes that were once considered too variable or unstructured to be automated effectively.
The Challenge: Enterprise-Readiness for Agentic AI
While classic deterministic automations enjoy enterprise-grade maturity, agentic AI adoption still faces hurdles. Prime among them is lifecycle management: agentic AI automations will also be evolving over time and should keep up with models, configurations, rules, grounding, guardrails, prompting, knowledge bases, etc., and require therefore ongoing governance. Enterprises should have scalable pipelines to develop, test, version, release, deploy, and monitor agentic AI automations — practices that are still emerging. Moreover, organizations must reconcile deterministic expectations of ERP with the agentic AI probabilistic nature.
Governance, risk and compliance, as well as real-time monitoring remain critical gaps. Enterprises require tracing, transparency andexplainability of decisions, especially in regulated domains like finance or healthcare. Yet agentic AI influence can be opaque. Without integrated tracing and observability across agentic AI components, errors may go undetected. Achieving enterprise-readiness requires end-to-end trust on both, central ERP systems as well as across supply chains.
Deterministic + Non-Deterministic: A Proven Synergy Towards an Agentic ERP
To understand why the bundle of both technologies form the most effective automation approach, consider this: RPA is a function f(x) that deterministically maps input x to an output. Agentic AI, on the other hand, represents a conditional probability distribution P(y|x), offering diverse but plausible outputs based on input x. When fused, this yields a novel composite automation pipeline:
Automation_Pipeline = RPA n(agentic AI(RPA n-1(Input), …), …)
Here, RPA n-1pre-processes data, agentic AI interprets or transforms it, and RPA nfinalizes the action. The two layers cooperate like an ensemble: RPA handles structure; agentic AI handles ambiguity.
This synergy can be quantified. Suppose a classic pareto distribution where 80% of a process is routine and handled by RPA with near 100% accuracy, while the remaining 20% is complex and handled by agentic AI with accuracy p. The overall automation accuracy becomes:
Automation Accuracy = (0.80×100%)+(0.20×p%)
With downstream deterministic validations, the risk of agentic AI errors slipping through is reduced further, effectively multiplying the agentic AI error rate by the probability RPA fails to catch it — a much smaller value. This results in a strongand resilient process flow that neither RPA nor agentic AI could achieve alone.
Ultimately, the simplified formula for a reengineered business process, being automated by both: RPA and agentic AI, would read as follows:
Reengineered_Business_Process = Automation_Pipeline n+ Automation_Pipeline n+1
Flagship Use Cases
Finance: Transforming Financial Processes
Finance functions are well-suited to this hybrid approach due to the dual nature of their work — highly structured yet requiring judgment. Traditional automation can tackle transaction-heavy workflows such as journal entries, reconciliations, and payment processing. Agentic AI adds cognitive capabilities like interpreting semi-structured invoices, identifying anomalies, or generating financial forecasts. Together, they help procure-to-pay, order-to-cash, and record-to-report cycles, improving accuracy and decision speed.
Supply Chain: Enhancing Agility and Accuracy in Operations
Supply chains are dynamic and data-rich — an ideal playground for traditional automation and agentic AI. RPA enables consistent execution of inventory updates, procurement, and shipment processes. Meanwhile, agentic AI enhances demand forecasting, identifies supply risks, and interprets market signals. When disruptions occur, agentic AI can propose mitigation strategies, which RPA could rapidly implement, creating a resilient and adaptive supply chain.
Analytics: Driving Real-Time Insights and Decision-Making
In business intelligence, the power of traditional automation and agentic AI is transformative. RPA automates the collection and cleansing of data from disparate systems. Agentic AI then interprets this data, uncovers patterns, and generates human-like narratives. This not only reduces the manual effort while generating reports but also elevates decision-making with real-time, context-aware insights. E.g., systems can automatically generate regular reports or one-time ones written in natural language, tailored to KPIs.
Outlook: Real-Time Intelligence, Proactive, and Adaptive Workflows
The next frontier lies in high agency automation. Agentic systems will not just react to transactional flows but anticipate them — spotting risks, generating alerts, and even initiating remediation workflows autonomously. Real-time integrations between ERP systems, middleware platforms, and workflow engines will allow issues to be addressed before they escalate.
Imagine a finance process reengineering that flags a liquidity risk, explains it in natural language, and routes it to a decision-maker with recommended actions — all in real time. Or a supply chain one that spots delays in Asia and auto-adjusts sourcing rules. These capabilities will likely be the hallmark of agentic ERP automations — real-time context-aware, predictive, and anticipatory.
Now Is the Time to Act
Traditional automation and agentic AI together offer not just efficiency, but transformation. They can unlock new process frontiers, drive smarter decisions, and create resilient operations. PwC invites you to explore how this hybrid model can reengineer your ERP workflows.
AI agents think. RPA automations do. Humans lead.
Let’s build the intelligent enterprise, together!
José Márquez is AI Director, Advisory, PwC US.