PwC becomes first to deliver reliable AI reasoning for enterprise-grade spreadsheets 

Breakthrough removes a long-standing bottleneck in applying AI to the world’s most critical business decisions, unlocking AI across enterprise workflows and high-stakes decision-making.

Frontier agent delivers up to 3x higher accuracy on real-world, enterprise-grade spreadsheets — using 50% fewer tokens.

Today, PwC US announced a major advance in applied artificial intelligence: a frontier agent that enables AI systems, apps, or other agents to understand and navigate the kind of large, complex, multi-sheet workbooks used every day across the enterprise. The frontier agent can surface insights, trace logic across sheets, and explain outcomes — all with greater accuracy, speed, and efficiency than previously possible for the real-world spreadsheets used inside the largest organizations.

For decades, the most important business decisions have lived inside spreadsheets. While AI has transformed how organizations work with text, images, and code, enterprise-scale spreadsheets remain beyond the reach of even the most advanced models. Many workbooks contain hundreds of thousands of rows, dense formulas, hidden dependencies, and multimodal inputs like charts and receipts. This level of complexity exceeds the reasoning limits of conventional AI systems and prevents them from reliably tracing logic, validating calculations, or producing defensible outputs.

As a result, large portions of enterprise data remain effectively inaccessible to AI, particularly in regulated, high-stakes environments where precision and defensibility are essential. For companies seeking broader value from their AI investments, this constraint has left significant opportunities for automation, deeper insight, and more advanced agentic operations unrealized within one of the most critical layers of enterprise work.

The frontier agent, a first in the industry, makes PwC the first organization to successfully bring reliable AI reasoning to enterprise-scale spreadsheets, unlocking new ways for practitioners to apply AI to work that previously required extensive human review. In internal research, PwC’s approach performed an average of 3x more accurately than previously published methods, while using up to 50% fewer tokens — enabling employees to run larger workloads, analyze bigger and more complex datasets, and tackle more sophisticated problems with greater speed, lower cost, and improved energy efficiency.

Solving one of enterprise AI’s hardest, most important practical challenges

Under the hood, PwC’s frontier agent combines multimodal pattern recognition with a retrieval-augmented architecture that mimics how experienced practitioners work: scanning, searching, jumping across tabs, integrating charts and receipts, and reasoning.

The frontier agent is designed to reason across enterprise-scale spreadsheets with both speed and precision. It can analyze up to 30 workbooks containing nearly four million cells and complex cross-sheet formulas — the scale commonly found in deals, supply chain modeling, healthcare analytics, and other mission-critical workflows.

“The AI industry has spent years teaching models to understand text, images, and code. But some of the most critical business decisions still live inside spreadsheets that are too complex, too large, and too messy for any AI to reason over reliably,” said Matt Wood, Commercial Technology & Innovation Officer, PwC Global & US. “That gap between what AI can do and what business actually needs — that's what we set out to close.”

Unlocking new value across Audit, Tax, and Advisory

Built by PwC’s engineers, this capability is already in use across many business areas, enabling the firm to tackle some of the most complex professional challenges using AI— all of which were previously not possible. For example:

• Automating complex engagement documentation at scale: PwC teams work with highly structured workbooks that document business processes, controls, and procedures for engagements. Though nominally standardized, these files can vary significantly in practice: column names shift, fields appear in different orders, and structures change between engagements. Many files are too large for any AI system to process directly. PwC's agent handles this in two stages: first, it reads the workbook to identify its structure — locating subprocesses, mapping row ranges, and building a schema. Then it extracts the specific control details, procedures, and descriptions needed, using targeted retrieval rather than trying to ingest the entire file. The result is reliable, repeatable extraction from documents that previously required manual review.
• Accelerating risk assessment from weeks to hours: Risk assessment workflows depend on reading through multiple heavily formatted files and extracting specific information for downstream analysis. The volume and complexity of these files previously required weeks of custom development effort to handle programmatically — and each new set of files could break existing approaches. PwC's agent now indexes and extracts the relevant data directly, regardless of formatting variations. What previously took weeks of development to configure can now be set up in hours. PwC evaluated alternative approaches, and none matched the accuracy or reliability of its retrieval-augmented system.

A new chapter in PwC’s technology leadership

This milestone builds on PwC’s ongoing investments in AI. By enabling AI to reason over enterprise spreadsheets, one of the most complex and critical components of modern business, PwC is extending the practical reach of AI into areas previously considered off-limits.

“This didn't start as a research project,” Wood added. “It started because our teams were spending weeks manually tracing logic through workbooks that no existing tool could handle. The fact that the solution turned out to be a genuine advance in the field tells you something about the kind of problems our practitioners face every day — and the kind of engineering talent we've built to solve them."  

About PwC

At PwC, we help clients build trust and reinvent so they can turn complexity into competitive advantage. We’re a tech-forward, people-empowered network with more than 364,000 people in 136 countries and 137 territories. Across audit and assurance, tax and legal, deals and consulting, we help clients build, accelerate, and sustain momentum. Find out more at www.pwc.com.

© 2026 PwC. All rights reserved.