Audit Evidence Graph

The result of the past fifty to seventy five years of constructing accounting information systems is craft-based processes where data integrity depends on human discipline, human bucket brigades which move information,  rather than based on good system design. 

The question is not whether there are better alternatives; there are.  The question is why the accounting and audit profession has tolerated this current situation for so long, and what it will take to transition to something fundamentally more sound.

Imagine being able to click on a link to view the audit evidence knowledge graph which attests to the fairness of a financial statement. Imagine an artificial intelligence (AI) agent or a human being able to interpret that information reliably.  Why imagine it? Why not build it?

The Nature of the Problem

Gartner points out (page 2) that the typical Fortune 1000 company uses about 800 electronic spreadsheets to create its compliance report.  Each of those 800 spreadsheets is, in essence, a self-contained silo of data, logic, and institutional knowledge. It contains:

  • Data: numbers pulled from the general ledger, sub-ledgers, external sources, or other spreadsheets
  • Logic: formulas, calculations, allocations, and transformations encoded in cell references and functions
  • Assumptions: judgment calls, estimates, and policy choices embedded in the structure of the spreadsheet
  • Institutional knowledge: the unwritten understanding of why the spreadsheet is structured the way it is, what the tabs mean, which cells are hard-coded overrides versus calculated values, and how it connects to the 799 other spreadsheets

The problems of this approach are well understood, well-documented, and severe:
  • Error rates: Academic research consistently finds material errors in a significant percentage of operational spreadsheets. The European Spreadsheet Risks Interest Group (EuSpRIG) has cataloged hundreds of cases where spreadsheet errors led to material financial misstatements, regulatory failures, and operational losses. In the context of those 800 spreadsheets used by those Fortune 1000 companies, even a low per-spreadsheet error rate compounds into a near-certainty of material errors somewhere in the chain.
  • Opacity: No single person understands all 800 spreadsheets. Each is typically maintained by a different individual or team, with its own conventions, its own undocumented logic, and its own fragilities. The knowledge of how they interconnect is distributed across dozens of people, many of whom may have left the organization.
  • Version control failures: Spreadsheets are copied, emailed, modified, and re-saved. At any given moment during the close process, multiple versions of the same spreadsheet may exist, with no reliable mechanism to determine which is current or authoritative. The result is a process where data integrity depends on human discipline rather than system design.
  • Fragile interconnections: The 800 spreadsheets are not independent; they are linked via cell references, copy-paste operations, and manual re-keying. A change in one spreadsheet can cascade through dozens of others in ways that are difficult to predict and impossible to trace systematically. These interconnections are typically undocumented and exist only in the tacit knowledge of the people who maintain them.
  • Audit difficulty: When an auditor examines the closing book, they are essentially reverse-engineering the logic of 800 interconnected electronic spreadsheets to determine whether the numbers in the compliance report are reliable. This is enormously time-consuming, requires the auditor to trust that the spreadsheet logic correctly represents the entity's accounting policies and transactions, and provides limited assurance that the interconnected system as a whole is coherent.
It is worth pointing out that the auditor documentation, the audit bundle the auditor prepares, which also tends to be electronic spreadsheets and documents has exactly the same problems as those 800 electronic spreadsheets used by those Fortune 1000 companies.

The Deeper Problem: Knowledge Without Structure

The 800-spreadsheet model represents what happens when complex accounting knowledge is encoded in a medium that has no inherent structure, no semantic richness, and no machine-readable logic. A spreadsheet cell containing "=SUM(B5:B17)" tells a machine nothing about what the sum represents, why it is calculated, or how it relates to other elements of the financial reporting framework. The meaning exists only in the mind of the person who built the formula in that specific electronic spreadsheet.

This is precisely the gap that XBRL and the Seattle Method's knowledge graphs of financial reporting information are designed to fill. The question is not whether a graph-based alternative to 800 spreadsheets is better. The question is why the profession has tolerated the spreadsheet model for so long, and what it will take to transition to something fundamentally more sound.

It is time to think beyond the electronic spreadsheet, beyond the document, separating the document "costume" that the financial statement wears from the information the financial statement contains, and think knowledge graphs instead.

It is time to build industrial processes rather than continue with the fragile, brittle, error prone, fragmented, disconnected. It is time to leverage the computational substrate accounting information systems provide rather than fight it.

Replace Documents with Graphs

A graph-based digital financial reporting framework, a digital closing book, and a digital audit bundle would replace those 800 disconnected spreadsheets and other documents and electronic spreadsheets in this financial reporting and compliance audit supply chain with a single, interconnected knowledge graph in which every financial data point is:
  • Tagged with its identity and meaning (what concept it represents)
  • Linked to its source (where the number came from; the general ledger, a sub-ledger, a calculation, a manual estimate)
  • Connected to its dependencies (what other data points feed into it or flow from it)
  • Constrained by the rules of the financial reporting scheme (what must be true for the graph to be internally consistent)
  • Verified using reliable automated processes which leverage the deterministic nature of accounting and the rich set of  standardized accounting terminology
  • Versioned with complete traceability/trackability (who changed what, when, and why)
In graph terminology, each financial data point is a node, and each relationship (calculation, dependency, source, assertion) is an edge. The financial reporting framework, closing book, audit bundle, and financial statement are not a collection of documents; they are living, navigable map of the entity's financial position and changes in that financial position, where every number can be traced to its origin and every relationship is explicit.

How All this Connects to the Seattle Method

The Seattle Method's proven, reliable framework provide leverage and a structural template for this graph based model using global open industry standard XBRL. The knowledge graphs that the Seattle Method constructs, from the basic accounting equation through progressively more complex representations which will eventually include all of US GAAP and IFRS define:
  • What nodes a complete closing book must contain
  • What edges (relationships) must exist between those nodes
  • What constraints must be satisfied for the closing book to be internally consistent
  • What the hierarchical structure of the financial reporting framework looks like, so that the closing book is organized according to the actual logic of accounting rather than the arbitrary structure of 800 spreadsheets
The closing book becomes, in effect, an instantiation of the financial reporting framework, a specific entity's financial model/data mapped onto the general structures (patterns) that the Seattle Method defines. This is a powerful architectural relationship: the Seattle Method provides the framework, and the closing book, the audit bundle, the financial statement provides the instance.

The Transformation this Enables

Rather than human oriented bucket brigades, new approaches to working within that financial reporting supply chain can be employed, including:
  • Real-time validation during the close: Rather than completing the close and then discovering errors during the subsequent audit, a graph-based closing book can be validated against the financial reporting framework continuously as data is entered. If a calculation relationship is violated, if a required node is missing, or if an inconsistency is detected between interdependent elements, the system can flag it immediately; when it is cheap and easy to correct, rather than weeks or months later.
  • Effective control: Electronic spreadsheets are too flexible to control, so it is hard to create scalable processes using electronic spreadsheets.  But ERP systems tend to be too inflexible; that is why we use electronic spreadsheets. But graph-based structures are controllable at scale.
  • Enforceable canonical artifacts and templates: Because of the control and because graph-based nodes and edges can be worked with directly good practices and best practices can be reliably specified and enforced.
  • Leverage Lean Six Sigma philosophies, techniques, practices: Because software can work with and interpret objects of a digital closing book, digital audit bundle, digital reporting framework, or digital report; work can be reliably and effectively automated. (e.g. Poka Yoke mistake proofing) Processes can be fixed rather than errors remediated of failures dealt with.
  • Automated completeness assessment: The financial reporting scheme defines what a complete set of financial statements looks like. A graph-based closing book can be automatically checked against this template to determine whether all required elements are present. This is not a syntactic check (are all tags present?) but a structural check (does the graph contain all the nodes and edges that the financial reporting framework requires?).
  • Traceable interconnections: When a number in the income statement changes, the graph immediately reveals every other node that is affected (e.g. fails loudly); the tax provision, the earnings per share calculation, the segment disclosures, the statement of cash flows. This traceability eliminates the hidden dependencies that plague spreadsheet-based processes, where a change in one spreadsheet silently corrupts calculations in others.
  • AI-readiness: A graph-based closing book, audit bundle, reporting framework, and financial statement are all natively consumable by both rules-based and generative AI:
    • Rules-based systems can traverse the graph, validate its structure, and check its consistency
    • LLMs can reason about the graph's contents, identify anomalies, and generate analyses with full knowledge of the underlying structure
    • Machine learning models can be trained on historical closing book graphs to identify patterns, predict issues, and optimize the close process
  • Industrial strength processes: Process improvement, improved quality, reduction in costs, increases in value to your organization.
This is the critical point: the spreadsheet-based closing book, audit bundle, financial reporting framework, and financial statement are essentially opaque to AI. An AI system confronted with 800 electronic spreadsheets must first parse and interpret each one independently, reverse-engineer their interconnections, and attempt to infer the underlying accounting logic, a task that is unreliable and error-prone. A graph-based artifacts, by contrast, presents the same information in a form that AI systems can consume directly, reliably, and at scale.

The XBRL-Based Digital Audit Bundle

The audit bundle, the collection of documentation that supports the auditor's opinion, is currently assembled as a package of electronic documents. Workpapers are organized by financial statement area (cash, receivables, inventory, revenue, etc.) or by assertion (existence, completeness, valuation, rights and obligations, presentation and disclosure). Each workpaper typically contains:
  • The data being tested
  • The procedure(s) performed
  • The evidence obtained
  • The conclusion(s) reached
This documentation is valuable but structurally limited. It is organized around the audit methodology rather than around the financial reporting logic. The connections between workpapers; how the revenue testing relates to the receivables testing, how the management estimate evaluation relates to the disclosure assessment; exist implicitly in the auditor's judgment but are not explicitly represented in the documentation structure.

An XBRL-based digital audit bundle would transform audit documentation from a collection of documents into a structured, tagged, navigable evidence graph in which:
  • Every piece of evidence is tagged to specific financial statement elements. Rather than organizing evidence by audit area, the evidence graph links each piece of evidence directly to the specific XBRL-tagged financial statement element(s) it supports. The receivables confirmation evidence is linked to the Accounts Receivable node; the depreciation calculation workpaper is linked to the Depreciation and Amortization node; the management representation letter is linked to every node it covers.
  • Every audit procedure is linked to specific assertions. Each procedure is tagged with the assertion(s) it addresses (existence, completeness, valuation, etc.), creating a matrix of evidence coverage that can be traversed in either direction, from financial statement element to supporting evidence, or from assertion type to the evidence that addresses it across all elements.
  • The evidence graph mirrors the financial reporting scheme. By aligning the structure of the audit bundle with the structure of the financial reporting scheme (as defined by the Seattle Method's knowledge graphs), the audit documentation becomes a parallel structure to the closing book itself. Every node in the closing book has a corresponding cluster of evidence in the audit bundle, and the completeness of audit coverage can be assessed by checking whether every node in the closing book has adequate supporting evidence in the audit bundle.
  • Traceability is bidirectional. An auditor, regulator, or AI system can start from any financial statement number and trace backward through the audit bundle to find every piece of evidence that supports it and can start from any piece of evidence and trace forward to see which financial statement elements it supports. This bidirectional traceability is impossible in a document-based audit bundle.

The Transformation this Enables

The benefits of this transformation, of making a paradigm shift, of all information within this supply chain of  compliance reporting include:
  • Automated completeness assessment. The most labor-intensive aspect of audit quality management is ensuring that all required procedures have been performed and all required evidence has been obtained. An XBRL-based audit bundle can be automatically checked against a template derived from the audit methodology and the financial reporting scheme to identify gaps in evidence coverage. This transforms completeness assessment from a manual review exercise into a computational check.
  • Regulatory inspection readiness. When a PCAOB inspector or other regulator reviews audit documentation, they are essentially trying to determine whether the audit evidence supports the auditor's opinion. An audit evidence graph that is explicitly linked to the financial reporting scheme makes this assessment dramatically more efficient. The inspector can navigate directly to the area of concern, see all the evidence that supports the relevant financial statement elements, and assess its sufficiency without having to reverse-engineer the auditor's documentation structure.
  • AI-assisted evidence evaluation. With an audit evidence graph, AI systems can:
    • Assess whether the evidence obtained is consistent with the financial statement assertions being made
    • Identify areas where evidence coverage appears thin relative to the risk profile
    • Cross-reference evidence across entities and periods to identify patterns that may indicate audit quality concerns
    • Generate summaries of evidence coverage for audit committees and regulators
  • Institutional memory. One of the most underappreciated problems in auditing is the loss of institutional memory between periods. When audit team members change, the tacit knowledge of why specific procedures were performed, why certain judgments were made, and what issues arose in prior periods is often lost. An evidence graph preserves this knowledge in structured form; every procedure is linked to its rationale, every judgment is documented in context, and the historical evolution of the audit is navigable.

Audit Evidence Graph

The most powerful implication of  a digital closing book, a digital audit bundle, and digital financial reporting framework should not be seen as independent innovations; rather they are three different views of the same underlying graph which all use the same underlying architecture; the global open industry standard XBRL.
  • The Financial Reporting Framework (enhanced using the Seattle Method framework) defines the structural template; what a complete, coherent financial report looks like, what elements it contains, and how those elements relate to each other.
  • The Digital Closing Book is an instantiation of that template; a specific entity's financial data mapped onto the framework, with every node populated, every edge verified, and every constraint satisfied.
  • The Audit Bundle is a parallel graph; the same nodes as the closing book, but with each node linked to the evidence, procedures, and conclusions that support it.
These three structures are natively interoperable because they share the same underlying architecture, the XBRL-tagged financial statement, a.k.a. Digital Financial Statement. This creates a seamless chain. 

Every link in this chain is machine-readable, navigable, and auditable. There are no manual handoffs, no format conversions, no information losses. The same structural logic that defines what a financial report should look like also defines what the closing book should contain and what the audit bundle should cover.

The Implications for AI

This convergence is what makes AI truly effective in accounting and auditing. Consider the difference between the current state and the future state:
  • Current state: An AI system must independently parse 800 spreadsheets (closing book), a separate document management system (audit bundle), and the accounting standards (reporting framework); three disconnected information sources with no shared structure. The AI must infer the connections between them, which is error-prone and often impossible.
  • Future state: An AI system operates on a single, integrated graph in which financial data (closing book), supporting evidence (audit bundle), and structural logic (financial reporting framework) are all connected through explicit, machine-readable relationships. The AI can traverse the graph in any direction, checking consistency, identifying gaps, and generating insights with full knowledge of the underlying structure.
This is the architecture where both rules-based AI and LLMs deliver their maximum value, and it is an architecture that is only possible with global open industry standard graph structures such as the graph structures provided by XBRL. Simple, elegant,  powerful.  For more information see Practical Knowledge Graphs in Practice.

XBRL does not replace the LLM; it makes the LLM more reliable by separating the data access problem (structured, deterministic) from the analysis problem (unstructured, judgment-intensive). This separation is exactly what makes the combination of rules-based AI and LLMs powerful, and XBRL is the enabler.

XBRL is the data infrastructure that makes the entire AI-driven transformation of accounting and auditing practically viable at scale. The Seattle Method is the tested, proven framework and playbook that shows you how. Think information Legos.


Additional Information:

Comments

Post a Comment

Popular posts from this blog

Internationalized Resource Identifier (IRI)

Image
An Internationalized Resource Identifier (IRI) is a standard proposed by the IETF ( RFC 3987 ).  An IRI is a unique sequence of characters that identifies a resource.   IRIs are a generalization of URIs ( RFC3986 ) that permits a wider range of Unicode characters. A  Uniform Resource Identifier (URI) is a unique sequence of characters that identifies an abstract or physical resource.  The difference between an IRI and a URI is that IRIs use the more modern  Universal Character Set (a.k.a. UNICODE)  and URIs use the older ASCII character set. The newer IRI format is compatible with the older URI format. The most  current version of RDF  uses IRIs to identify resources. Don't confuse IRIs with URLs and URNs.  A URL is a web address. A URN is a Uniform Resource Name. This Venn diagram shows the relationship between these different tools: Why are IRIs and URIs important?  If you are thinking in terms of a single database, a silo ...
Read more

Example Financial Statement Holon

Image
As I have mentioned in other posts on my blog, a financial statement is a formal semantic structure , in fact a financial statement is a special type of formal semantic structure referred to as a holon . As I mentioned in my blog posts, Kurt Cagle and Chloe Shannon are providing a lot of very good information about holons on their substack. ( Here is Kurt Cagle's and Chloe Shannon's answers to my questions. Here is an example Kurt Cagle created of a financial statement as a DataBook.) I have provided a bunch of good examples of financial reporting frameworks and reports   In this blog post I want to focus on one specific financial statement and see if I can explain that a financial statement is a "container" of  blocks of information and the "information blocks" that the financial statement contains. Cagle and Shannon point out that a holon is made up of four distinct knowledge graphs that maximize coherence and operationalize the holon which are: The Sc...
Read more

Inhabiting Babel, A Manifesto for Responsible Meaning Engineering

Image
Inhabiting Babel, A Manifesto for Responsible Meaning Engineering is a book written by Nicolas Figay . This book is very practical and easy for a motivated business professional to read. For me, this book has both tempered my expectations as to what is possible and also reassured me of the real possibilities that exist if you do the required work. Accountants and auditors tend to be on the conservative side of the spectrum.  Me included. I am  actually rather conservative in terms of what technology might bring to the table for accounting, reporting, audit, and analysis. I have learned how to "cut the cards" when dealing with sales people and software vendors. What this book does is bring the wisdom from the experiences of an expert in system interoperability, enterprise architectures, and complex systems to helps motivated business professionals and technical professionals that truly want to understand artificial intelligence, cutting through the rhetoric and hype. This boo...
Read more