The DoD’s AI Auditor: Why the Time for a New Approach is Now
The AI Audit Operating System Thinks Like an Auditor, Acts like a Machine.
In a dangerous world, any Secretary of Defense should be able to get an immediate answer to a simple question: How much of each munition do we have right now—and where are they? This seemingly straightforward question has enormous implications for deterrence, resupply, and decision-making. Yet it triggers a cascade of emails, spreadsheets, and manual data pulls from dozens of disconnected systems, creating a strategic vulnerability that harms our ability to make time-sensitive decisions about force deployment, supply chain resilience, and resource allocation.
Now instead of artillery shells and missiles, think about dollars. The same problems that prevent us from tracking shells from the factory to the firing line plague the Department of Defense’s (DoD) financial statement audit: fragmented systems, disconnected processes, and zero real-time visibility into how money moves, what has been bought, and what cash is on hand. These problems have caused DoD to flunk seven audits in a row.
For decades, the DoD has attacked this problem with conventional solutions: setting aside billions for plans to consolidate financial and business platforms, developing data standardization frameworks, and hiring armies of consultants. While well-intentioned, this approach has delivered incremental progress at best. Already overworked staff scramble to fulfill requests with inadequate tools. Senior leaders get inadequate answers in return. Bad decisions get made on the basis of inadequate information. The cycle continues.
But what if that question didn’t trigger a chain of emails and spreadsheets? What if it triggered an AI agent? More precisely, an AI Auditor: a persistent, intelligent software agent built to continuously monitor, reconcile, and reason over every financial transaction and business event across the department. Unlike dashboards or static reports, the AI Auditor doesn’t wait for quarterly reviews. It lives inside the system. It understands how purchase requests become obligations, how invoices connect to disbursements and general ledger posting logic, and how financial data relates to operational reality as well as a financial statement line item. It flags anomalies before they become findings, drafts corrective actions, makes the fix before IGs issue memos, and learns over time from the best source of insight in the DoD: the people who do the work.
The DoD doesn’t need more dashboards. It needs decisions and action. It needs an AI audit operating system—one that can think like an auditor, scale like software, and work nonstop to get the DoD to pass a financial statement audit by 2028. To build that, the focus must shift away from adding more consultants and toward data integration and automation.
I helped build Advana with a small team in 2018 to unify the DoD’s fragmented business data. The initial goal was to ingest and standardize financial and business data to provide auditors with something usable—a “universe of transactions”—even if imperfect. Advana grew quickly, ingesting data across business systems, enabling commercial software, and becoming the backbone of financial and performance reporting. The initial results were promising. Lt. Gen. James Adams, Deputy Commandant for Programs and Resources, testified that the Marines’ unmatched and undistributed transactions dropped from $4.2 billion to just $200-$300 million because of Advana.
But while Advana integrated data and provided unprecedented visibility, it revealed a deeper truth: data aggregation alone cannot solve the audit. The problem extends beyond seeing the data—it requires understanding what it means in operational context and automatically taking corrective actions when problems arise. Aggregating data in one place improved transparency, but it didn’t encode the audit logic—the way transactions flow, the controls they pass through, or how audit findings should be automatically corrected. That requires reasoning, structure, and intelligent automation. Another key limitation of Advana was its inability to deploy tools that collected missing data directly from operators when no authoritative system existed—the kind of frontline inputs needed to resolve root-cause audit issues like unrecorded inventory, unmatched property transfers, or undocumented asset receipts. To truly solve the audit, Advana must evolve into an authoritative transactional system—not just aggregating existing data, but capturing it in an auditable, structured way as it’s generated by the operators.
Here’s the blunt reality: if the DoD wants to pass a financial statement audit by 2028, we don’t have time to retire legacy IT systems (the DoD has over 1,700 business IT systems). We don’t have time to reengineer processes. We should not hire more consultants or system integrators. And we definitely don’t have time to wait for the next monolithic system plan to fail. Each year, the audit adds new findings. Corrective actions stretch out. Audit teams rotate. Institutional knowledge disappears. The backlog grows (over 80% of the 2024 financial audit findings were audit issues found in previous years’ audits).
This isn’t just about financial compliance. Each finding is a fundamental impact to readiness, resource planning, and operational risk. Even successful audits of the Marines and DLA came at enormous cost. Thousands of Marines and civilians manually pieced together data that should have been integrated and automated from the start. That was heroic work, but heroism is not a scalable audit strategy. We need a paradigm shift to achieve a clean audit opinion at the department level, with complexities dwarfing those of the Marine Corps and DLA. Only then will we unlock the full operational value of a successful audit.
This is why Advana must now evolve into an AI audit operating system. The foundation is already there—the data, the platform, the users. The next step is encoding how auditors and managers think: the rules they follow, the exceptions they know, the risks they intuitively spot. That knowledge must become machine-readable and continuously applied to fix audit findings automatically. The kind of AI agent-based architecture needed to do that simply didn’t exist when Advana was born. Now it does.
Here’s how the AI Auditor works:
Data Integration With Operational Context: The AI Auditor consumes every financial and business transaction across systems—from budget execution to procurement to payroll, with the plans and justifications behind all spending—leveraging the vast data already ingested and maintained within Advana. It connects every dollar spent to mission objectives, strategic plans, and operational requirements. This means it can flag not just accounting discrepancies but also misalignment between spending and strategic priorities.
Ontology Creation: At the core of the AI Auditor is a dynamic framework known as an ontology—a living map of the audit’s key entities, relationships between financial elements, and the actions auditors and management take to verify them. The ontology captures both entities and relationships: funding authorizations flowing to contract line items, becoming obligations, generating invoices and receiving reports, triggering disbursements, and ultimately appearing in financial statements. Each connection creates a traceable path from appropriation to battlefield impact. What distinguishes this approach from traditional data models is its ability to encode expert actions. The ontology doesn't just represent what exists—it captures what should happen when discrepancies arise. It models the decision logic of experienced auditors: which exceptions matter, which patterns indicate risk, and which corrective actions resolve specific issues. By connecting transactions, business events, policies, people, and systems, it transforms fragmented data into meaningful, machine-readable intelligence. This enables AI agents to reason over financial information with auditor-level expertise, automatically resolving issues that previously required manual intervention. That's how static audit trails become dynamic audit intelligence.
Automated Reconciliation: The AI continuously checks every line item and business event against the general ledger posting logic and business feeder systems, flagging mismatches and missing data or evidence based on rules and learned patterns.
Anomaly Detection: Algorithms identify patterns, pinpoint potential fraud, pinpoint compliance issues, and surface high-risk transactions for immediate review.
Corrective Action: Instead of waiting for quarterly reports, the AI Auditor issues real-time alerts and fixes issues automatically. The assertion of existence and completeness now becomes a metric that is tracked daily.
Audit Trail: Every action, every transaction, every anomaly is logged and tracked, creating an indisputable audit trail based on a unified structure.
Human-AI Teaming: The AI Auditor works alongside human auditors and management. Auditors and management review flagged transactions, validate or override recommendations, and provide feedback that continuously improves the system. Over time, the AI learns individual agency policies, common exceptions, and auditor preferences—not just from user interactions, but by reading and interpreting the hundreds of training documents, audit guides, policies, and app-specific PDFs scattered across the DoD. It captures and scales institutional knowledge that would otherwise remain siloed or disappear altogether.
Data Collection When No Authoritative Source Exists: The AI Auditor doesn’t just work with existing data—it identifies where no authoritative data source exists and actively collects what’s missing. Whether it’s a receiving report, a property transfer, or asset condition data, the AI agent prompts end users at the point of action to enter what’s needed, creating structured, auditable records in real time. This is particularly vital for inventory and property, where the Army is already piloting efforts to digitize frontline data capture for items that were never properly recorded. The AI Auditor scales that effort by turning human-in-the-loop inputs into authoritative, audit-ready data—filling critical gaps that legacy systems were never designed to close.
Multi-Agent Environment: The system operates as a network of specialized AI auditor agents, each assigned to specific financial and material weakness areas like payroll, contracts, asset management, and budget execution reconciliation. These agents work independently but share a common operating environment, cross-referencing findings and escalating potential risks to higher-level agents or human auditors/management.
Advana 2.0 (or should I say, AdvanAI) isn’t a data lake or a dashboard. It’s a full-fledged AI audit operating system—one that would transform the DoD from a reactive audit preparer into a proactive financial control engine. Built on a foundation of ontology and intelligent AI agents, this system addresses the root causes of persistent audit failures by identifying data quality issues at their source, pinpointing process breakdowns during system handoffs, highlighting unclear or contradictory policies, and adapting to rapidly evolving organizational structures. Unlike traditional approaches that simply flag discrepancies, AI Auditor incorporates rules, procedures, doctrine, and organizational context into its knowledge base, enabling it to explain anomalies and target likely issues for immediate resolution. This depth of understanding—far beyond typical financial review—transforms raw data into structured intelligence that AI agents can reason with and act upon, turning what would be educated guesses into expert-level financial oversight and automated corrective action.
The value of a clean financial statement audit opinion is decision advantage. It’s resourcing the warfighter faster. It’s visibility, accountability, and speed. The DoD doesn’t need more systems or consultants. It needs better thinking and action—at scales that can only be achieved by machines. When commanders have accurate, real-time information about resource availability, they can optimize deployment schedules, redirect assets to emerging priorities, and ensure forces never lack critical supplies in contested environments. Funds trapped in accounting errors become immediately available for modernization efforts and emerging capabilities, while improved visibility into assets and inventories strengthens readiness postures across all domains.
We won’t get there with spreadsheets. We won’t get there with dashboards. We won’t get there with more consultants. We'll get there with an AI Auditor working 24/7, learning, flagging risk, enforcing discipline, and automatically fixing issues before they impact mission execution. This system watches all, remembers all, and never lets a dollar go unaccounted for, making the audit continuous rather than episodic. It doesn't rip and replace legacy systems or ERPs—it rides above them, providing immediate value without disrupting operations. And it doesn't wait—it works now, enabling faster acquisition cycles, reducing administrative burden on operational units, and providing leaders with confidence in their resource decisions during competition, crisis, and conflict. That's what the warfighter deserves. That's what the taxpayer demands. And that's what the DoD can finally deliver—if we deploy it now.