The data layer AI actually needs: what Engage told us about the next phase of audit
Tomorrow's Audit

The data layer AI actually needs: what Engage told us about the next phase of audit

The conversation at AICPA Engage 2026 was different this year. Not because AI was on every agenda. It has been for two years. The difference was the tone.

Last year the energy was speculative. This year it was urgent. And the question underneath every session was the same: when every vendor is promising AI for audit, how do practitioners know what actually works?

The answer is not in the AI. It is in the data underneath it.

Want the short version?
See how Validis delivers AI-ready audit data

Or read on for the full breakdown…

From buzz to strategy: what Engage confirmed about where audit firms actually are

The recognition of AI’s importance is near-universal. The Thomson Reuters Future of Professionals Report 2025, drawing on more than 460 tax, audit, and accounting firm professionals, found that 80% believe AI will have high or transformational impact on their work.The gap is not belief. It is execution: only 14% of those firms have a defined AI strategy in place.

Most firms are somewhere between awareness and action, overwhelmed by the pace of change and afraid of making the wrong decision. Their technology change cycle used to run once a decade. AI is running on a different clock. The firms that have done nothing beyond a general-purpose AI assistant are already considered behind. The session rooms with the most energy were the ones that stopped asking whether to adopt and started asking how.

Firms with a visible AI strategy are more than three times as likely to realize positive ROI compared to those without one, according to the same Thomson Reuters research. The implementation gap is not just a competitive disadvantage. It is a compounding one.

 

The AI-native firm: what the direction of travel actually looks like

The standout framing at Engage this year was the AI-native firm. Not AI as a tool bolted onto existing workflows, but AI as an operating model built into the firm’s structure from the ground up.

Fieldguide’s presentation on this theme addressed new roles that a firm needs to operate at this level: agent orchestrator, AI skills coach, governance and quality lead. And it made a point that resonated across the floor: adoption without governance is not a strategy. It is a risk.

The AI-native framing will become the new benchmark. Firms not moving toward it now will not just be behind their competitors. They will be visibly so.

Augmentation, not replacement: the workforce argument audit needs to hear

The macro view at Engage was direct: AI is a tool for augmentation, not replacement. The accounting profession has faced a structural talent and capacity crisis for years. AI addresses it not by cutting headcount but by removing the work that was consuming people, and by raising the quality ceiling for the staff who remain.

Goldman Sachs CEO David Solomon stated publicly in May 2026 that fears of an AI-driven jobs wipeout are overblown. The more likely outcome is an evolution in what talent means: rising demand for professionals who can bridge domain expertise and AI implementation. The audit and accounting profession is particularly well placed. The work that makes accountants valuable, judgment, client relationships, complex analysis, is precisely what AI does not replace.

The firms winning right now are not the ones automating the most. They are the ones using AI to extend what their existing people can do.

 

Why AI in audit keeps failing at the data step

Large language models are probabilistic. Given the same input twice, they can produce different outputs. For drafting, summarizing, or researching, that variability is acceptable. For audit, it is not.

Audit requires determinism. Every procedure needs to trace back to source data. Every conclusion needs to be reproducible. Every step needs an audit trail. When AI operates on inconsistent or unstandardized financial data, those requirements break down.

Multi-step AI workflows compound the problem. Where one model’s output feeds the next, errors amplify at every step. If the data entering the first step is not clean, structured, and standardized, the degradation accelerates across the entire workflow.

This is not a failure of AI capability. It is a data foundation problem. Research on AI in audit data analysis confirms that the core challenge is ensuring AI operates on clean, structured inputs with human oversight maintained throughout.

See how Validis solves the data foundation problem

 

What ‘AI-ready’ actually means for audit data

AI-ready financial data has four properties that general accounting exports do not reliably provide.

Standardized across sources: A trial balance from QuickBooks and a trial balance from Sage do not share the same chart of accounts, field names, or data structure. An AI agent querying both without a common data model produces incomparable outputs. Standardization across accounting systems is the precondition, not an optional feature.

Complete at transaction level: Audit AI workflows need GL, AR, and AP at sub-ledger depth. Summary-level exports hide the transactions where risk lives. Risk-focused sampling, anomaly detection, and automated procedures all require the underlying population.

Traceable to source: Every data point needs clear lineage from the source accounting system. When a procedure flags a variance or an AI agent surfaces an anomaly, the auditor needs to trace that finding directly back to the originating transaction. Without lineage, the output is not auditable.

Never used to train AI models: Audit clients’ financial data is sensitive. Data used in audit workflows must not feed model training. This is a governance requirement, not a preference.

 

MCP: the infrastructure shift audit firms need to understand now

Model Context Protocol (MCP) is becoming the standard way that AI agents connect to external tools and data sources. It is a universal interface through which an AI can query a database, retrieve a document, or pull structured financial data, regardless of where that data lives.

For audit, the practical implication is significant. Audit AI workflows today require data exports, reformatting, and manual import steps before AI can engage with the data. MCP removes those steps. An AI agent running inside Caseware, DataSnipper, or a firm’s internal tooling can query standardized financial data directly through a permissioned, auditable connection.

The audit workflow no longer starts with a data preparation step. It starts with analysis. Every subsequent AI procedure operates on the same standardized, source-connected data. The audit trail is built in.

CPA Practice Advisor’s Accounting Technology Lab described MCP in February 2026 as one of the most important AI developments for accounting firms, noting that every connection is permissioned and auditable, and that firms control exactly what data AI can see or act on. Firms evaluating AI tooling in the second half of 2026 will increasingly ask their vendors one question: do you support MCP?

Validis MCP is live

 

The hybrid model: deterministic extraction and LLM intelligence

There is a false choice in parts of the current AI-in-audit conversation: either use deterministic rule-based systems, reliable but limited, or use LLMs, capable but unpredictable.

The firms moving fastest are not choosing. They are combining.

Think of it as fuel for an engine. You can build the most sophisticated engine in the world, but if the fuel is inconsistent in quality or composition, the engine will not run correctly. Deterministic data extraction is the fuel: pulling complete, structured, standardized financial data directly from accounting systems. LLM intelligence is the engine: risk stratification, anomaly detection, engagement documentation, procedure guidance. The AI does what it is actually capable of, because the data step is already done.

Auditors get AI-generated insights they can trust, because those insights are grounded in data they can trace. As Thomson Reuters and Validis have demonstrated through their Audit Intelligence partnership, clean structured data from accounting systems is what enables AI to surface risk, automate documentation, and support planning at scale.

 

Validis is the start of the process, not the end

Every AI tool an audit firm evaluates, whether it is Fieldguide, Caseware, or an internal build, needs clean, standardized, complete financial data to function correctly. That need does not diminish as AI capability increases. It intensifies. The better the AI, the more it exposes the quality of the data underneath it.

The Caseware partnership with Validis reflects exactly this: audit-ready data from diverse sources, clean and available for analysis, delivered to the tools firms already use.

Firms that standardize their data foundation now are not just solving a current problem. They are building the infrastructure that makes every future AI investment work as intended.

And the barrier to getting started is lower than most firms expect. A Validis portal can be live and connected to a client’s accounting system in under 24 hours. White-glove onboarding is in place throughout. The data layer is not the hard part. Waiting to build it is.

 

What Validis provides in this architecture

Validis connects directly to over 100 accounting systems, cloud and on-premise. GL, trial balance, AR, and AP are extracted at transaction level, standardized to a common audit data model, and delivered to the tools audit teams already use: Caseware, DataSnipper, Excel, and internal LLMs.

The same data layer that powers manual audit procedures powers the AI workflows being built on top of them. Read-only connections. Encrypted in transit and at rest. Never used to train AI. Permissioned, traceable, and tied back to source.

When an audit AI agent queries financial data through Validis, it gets the same standardized, complete dataset that the audit team uses for their own procedures. One source of truth. One data model. Across every client engagement.

Ready to build your data foundation?

 

Resources

Book your demo today.