AI Accuracy Confidence Gap: What Finance Firms Must Address in Crypto Accounting Software
Eighty-four percent of finance, audit, and sustainability executives say they trust AI-generated outputs in annual reports, even without a human reviewer in the loop. That figure, from a survey published by Workiva in August 2026, sounds reassuring until you read the next data point: 26% of those same executives report that internal audits have already detected AI errors that reached the board or external audiences. For any accounting firm or CFO deploying crypto accounting software, this confidence gap is not background noise. It is a direct operational and reputational risk that deserves immediate attention.
What the Survey Actually Found
The Workiva report drew on responses from executives across finance, audit, and sustainability functions. The headline finding is that 84% expressed confidence in AI accuracy, split between those who are "somewhat confident" (39%) and those who are "very confident" (45%). On its own, that might suggest AI adoption in financial reporting is maturing steadily.
The contradiction inside the data
The problem is that confidence does not appear to be grounded in strong internal controls. The same cohort provided a second data point that sits in direct tension with their stated confidence: more than one in four said their own internal audit function had caught AI-generated errors that had already made it to the board table or to external stakeholders. Junko Swain, Chief Accounting Officer at Workiva, described that figure as setting off alarms, noting that AI errors are not hypothetical risks but an actual control problem organisations must address now.
Data readiness: the root cause
Swain pointed to a related finding that helps explain the contradiction. Only 11% of respondents felt their own organisation's data was genuinely AI-ready, and 71% said poor data quality has at least moderately impacted AI outputs in financial and sustainability reporting. The implication is clear: AI does not correct bad inputs. It processes them quickly and produces outputs that look credible even when the underlying data is flawed. In Swain's words, AI often looks polished even when the inputs are wrong, and generic tools typically cannot trace numbers back to their source, making errors harder to catch and harder to explain after the fact.
A Pattern, Not an Outlier
The Workiva findings do not stand alone. Several other surveys from 2025 and 2026 point to the same structural tension between stated confidence and observed reality.
Governance maturity versus incident rates
A separate report from a Top 50 accounting firm found that 90% of AI governance professionals said their organisations had already allocated dedicated funding for AI governance, and 74% believed they could pass an AI compliance audit today. Yet only 27% described their governance programmes as fully mature. The consequence was visible: 65% of respondents said their organisations had experienced AI-related incidents or near-misses over the preceding 24 months.
Data quality costs and delayed reporting
Research from corporate performance management provider OneStream added financial texture to the problem. It found that 79% of executives believed their data governance could support large-scale AI adoption, but 61% admitted to second-guessing their data at least once a month, and 11% questioned it daily. Seventy-two percent reported that bad data had cost their organisation $500,000 or more, with more than a third reporting losses exceeding $1 million. The downstream effects included delayed financial closes (cited by 44% of respondents), lost revenue opportunities (41%), reduced trust in automated insights (38%), and compliance issues (35%). A further survey of US business leaders found that 88% were confident in the accuracy of data feeding their AI systems, yet 69% had discovered errors in document-derived data at least sometimes.
Taken together, these surveys describe an industry-wide pattern: stated confidence in AI governance consistently outruns actual governance maturity, and the gap produces real errors in real reporting cycles.
Why This Is Amplified for Crypto Accounting Software
The confidence gap matters in any financial reporting context. It carries extra weight in the digital asset space for three specific reasons.
Volatile and complex source data
Crypto transactions generate source data that is far more complex than traditional financial records. A single portfolio might span dozens of chains, multiple wallet addresses, decentralised exchange interactions, staking rewards, and airdrops, all requiring classification under standards such as ASC 350-60 or IFRS guidance. If only 11% of organisations consider their general financial data AI-ready, the proportion with crypto-specific data that meets that bar is almost certainly lower. Any AI layer sitting on top of incomplete or misclassified on-chain data will amplify the underlying inaccuracies, not correct them. For more on the unresolved terrain under current standards, see our analysis of crypto accounting under ASC 350-60.
Regulatory scrutiny is intensifying
The SEC's proposed Reg Crypto rulemaking and ongoing standard-setting activity mean that digital asset disclosures are moving closer to the centre of regulatory attention. An AI error that reaches an external audience or the board in a traditional reporting context is damaging. The same error in a digital asset disclosure during a period of active rulemaking — where regulators are looking closely at whether firms have adequate controls — carries a materially higher compliance risk. Firms tracking that regulatory trajectory should also review our coverage of the SEC Reg Crypto proposed rulemaking.
Data lineage is harder to establish
Swain's point about generic AI tools lacking the ability to trace numbers back to their source is particularly acute for digital assets. On-chain data is pseudonymous and distributed. Establishing a clean audit trail from a wallet transaction to a financial statement line item requires deliberate data architecture, not just an AI query layer. Without that lineage, an auditor cannot efficiently validate AI-assisted outputs, and a firm cannot confidently stand behind a disclosure.
Practical Steps for Accounting Firms and CFOs
The survey findings point to specific actions rather than general caution.
Treat data governance as a prerequisite, not a parallel workstream
Swain's framing is worth internalising: data governance is not a back-office concern, it is the foundation. For firms using digital asset accounting software, that means mapping data ownership clearly before expanding AI-assisted workflows. The key questions are whether the data feeding any AI tool is accurate, who is authorised to consume it, and whether a clear lineage exists from source to output. If those questions cannot be answered confidently, the firm does not yet have the controls in place to stand behind AI-assisted disclosures.
Build mandatory human review checkpoints
The fact that 26% of audit functions are catching errors that have already reached external audiences suggests review checkpoints are either absent or positioned too late in the workflow. Firms should require a qualified reviewer to sign off on any AI-assisted output before it leaves the finance function, not after it reaches the board pack or a draft filing. For crypto bookkeeping software specifically, that review should include a line-level check that transaction classifications are consistent with the firm's stated accounting policy.
Audit the AI workflow, not just the output
One finding from the separate governance report is particularly instructive: the reason AI processes often fail audits is not architectural but procedural. Auditing a process the organisation has not formally defined is impossible. Firms should document the AI-assisted steps in their crypto reporting workflow, assign responsibility for each step, and include those steps in the scope of internal audit. That documentation also provides a defensible record if a regulator or external auditor raises questions.
Assess vendor capability on data lineage
When evaluating or renewing crypto accounting software, firms should ask vendors directly how their tools trace AI-generated outputs back to source transactions. Generic large-language-model-based tools that cannot provide that lineage are not appropriate for regulated financial reporting. Specialised platforms built for digital asset accounting should be able to demonstrate a clear chain from raw on-chain data to a financial statement entry.
The Governance and Accountability Dimension
The survey data suggests that the gap between perceived and actual AI governance is not primarily a technology problem. It is a governance and accountability problem. Executives are confident in the abstract, but that confidence does not translate into mature controls at the organisational level. Swain's framing captures it precisely: leaders must stop asking only whether AI is accurate in general and start asking whether they can stand behind this specific output, with this specific data, in front of this specific audience.
For accounting firms advising clients on digital asset reporting, this is also a professional liability question. If a firm uses AI-assisted crypto accounting software to prepare or review financial statements or tax disclosures, and an error reaches a regulator or external auditor, the question of whether adequate human oversight was in place will be central to any enforcement or disciplinary review. The Workiva data suggests the profession as a whole has not yet answered that question satisfactorily.
The misinformation risk perception data reinforces the governance point. The proportion of respondents identifying misinformation as a top external threat rose from 24% to 31% in the US between survey periods, and from 23% to 26% globally. Executives recognise the risk at the macro level. The work now is to build internal controls that match that recognition at the organisational level.
FAQ
What does the 26% figure mean for an accounting firm's audit risk?
It means that in more than one in four organisations surveyed, AI-generated errors bypassed internal controls and reached either the board or external stakeholders. For an accounting firm, that represents a direct professional liability risk. If AI-assisted outputs in client engagements are not subject to documented human review, the firm may not be able to demonstrate that it exercised appropriate professional judgement.
How does poor data quality affect crypto accounting software specifically?
Crypto transaction data is structurally complex. Missing cost-basis records, unclassified token types, and incomplete wallet mappings are common. AI tools that process this data will generate outputs that sound plausible but may misclassify transactions, misstate fair values, or omit taxable events. The Workiva finding that only 11% of organisations consider their data AI-ready makes this a near-certain issue for most firms without a dedicated data preparation layer.
What is data lineage and why does it matter for digital asset disclosures?
Data lineage is the documented trail that connects a financial statement figure back to its source transaction. For digital assets, that means tracing a reported fair value or gain figure back to a specific on-chain transaction, through any classification or conversion steps, to the final disclosure. Without that trail, neither an internal auditor nor an external reviewer can efficiently validate AI-assisted outputs, and the firm cannot defend its figures to a regulator.
Should firms pause AI adoption in crypto reporting until governance matures?
The Workiva findings do not suggest pausing adoption. They suggest sequencing it correctly. Governance infrastructure — data ownership mapping, lineage documentation, and mandatory human review checkpoints — should be in place before AI tools are applied to regulated outputs. Firms that have already deployed AI without that infrastructure should treat closing the gap as an urgent control remediation, not a future project.
How should a CFO respond if an AI error has already reached the board or an external audience?
The immediate steps are to identify the specific output, trace it back to the source data to understand how the error arose, assess whether any regulatory disclosure or financial statement is affected, and determine whether a correction or supplemental disclosure is required. The broader remediation is to document what control failed, implement a review checkpoint to prevent recurrence, and brief the audit committee on both the error and the remediation plan.
Source: Accounting Today