A finance team runs a quarterly report through an AI tool. The draft looks clean. It goes to the board. Then an internal audit finds the mistake, after the numbers already left the building.
That story is more common than most owners would guess. In Workiva's 2026 Midyear Executive Benchmark Survey of 2,272 finance, risk, and sustainability professionals, 26% of executives said internal audits detected AI errors that had reached external audiences or board members. About one in four.
The next finding should get your attention. In the same survey, 84% of executives said they were at least somewhat confident in AI output without human review. Only 11% believed their data quality was good enough to use for AI in the first place. High confidence sitting on top of thin evidence.
"Confidence in AI without control over data quality is a liability, not a strategy," said Barbara Larson, Chief Financial Officer at Workiva.
The owner bottleneck
In an owner-led company, one kind of work never really leaves your desk. It comes back every month and every quarter, and you are usually the last person to touch it before it goes out.
The recurring workflow is the reporting that leaves your company: the monthly financials, lender updates, investor notes, and board materials your team assembles from raw data.
You are the final read. You know which numbers move, which line items get questions, and which claims a partner or a bank will actually check. So the report waits for you. When you are busy, it waits longer. When you are traveling, it either ships without your read or it does not ship at all.
AI changes the speed of the draft, but it does not change who owns the claim. That is still you.
And the exposure is rising. The Conference Board and ESGAUGE found that 72% of S&P 500 companies disclosed at least one material AI risk in their 2025 filings, up from just 12% in 2023. Reputational risk was the most-cited concern, named by 38% of firms. Eleven companies went further and named hallucinations and inaccurate outputs as a specific hazard. Bigger companies are writing this risk into legal documents. Owner-led firms feel the same pressure, without the legal team.
of S&P 500 companies disclosed at least one material AI risk in their 2025 filings, up from just 12% in 2023.
Source: The Conference Board and ESGAUGE, 2025What this changes for the business
The outcome worth protecting is simple. Fewer wrong numbers reaching the people who fund you, review you, or bet on you.
Investors already treat this as a real risk. In the Workiva survey, 89% of institutional investors said they were concerned about AI accuracy in corporate disclosures. When almost nine in ten of the people reading your reports are worried about accuracy, a visible way to show your work is worth something.
Data quality is where the drafts break down. A Workiva survey found that 27% of executives said poor data quality had significantly blocked AI deployment in key workflows. And 71% said poor data quality had at least moderately affected their use of AI in financial and sustainability reporting. The tool is only as good as what you feed it.
So the honest role for a managed system is narrow and useful. It produces the first pass. It drafts, summarizes, pulls the numbers together, and gets you to a review-ready document faster. It does not own the final claim, and it does not get the last word before something goes to the board.
"Generic AI isn't enough for financial reporting," said Jason Darby, Chief Financial Officer at Amalgamated Bank.
Read that as a floor, not a ceiling. A system that drafts reporting for your business needs to be pointed at your systems of record, not a general model guessing at your books.
The tool is only as good as what you feed it.
Where people stay in control
The control point sits at one specific spot. Assurance.
A person verifies AI output against the systems of record before it becomes a board material or a public number. Not confidence in the tool. Verification against the source. That is the load-bearing step, and it stays human on purpose.
What that looks like in practice: the AI hands you a draft, and every figure that counts traces back to something you can open and check. The bank statement. The ledger. The signed contract. If a number cannot be traced, it does not go out.
This is also where a real shortfall shows up at the top. The Conference Board found that disclosed AI expertise among S&P 500 directors rose only from 1.5% in 2021 to 2.7% in 2025. Over the same period, disclosed technology expertise on those boards climbed from 20% to 51%. The rooms reviewing this work are still thin on people who understand it. In an owner-led company, that reviewer is often just you. Which makes the verification step more important, not less.
You are not becoming an AI engineer here. You are keeping the one judgment call that has always been yours: deciding whether a number is true enough to put your name on.
The next step
Take your most recent board or lender report. Pick the five numbers that would cause the most damage if they were wrong. For each one, ask a plain question: can I trace this back to a source I can open right now? If the answer is yes for all five, your review step is doing its job. If it is no for even one, that is where your verification process needs to be before AI drafts anything for you.
Sources
- One in Four Executives Say AI Errors Have Reached External Audiences or Boards (Workiva)
- AI Risk Disclosures in the S&P 500: Reputation, Cybersecurity, and Regulation (Harvard Law / The Conference Board, ESGAUGE)
- Governing AI in the Corporation (The Conference Board)
Frequently Asked Questions
Does this mean AI cannot be trusted with reporting?
It means AI can draft reporting well and should not sign off on it. The survey findings say as much. Plenty of confidence, not much verified data quality underneath it. Use the draft. Keep the review.
We are small. Do board-level findings even apply to us?
The reader changes, the risk does not. Swap "board" for your lender, your investors, or the partner who audits your numbers. A wrong figure reaching any of them costs trust you spent years building.
Where do we start if data quality is the weak point?
Start with the reports that leave the building most often. Pick one. Map where each number comes from and confirm you can trace it to a system of record before AND after AI touches it.
Bring us the work that keeps coming back to you.
We will trace the workflow, identify where a system can create capacity, and decide what should remain with your people.