Business Reporting Statistics

A curated page of the statistics we reference most often when discussing reporting – the burden it places on teams, the quality of the data behind it, the cost of poor data, and the effect of AI on report production. Every figure is sourced, dated and, where the source is a vendor benchmark, accompanied by its sample size.

This page exists because reporting statistics circulate widely and are often repeated without a source or a date. Where we could not attribute a figure precisely, we have left it out. Use these figures freely, and cite them with the attribution given.

Figures change. A statistic is a snapshot of a source at a date, not a settled fact. Always cite the date.

How to use this page

• Cite the source and the date, not just the number.
• State the sample size where the source is a survey, especially a vendor survey.
• Check the date before using a figure in a current argument; treat anything over two years old as historical.
• Distinguish vendor benchmarks from independent research, and label them accordingly.

We update this page when a source publishes a new edition. Where a figure has been superseded, we replace it rather than keep both.

The burden of reporting

The most common question about reporting is how much time it consumes. These figures are the clearest answer available.

Finance teams and manual work

51% of the finance week is spent on manual work such as reconciliation and report stitching.

Source: Intuit Enterprise Suite, Future of Finance 2026 Report (survey of 2,000 CFOs, controllers and VPs of Finance at US businesses over $2.5M revenue, May 2026).

This is the headline figure for reporting burden. It means that in a typical finance function, around half the available time goes to assembling and checking rather than analysis or decision support – and most of that effort sits in recurring reports.

Analysts and data preparation

78% of analysts’ time goes to data preparation, validation and tool navigation, rather than insight.

Source: dbt Labs and Quietly, The Analyst Revolution (Harris Poll, 2026).

The pattern is the same as in finance: the analytical workforce spends the majority of its time preparing data to be reported on, not interpreting it. It is a vendor-commissioned study, so the figure is best read as a directional benchmark.

Senior staff and manual data work

36% of senior finance staff spend 31-50% of their time on manual data work, and 40% report a close that takes seven days or more.

Source: Financial Education & Research Foundation / Financial Executives International, September 2026.

This gives reporting burden a second dimension: the seniority of the people affected. Time spent on manual data work at senior level is time not spent on judgment.

Data scattered and unreachable

70% of finance leaders say their critical data is scattered, with no single source of truth, and 57% say time-sensitive action has been missed because of delays in getting data.

Source: Intuit Enterprise Suite, Future of Finance 2026 Report (2,000 finance leaders, May 2026).

These two figures explain why reporting is slow. The data exists but is fragmented, and the delay has a consequence beyond the report itself: decisions are taken late, or not at all.

Data quality

Data quality is the least visible and most expensive input to reporting. These figures quantify it.

The cost of poor data quality

The average annual cost of poor data quality is estimated at $12.9 million per organization.

Source: Gartner (2020), based on a survey of 154 customers.

This is the most cited figure on the cost of poor data quality, and it is worth reading carefully: it is an average across organizations, and it includes costs well beyond reporting – in operations, compliance and lost opportunity. For a reporting argument, it establishes the order of magnitude.

Data quality as a top concern

43% of chief operating officers rank data quality as their top data challenge, and more than 25% report losing over $5 million a year to poor data quality.

Source: IBM Institute for Business Value (2025).

That senior operators rank data quality above every other data challenge is the strongest available evidence that it is a business problem, not a technical one.

Errors in newly created records

47% of newly created data records contain at least one critical error, and only 3% of records meet basic quality standards.

Source: Nagle, Redman and Sammon, “Only 3% of Companies’ Data Meets Basic Quality Standards,” Harvard Business Review (2017).

The second figure is the more striking: a single-digit share of records meeting basic standards means that error is the default state of new data, and correction is the norm.

Time lost to finding and fixing data

Knowledge workers spend around half their time finding, correcting and assembling data.

Source: Redman, “Bad Data Costs the U.S. $3 Trillion Per Year,” Harvard Business Review (2016).

Redman describes this as work in “hidden data factories” – the unglamorous, unmeasured effort that precedes every report. It is the reason a report takes days rather than hours.

Spreadsheet risk

A large share of reporting runs through spreadsheets, and spreadsheet error is well documented.

Spreadsheet error rates

Spreadsheet error research has found error rates as high as 94% in audited spreadsheets.

Source: widely cited research on spreadsheet error by Ray Panko (University of Hawaii).

The 94% figure is often quoted as “94% of spreadsheets contain errors.” Read precisely, it reflects error rates found in audited spreadsheets in academic studies – high enough to establish that unaudited spreadsheets are a risk, not proof that every spreadsheet is wrong.

Confirmed errors in real spreadsheets

A review of 25 spreadsheet audits found 117 confirmed errors, the largest with an impact of over $100 million.

Source: Powell, Baker and Lawson (2009).

This is the concrete case behind the error-rate research: real, audited spreadsheets, with errors that were confirmed and, in one case, materially expensive.

Why it matters for reporting

Spreadsheet risk is not an argument against spreadsheets. It is an argument for controls: a check before issue, a version record, and a named owner. See Data-Quality Checks for Report Production.

Decision-making and reporting

Reporting exists to support decisions, and the figures on decision delay are a direct measure of its cost.

Decisions taken late

57% of finance leaders report that time-sensitive action has been missed because of delays in getting data.

Source: Intuit Enterprise Suite, Future of Finance 2026 Report (2,000 finance leaders, May 2026).

Repeated deliberately, because it is the figure that turns reporting from an efficiency question into a decision question. A late report is not just expensive to produce; it changes what the business does.

Reporting without a single source

70% of finance leaders report no single source of truth for critical data.

Source: Intuit Enterprise Suite, Future of Finance 2026 Report (2,000 finance leaders, May 2026).

Where there is no agreed source, every report is a small reconciliation project, and discrepancies between reports become routine. See What Is a Single Source of Truth?.

AI in reporting

AI is now widely adopted for drafting, and the figures on both adoption and error are relevant.

Adoption and limited use

77% of finance leaders report no or limited use of AI in their finance processes.

Source: Intuit Enterprise Suite, Future of Finance 2026 Report (2,000 finance leaders, May 2026).

Adoption in reporting lags adoption in general knowledge work, which is unsurprising given the accuracy requirement. It is also where the opportunity sits: the mechanical parts of report production are the easiest to assist.

Hallucination rates

Reported hallucination rates in large language models range widely – from roughly 22% to 94% depending on the task and the measurement method – with one benchmark finding 13.6% of responses grounded.

Source: Stanford HAI, AI Index 2026, together with Vectara’s hallucination leaderboard and OpenAI’s own model documentation (2025-2026).

The range is wide because the measurement is genuinely difficult. What matters for reporting is the implication: no generative model output can be published without human verification of every figure and claim.

Model-specific error rates

Published evaluations have reported hallucination rates of around 33% for one reasoning model and around 48% for another on specific benchmarks.

Source: OpenAI model documentation and independent evaluations, as reported in 2025-2026.

These figures are model- and task-specific and should not be generalized to all AI use. They are included because the honest answer to “is AI accurate enough for reports?” is “not without verification.”

Governance frameworks

• NIST AI Risk Management Framework – a US framework for managing AI risk, including the risks of AI-assisted content.
• ABA Formal Opinion 512 (American Bar Association) – guidance on the professional obligations of lawyers using generative AI.
• ICMCI Code of Ethical Conduct (International Council of Management Consulting Institutes) – professional conduct expectations for consultants, including technology use.

These frameworks are the reference points for a report governance policy. See Report Governance for AI.

Confidential data in AI tools

26.4% of employees have pasted confidential company data into a generative AI tool.

Source: Harmonic Security (2025 study).

For reporting, the implication is direct: a report produced with AI may carry confidential figures into a third-party service. Policy and controls are the response, not prohibition alone. See Confidential Data in Report Production.

Adjacent: proposal and bid reporting

Proposal production is a sibling reporting workload with its own published benchmarks.

Proposal effort and outcomes

Among 1,500+ proposal teams surveyed: 79% have adopted generative AI in proposal work, average win rates sit near 45%, and a typical proposal consumes 25-41 hours of work, with roughly 55% of pursuits reaching submission.

Source: Loopio, 2026 RFP Trends Report (survey of 1,500+ proposal professionals).

These figures are for proposal reporting specifically, and they vary widely by sector. They are included because the structure of the problem – a recurring, deadline-driven, evidence-heavy document – is the same as the reporting problem.

See The RFP Win Desk for how the same production discipline applies to bids.

How to cite these figures

A short citation format that keeps you honest:

• “According to [source], [figure] ([date]).”
• Add the sample size where the source is a survey: “(survey of 2,000 finance leaders).”
• Label vendor research as such: “a 2026 vendor-commissioned survey of…”
• Do not re-round a figure upward; a “47%” is not “nearly half” in a precision-sensitive context.

If a figure has no source you can name, do not use it. That rule alone would remove most of the unreliable reporting statistics in circulation.

Frequently asked questions

What percentage of finance time is spent on manual work?

51%, according to a May 2026 Intuit survey of 2,000 finance leaders at US businesses over $2.5M revenue. The figure covers manual work such as reconciliation and report stitching, much of which sits in recurring reports.

What is the cost of poor data quality?

Gartner (2020) estimated the average annual cost at $12.9 million per organization, based on a survey of 154 customers. The figure spans all costs of poor data quality, not reporting alone.

What percentage of spreadsheets contain errors?

Research on spreadsheet error rates (Panko) has found rates as high as 94% in audited spreadsheets, and a review of 25 spreadsheet audits (Powell, Baker and Lawson, 2009) found 117 confirmed errors. The practical lesson is the need for checks, not the abandonment of spreadsheets.

How accurate is AI at producing report content?

Reported hallucination rates vary widely – roughly 22% to 94% depending on task and measurement, with some benchmarks finding around 14% of responses grounded. The safe operating assumption is that AI output requires human verification of every figure and claim.

What percentage of companies use AI for reporting?

77% of finance leaders reported no or limited use of AI in finance processes (Intuit, May 2026), and separate research on proposal teams found 79% had adopted generative AI in proposal work (Loopio, 2026). Adoption varies sharply by function.

How do you cite a reporting statistic properly?

Name the source, the date and the sample size where applicable, and label vendor-commissioned research as such. If you cannot name the source, do not use the figure.

How often is this page updated?

When a source we cite publishes a new edition, or when a new figure on reporting, data quality or AI in reporting is published by a source we can attribute. Superseded figures are replaced rather than retained.

Can these figures be used in a business case?

Yes, with attribution. Pair the burden figures (time on manual work) with the cost figures (poor data quality) to build the case for a controlled reporting process, and cite each figure with its source and date.

Next step

If these figures resonate, the next step is to see them against your own numbers. Subscribe for new figures and reporting guides as they are published, or book a pilot call to measure the burden in your own reporting cycle.

Sources

• Intuit Enterprise Suite, Future of Finance 2026 Report (survey of 2,000 CFOs, controllers and VPs of Finance at US businesses over $2.5M revenue, May 2026): 51% of the finance week on manual work; 70% no single source of truth; 57% missed time-sensitive action; 77% no or limited AI use.
• dbt Labs and Quietly, The Analyst Revolution (Harris Poll, 2026): 78% of analysts’ time on data preparation, validation and tool navigation.
• Financial Education & Research Foundation / Financial Executives International (September 2026): 36% of senior finance staff spend 31-50% of time on manual data work; 40% report a close of seven days or more.
• Gartner (2020, survey of 154 customers): average annual cost of poor data quality $12.9 million.
• IBM Institute for Business Value (2025): 43% of COOs rank data quality as their top data challenge; more than 25% report losses over $5 million a year.
• Nagle, Redman and Sammon, “Only 3% of Companies’ Data Meets Basic Quality Standards,” Harvard Business Review (2017): 47% of new records contain at least one critical error; 3% meet basic quality standards.
• Redman, “Bad Data Costs the U.S. $3 Trillion Per Year,” Harvard Business Review (2016): around half of knowledge-worker time spent finding, correcting and assembling data.
• Panko (University of Hawaii), spreadsheet error research, as widely cited: error rates up to 94% in audited spreadsheets.
• Powell, Baker and Lawson (2009): 117 confirmed errors in a review of 25 spreadsheet audits, the largest with an impact over $100 million.
• Stanford HAI, AI Index 2026, with Vectara’s hallucination leaderboard and OpenAI model documentation (2025-2026): hallucination rates reported between roughly 22% and 94%, with one benchmark at 13.6% grounded responses; model-specific rates of around 33% and 48% on named benchmarks.
• Harmonic Security (2025): 26.4% of employees have pasted confidential data into a generative AI tool.
• Loopio, 2026 RFP Trends Report (survey of 1,500+ proposal professionals): 79% generative-AI adoption; average win rate near 45%; 25-41 hours per proposal; around 55% of pursuits submitted.

Figures are cited from their sources and dated. Where a source is vendor-commissioned, that is stated. This page is not a substitute for the primary sources.