AI Adoption Statistics page (curated, sourced, dated)

A curated page of the statistics we reference most often when discussing AI adoption — how many businesses have adopted it, what returns they report, where the risk sits, and how much time the underlying work consumes. Every figure is sourced, dated, and where the source is a vendor survey, accompanied by its sample size where published.

This page exists because AI statistics circulate widely and are frequently repeated without a source. Where we could not attribute a figure precisely, we have left it out.

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.
• Read adoption figures with care. AI adoption rates vary enormously depending on what is measured — trying, using, or having embedded it in operations.
• Distinguish vendor-commissioned research from independent research, and label it.
• Treat anything over two years old as historical.

Adoption: where businesses actually are

The most common misunderstanding about AI adoption is that a single number describes it. It does not: the figures below measure different things, and the spread is the finding.

73% of small businesses want AI training and implementation help.

Source: Goldman Sachs 10,000 Small Businesses Voices, AI survey (March 2026).

This is the demand figure, and it is the strongest available evidence that the gap is not awareness but capability. It also makes it plain that “we should do something about AI” is the majority position rather than an early-adopter one.

Approximately one in three small businesses do not know where to start with AI.

Source: Deloitte, The AI edge for small business (November 2025).

Demand without a route produces inaction, which is why the assessment and the first use case matter more than the strategy document.

Small-business AI adoption ranges from about 8.8% on a production-use measure (US Census) to around 76% using or exploring (Deloitte, November 2025).

The spread is not a contradiction; it is a definition problem. A business that has tried a tool and a business with AI embedded in a workflow are both “using AI” under some measures. Treat any single adoption headline sceptically and ask what it counted.

44% of small businesses have AI acceptable-use policies, and 24% cite data security and compliance as a top barrier to adoption.

Source: GTIA (2026).

Read together, these describe the position most businesses are in: using AI without a written rule, while knowing that the data question is the one that matters.

Return: what businesses report

56% of CEOs saw no significant financial benefit from AI, and only 12% gained both cost and revenue improvements.

Source: PwC, 29th Global CEO Survey (January 2026, n=4,454).

This is the outcome figure that should be printed next to the spending figure below. It is the clearest evidence available that buying tools is not the same as adopting them.

Global AI spending was estimated at $2.5 trillion in 2026.

Source: Gartner (2026).

Spending is not the measure of success; it is the measure of the size of the bet. The two figures together — record spend and thin reported returns — describe the adoption gap precisely.

The work AI is meant to reduce

51% of the finance week goes to 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).

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

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

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).

Around 60% of a sales rep’s week is non-selling activity, with roughly 40% spent selling.

Source: Salesforce, State of Sales Report 2026 (n=4,050).

The annual US cost of admin burden and repetitive tasks is estimated at $818 billion, at more than 5.5 hours per worker per week; 50% of workers have considered leaving because of admin load.

Source: Fyxer Admin Burden Index (February 2026).

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). Older data, directionally useful.

These figures are the argument for the work rather than the tool: the hours exist, they are documented, and they are concentrated in repetitive knowledge work.

Risk: where the exposure sits

The global average cost of a data breach reached a record $4.99 million in 2026, up 12%, with US organizations averaging $11.5 million and organizations under 500 employees averaging $3.31 million.

Source: IBM, Cost of a Data Breach Report 2026 (July 2026).

AI-driven attacks rose 56%, and 31% of CEOs report high cyber-risk exposure, up from 21% two years ago.

Source: PwC / IBM (2026).

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

Source: Harmonic Security (2025).

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

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

Published evaluations have reported model-specific hallucination rates of around 33% and 48% on named benchmarks.

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

Read together, these figures explain why the data rule and the verification step come before scale rather than after it. The behavior is already common and the consequence is already priced.

Data quality: the upstream problem

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

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

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).

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).

For an AI adoption case, these figures matter because AI amplifies the data underneath it. Automating on top of unreliable data produces confident output from a weak source.

Adjacent: where AI is expected to help most

77% of small businesses say marketing and customer engagement is where AI helps most, and 84% are willing to automate content creation.

Source: PayPal (2025).

86% of chief operating officers say day-to-day tasks crowd out long-term thinking.

Source: PwC Pulse.

These are expectation figures rather than outcome figures, and they are useful for one purpose: showing where demand is concentrated, which is not necessarily where the verifiable return is.

How to cite these figures

A short format that keeps the citation 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: “a 2026 vendor-commissioned survey of…”
• Do not re-round upward. A 47% is not “nearly half” in a precision-sensitive context.
• Give the measurement. For adoption figures, state what was measured — trying, using, or embedded.

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

What the figures do, and do not, tell you

Four cautions, because statistics about AI are unusually easy to misuse.

• Adoption figures measure different things. “Exploring”, “using” and “embedded” are three different claims, and the spread between them is larger than most people expect.
• Demand is not adoption. 73% wanting help describes intent; the 5-14% embedded figure describes what has happened. Both are true, and reading one as the other produces a false conclusion.
• Vendor research has an interest. Where a figure comes from a vendor-commissioned survey, label it. The methodology may be sound and the incentive is still real.
• Old figures are flagged as old. HBR’s 2016 estimate on time spent finding and correcting data remains directionally useful; it is not current.

The practical rule for using any figure in an argument: name the source, state the date, and say what it measured. Where those three are missing, the number will not survive its first sceptical reader.

Where the figures point

Read as a whole, the numbers describe a market with high demand, low embedded adoption, and a risk profile that most businesses have not yet addressed. Three conclusions follow.

• The constraint is capability, not awareness. The demand figures are unambiguous, and the outcome figures suggest that what businesses bought was access rather than adoption.
• The work is still there. The time-use figures — half a finance week, three-quarters of an analyst’s time, most of a sales week — describe the hours that a designed workflow is meant to reduce.
• The exposure is already present. Confidential data is already in tools nobody approved, and the cost of the resulting incident is significant even for firms under 500 employees.

Which is the case for the sequence this program recommends: choose one use case, design the workflow, publish the data rule, train the people, and measure the result.

Frequently asked questions

What percentage of businesses use AI?

It depends entirely on the measure. Small-business adoption ranges from around 8.8% on a production-use measure (US Census) to around 76% using or exploring (Deloitte, November 2025). Only a small minority have it fully embedded.

How many small businesses want AI help?

73% want AI training and implementation help (Goldman Sachs 10,000 Small Businesses Voices, March 2026), and around one in three do not know where to start (Deloitte, November 2025).

Do businesses see a return from AI?

PwC’s 29th Global CEO Survey (January 2026, n=4,454) found 56% of CEOs saw no significant financial benefit from AI, and only 12% gained both cost and revenue improvements — against roughly $2.5 trillion in global AI spending in 2026 (Gartner).

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. IBM’s Institute for Business Value (2025) found 43% of COOs rank data quality as their top data challenge.

How common is AI hallucination?

Reported rates range widely, roughly 22% to 94% depending on task and measurement, with one benchmark finding 13.6% of responses grounded (Stanford HAI, AI Index 2026). The spread reflects measurement difficulty; the implication — verify output — does not change.

How many employees have put confidential data into an AI tool?

26.4%, according to a 2025 Harmonic Security study. It is the most useful single figure for making the case for a data rule.

What is the cost of a data breach?

A record global average of $4.99 million in 2026, up 12%, with US organizations averaging $11.5 million and organizations under 500 employees averaging $3.31 million (IBM, Cost of a Data Breach Report 2026).

How often is this page updated?

When a source we cite publishes a new edition, or when a new attributable figure on AI adoption, risk or return is published. Superseded figures are replaced rather than retained.

Next step

If these figures are useful, the next step is to see them against your own numbers. Subscribe for new figures and adoption guides as they are published, or book an AI adoption call to measure the position in your own business.

Sources

• Goldman Sachs 10,000 Small Businesses Voices, AI survey (March 2026): 73% of small businesses want AI training and implementation help.
• Deloitte, The AI edge for small business (November 2025) and IDC SMB AI research (2026): around one in three do not know where to start; adoption ranges from about 8.8% (US Census, production use) to around 76% (using or exploring).
• PwC, 29th Global CEO Survey (January 2026, n=4,454): 56% of CEOs saw no significant financial benefit from AI; 12% gained both cost and revenue gains.
• Gartner (2026): global AI spending estimated at $2.5 trillion in 2026.
• GTIA (2026): 44% of small businesses have AI acceptable-use policies; 24% cite data security and compliance as a top AI barrier.
• Intuit Enterprise Suite, Future of Finance 2026 Report (2,000 finance leaders, 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.
• 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.
• dbt Labs and Quietly, The Analyst Revolution (Harris Poll, 2026): 78% of analysts’ time on data preparation, validation and tool navigation.
• Salesforce, State of Sales Report 2026 (n=4,050): around 60% of a sales rep’s week is non-selling; 35% of teams fully trust CRM data; 47% say accuracy has worsened.
• Fyxer Admin Burden Index (February 2026): $818 billion annual US cost of admin burden and repetitive tasks; 5.5+ hours per worker per week; 50% of workers have considered leaving due to admin load.
• Redman, “Bad Data Costs the U.S. $3 Trillion Per Year,” Harvard Business Review (2016): around half of knowledge-worker time finding, correcting and assembling data (older data, directionally useful).
• IBM, Cost of a Data Breach Report 2026 (July 2026): global average breach cost $4.99 million (up 12%); US average $11.5 million; organizations under 500 employees average $3.31 million. PwC/IBM (2026): 56% rise in AI-driven attacks; 31% of CEOs report high cyber-risk exposure.
• Harmonic Security (2025): 26.4% of employees have pasted confidential company data into a generative AI tool.
• Stanford HAI, AI Index 2026, with Vectara’s hallucination leaderboard and OpenAI model documentation (2025–2026): hallucination rates roughly 22% to 94% depending on task and measurement; 13.6% grounded on one benchmark; model-specific rates of around 33% and 48%.
• 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.
• PayPal (2025): 77% of small businesses say marketing and customer engagement is where AI helps most; 84% willing to automate content creation.
• PwC Pulse: 86% of chief operating officers say day-to-day tasks crowd out long-term thinking.

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.