Vardonne

Insights · Issue 01 · September 2026

More firms use AI. Few use it deeply.

A third of UK firms with ten or more staff now use AI, up from one in eight in autumn 2023. One in ten of them uses it extensively. The gap worth watching is no longer adoption; it is depth, and the distance from a pilot to something the business runs on.

1 · The number

A third of firms, from one in eight

Among UK businesses with ten or more employees, 34.9 per cent used at least one AI technology in June 2026, up from 11.9 per cent in September 2023.2,3 Across all businesses, including the smallest, it was 29 per cent, up eight points on a year earlier.1 The line has not flattened.

Figure 1 Share of UK businesses using at least one AI technology
Share of UK businesses with ten or more employees using AI, September 2023 to June 2026Rises from 11.9 per cent in September 2023 to 34.9 per cent in June 2026. All businesses, derived, rise from 9.4 to 28.9 per cent. 0%10%20%30%40% 11.9% 34.9% Sep 2023Jun 2024Jun 2025Jun 2026

Ten or more employees (ONS) All businesses (Vardonne’s derivation) Quarterly survey periods. Source: ONS2,3

  • 49%of businesses with 250 or more employees use AI, up 13 points in a year1
  • 17%of all businesses use text generation by large language model, the most common use1
  • 58% v 13%of businesses in information and communication use AI, against construction2

2 · Where a 50–500 firm sits

Closer to the largest firms than the smallest

ONS does not publish a figure for firms of 50 to 249 people. The survey dataset carries the bands, and deriving them the same way ONS’s own published figures reconcile puts firms of 50 to 99 people at 42.7 per cent and firms of 100 to 249 at 45.1 per cent.3

Figure 2 Using AI now, by number of employees, June 2026
  1. 0 to 9 28%ONS
  2. 10 to 49 33.1%derived
  3. 50 to 99 42.7%derived
  4. 100 to 249 45.1%derived
  5. 250 or more 49%ONS

Bands marked “derived” are Vardonne’s calculation from the ONS dataset; the others are published by ONS. Source: ONS2,3

Planned adoption over the next three months is 19.9 per cent for firms with ten or more staff, but 21.9 per cent of those firms answered “not sure”, and among the largest firms the unsure share reaches 42.8 per cent.3 Intent is not the constraint. Knowing what to do next is.

3 · Broad, not deep

Most use is a tool on the side of the work

10%of firms with ten or more staff that use AI report using it extensively2

15%of firms with ten or more staff that use AI say more than half their people use it in their daily work2

ONS describes use as still “relatively shallow”.2 The pattern matches what the use-case library shows: the cases that publish measured outcomes are the ones where AI sits in the path of a process, with an owner and a measure, not the ones where a licence was rolled out and counted.

4 · Public money

A record year for public-sector AI contracts, by August

We know of no official statistic that counts UK tenders or contracts involving AI. The nearest measure is a commercial one: Tussell’s tracker of awarded contracts, which reports that 2026 was already the largest year on record by 3 August.4

Figure 3 UK public-sector AI contract awards by value
  1. 2024 £582m338 contracts
  2. 2025 £1.18bn521 contracts
  3. 2026 to 3 Aug £1.41bn453 contracts

Awarded values, selected by keyword and categorised with a model under human review. One vendor’s method, not an official statistic. Source: Tussell4

Tussell highlights one contract awarded since its June update: £456 million from the Cabinet Office to KPMG and EY for civil service training, including digital and AI training.4 If it sits in the 2026 total, as its timing suggests, the rest of 2026 comes to about £954 million from 452 contracts, still below 2025’s £1.18 billion (Vardonne’s arithmetic from the tracker’s totals). The value record rests on that one contract; Tussell says the number of contracts “is also on track to exceed the record set in 2025”.4

5 · The production gap

Most pilots do not become production

Figure 4 Of every hundred organisations studied, how far enterprise AI tools got
60 of 100 evaluated enterprise AI tools, custom or vendor-sold; 20 reached pilot; 5 reached production
  1. 60 evaluated tools
  2. 20 reached pilot
  3. 5 reached production

Enterprise-grade systems, custom or vendor-sold. The report calls its figures “directionally accurate based on individual interviews rather than official company reporting”. Source: MIT NANDA5

No single study settles the rate, and they measure different things. Read together, they point the same way.

What each source measured
SourceFindingKind of evidence
Gartner, 20246At least 30% of generative AI projects abandoned after proof of concept by the end of 2025Prediction
Gartner, 20257Over 40% of agentic AI projects cancelled by the end of 2027Prediction
S&P Global, 2025842% of companies abandoning most of their AI initiatives; 46% of proofs of concept scrapped on averageSurvey, 1,000+ respondents, as reported
RAND, 20249“More than 80 percent” of AI projects fail, twice the rate of other IT projectsCiting other estimates
McKinsey, 20261044% say AI is scaling across the enterprise; 37% attribute any EBIT impact to it; about 6% are high performersSurvey, 1,719 participants

The reasons Gartner gives are the ones this practice meets: poor data quality, inadequate risk controls, escalating costs and unclear business value.6 None of them is a model problem. Guide I, the production gap, sets out the gates between a pilot and production.

6 · Reached production

Four from the library

All twenty-four use cases, by industry

7 · What changed

July to September

  1. 7 July

    DSIT and the NCSC launch the Cyber Resilience Pledge, signed by more than 60 businesses, including over 20 of government’s 39 strategic suppliers. The release notes that AI “is also lowering the barriers for attackers”.11

  2. 20 July

    ONS publishes its two-year view of AI in UK businesses: adoption among firms with ten or more staff up from about 12 to about 35 per cent.2

  3. 27 July

    The EU AI Omnibus comes into force. High-risk obligations move to 2 December 2027 for Annex III uses and 2 August 2028 for AI in regulated products.12

  4. 2 August

    The EU AI Office and national authorities begin enforcing the AI Act, and its transparency rules apply.13

  5. 11 August

    Tussell reports 2026 as the record year for public-sector AI contract awards, at £1.41 billion from 453 contracts.4

  6. 25 August

    McKinsey’s State of AI 2026: scaling up to 44 per cent of respondents, EBIT impact flat at 37 per cent.10

8 · What it means

For a firm of 50 to 500 people

  1. You are not early.

    Between four and five in ten firms your size already use AI in some form.3 The question has moved from whether to how deep, and where.

  2. Count depth, not licences.

    Two numbers are worth tracking inside your own firm: the share of people who use AI in their real work each week, and the number of processes where it sits in the path of the work with an owner and a measure. Prompt counts and seats are activity.

  3. Price the production gap before the pilot.

    If most pilots stop, the cost that decides success is the one after the pilot: integration, evaluation, ownership and support. Budget and name it at the start. Guide I.

  4. If you bid for public work, have the answer written.

    Spending is at a record, and central government buyers may ask how you use AI in writing the tender and in delivering the work.14 Guide II.

  5. EU customers: the dates moved, the reach did not.

    The high-risk deadlines are later; enforcement and transparency duties started in August.12,13 Guide V.

Sources

  1. Business insights and impact on the UK economy: 2 July 2026. Office for National Statistics, 2 July 2026.
  2. Artificial intelligence in UK businesses: 2023 to 2026. Office for National Statistics, 20 July 2026.
  3. Business insights and impact on the UK economy, dataset, Wave 159 (sheets “AI Current Usage TS (WTD)” and “AI Plan to Adopt TS (WTD)”). Size bands and the all-business series are Vardonne’s derivation: 100 per cent less “not sure” less “does not use”, which reproduces every figure ONS publishes. Office for National Statistics, 2 July 2026.
  4. UK Public Sector AI Procurement Tracker (awarded contracts selected by keyword; a vendor method, not an official statistic). Tussell, updated 11 August 2026.
  5. The GenAI Divide: State of AI in Business 2025 (“directionally accurate based on individual interviews rather than official company reporting”). MIT NANDA, July 2025.
  6. Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025. Gartner, 29 July 2024.
  7. Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027. Gartner, 25 June 2025.
  8. AI project failure rates are on the rise (reporting S&P Global Market Intelligence’s 2025 survey). CIO Dive, 14 March 2025.
  9. The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed. RAND, 13 August 2024.
  10. The state of AI in 2026 (survey of 1,719 participants, 4 May to 8 June 2026). McKinsey & Company, 25 August 2026.
  11. Businesses across Britain sign up to Cyber Resilience Pledge. DSIT and NCSC, GOV.UK, 7 July 2026.
  12. AI Omnibus enters into force. European Commission, 27 July 2026.
  13. Commission starts enforcing AI Act rules and new transparency requirements on 2 August. European Commission, 31 July 2026.
  14. PPN 017: Improving transparency of AI use in procurement. Cabinet Office, GOV.UK, 17 February 2025.
↑