Lots of trying, little value.
Official statistics put business AI use at between one in eight and one in four firms. Controlled trials show clear speed-ups on single tasks. Yet only about 6% of organisations see AI make a real difference to profit. Most of that gap sits in how work moves through an organisation, and the models are the smaller problem.
43% or 12%? It depends who you ask.
The National AI Centre's tracker counts SMEs that say they are adopting AI in any way. The ABS asks whether a business uses AI at all, across almost 7,000 firms. Both are right. Only the ABS tells you how far AI has spread.
Sources: National AI Centre, AI adoption insights: December 2025 to February 2026 (published 7 May 2026); ABS, Characteristics of Australian Business 2024-25 (released 25 Jun 2026)
Large businesses nearly quadrupled their use in three years. Small businesses, most of the economy, sit at 11%.
Source: ABS, Characteristics of Australian Business 2024-25 (released 25 Jun 2026)
One in four firms in the US, one in five in Europe, both climbing.
The US Census Bureau asks businesses every fortnight. Since it broadened its question in November 2025, use has crept from 18.9% in February to 24.6% in the fortnight to 18 October. Large firms use AI at 1.8 to 2.9 times the average rate.
Sources: US Census Bureau, Business Trends and Outlook Survey; Fortnightly BTOS figures as compiled by programs.com
Each country asks a slightly different question, so compare the shape, not the decimal points. The EU figure is businesses with 10 or more staff and updates automatically from Eurostat (last updated 2026-06-15).
Sources: ABS, Characteristics of Australian Business 2024-25 (released 25 Jun 2026); Census BTOS wave 5 to 18 Oct 2026 (published 22 Oct), as reported by FourWeekMBA; Eurostat, Use of artificial intelligence in enterprises (isoc_eb_ai)
Nearly everyone uses AI somewhere. Six in a hundred see it in profit.
McKinsey's survey of 1,993 executives found 88% use AI regularly in at least one function. Only 39% see any effect on earnings, and around 6% get 5% or more of their EBIT from it.
100 of 100 organisations surveyed
Source: McKinsey, The state of AI in 2025 (n=1,993, surveyed Jun to Jul 2025, published Nov 2025)
This is a survey of executives, and the profit effect is their own estimate. What separates the 6% is mostly how they run the work: 55% of them have fundamentally redesigned workflows, against 20% of everyone else.
The speed-ups are real, but narrower than the headlines.
In controlled trials, AI cut the time for well-defined tasks by 40 to 56%. On real work in a large codebase, experienced developers using early-2025 tools were 19% slower, while believing they were 20% faster.
A separate study of 5,179 support agents found 14% more issues resolved per hour, rising to 34% for the least experienced, with little change for the most experienced.
METR has since said that result is outdated: with newer tools, speed-ups now seem likely, though it considers its own new figures unreliable. The lasting lesson is the gap between how fast people feel and how fast they are, which is why measuring the whole flow matters.
Sources: Peng et al. (2023), GitHub Copilot RCT, n=95; Noy & Zhang (2023), Science, n=453; METR (Jul 2025), 16 experienced developers, 246 tasks; METR (Feb 2026): earlier slowdown now outdated; Brynjolfsson, Li & Raymond, NBER w31161, n=5,179
Faster drafting barely moves delivery while work waits in queues.
A piece of work spends most of its life waiting for review and approval. Writing it is the quick part. Almost half of managers now spend more time reviewing their team's output, and only a quarter of firms have moved even 40% of their AI pilots into production.
Drafting is 50% faster. The whole job is 13% faster.
The hours are assumptions for illustration, not measured data. Set the queue to zero to see what redesigning approvals does.
Sources: Nielsen survey of 1,700 managers for monday.com, 2026; Deloitte, State of AI in the Enterprise 2026
- Surveys define "using AI" differently, and Australian, US and EU figures cover different periods and firm sizes.
- US figures before November 2025 used a narrower question and are left out.
- Fortnightly US figures come from two independent trackers of Census data, which agree. The European figures refresh from Eurostat once a day.
- The lead-time model is an illustration with assumed hours, not a measurement.
Figures checked on 7 November 2026. Spot something out of date? pierre@aienabler.au