INSIGHT · AI GOVERNANCE & COMPLIANCE

In 2024-25, 6.4% of construction businesses used AI.
Start where an error does least harm.

9 October 2026

The ABS reports that 6.4% of construction businesses used artificial intelligence (AI) in the year ended 30 June 2025, against 12.1% of all businesses (Characteristics of Australian Business, 2024-25, released 25 June 2026). None of the official sources we read says which workflows builders hand to AI first. The National AI Centre's guidance says to choose by the impact if something goes wrong; its worked examples of drafting, summarising and extracting fields from documents each have a person checking, and it lists supporting decision-making among higher impact uses.

Hands working with printed documents beside a laptop on an office desk

Analysed 9 October 2026 · ABS Characteristics of Australian Business 2024-25 (released 25 June 2026), its methodology and data cube 6 · National AI Centre AI adoption tracker method page and planning guidance · Jobs and Skills Australia Our Gen AI Transition (14 August 2025) · all pages read 9 October 2026.

Key takeaways

Construction is in the bottom four industries for AI use (our calculation): In the year ended 30 June 2025, 6.4% of construction businesses reported using AI, against 12.1% of all businesses. Of the 17 industries the ABS publishes, construction is 14th (our calculation from the ABS table).
The ABS figure counts use, not depth: Businesses ticked AI from a list of technologies. The ABS states the question "is not designed to measure intensity or extent of use within the business". One person using a drafting tool counts the same as a business-wide system.
None of the official sources we read ranks construction workflows: Jobs and Skills Australia states that "there is limited evidence in Australia to show the extent to which Gen AI is being applied to specific tasks". The ABS and the National AI Centre (NAIC) tracker publish adoption rates, not a list of which jobs builders give to AI.
The guidance chooses by the cost of a wrong answer: The NAIC says to think about "the impact if something goes wrong" and to start with the simplest option. Its worked examples include drafting, summarising, extracting fields from documents and drafting quotes, each with a person checking before anything is sent.
We would start with one document type and a measured sample: A system-design recommendation: pick one document a building business handles in volume, define the fields, measure accuracy on a sample, and record who confirmed each output. That record is what makes a second workflow a decision rather than a guess.

What the ABS measured,
and what it did not.

Characteristics of Australian Business, 2024-25 is the first release of the redeveloped Business Characteristics Survey. It covers all employing businesses in the Australian economy, with exclusions, and the reference period is the year ended 30 June 2025. The methodology reports a response rate of 87.9%.

The release states the earlier comparison figure for the table below two ways: its key statistics compare 12% with "1% in 2022-23", while its ICT section compares it with "1% in 2021-22". The release does not say which year is correct.

Figure	As published	Where
Construction businesses that used AI	6.4%, year ended 30 June 2025	Data cube 6, Table 2
All businesses that used AI	12.1% in the data cube; "12%" in the release text	Data cube 6, Table 2; release
All businesses, 0 to 4 persons employed	10.7%	Data cube 6, Table 2
All businesses, 20 to 199 persons employed	22.3%	Data cube 6, Table 2
All businesses, 200 or more persons employed	34.7%	Data cube 6, Table 2

How AI was asked about: "Business use of AI was one response option from a selection of ICTs used by business. Businesses could select more than one option." The ABS adds that the question "is not designed to measure intensity or extent of use within the business".
What counts as AI: The methodology glossary defines AI as "a machine-based system that can perceive and learn about its environment and then make relevant predictions, recommendations, content, or decisions for a given set of objectives", software-based or embedded in devices.
Construction is not split by business size: The table publishes AI use by industry and by employment size separately. It does not give the rate for a construction business with 20 to 199 people, which is the size of many building groups.

Construction against
the other sixteen industries.

Ranked by the share of businesses that used AI in the year ended 30 June 2025, construction sits 14th of 17 industries in the ABS table (our calculation). Its rate is about half the all-business rate: 6.4 divided by 12.1 is 0.53 (our calculation).

Selected industries from ABS Table 2, which lists all seventeen; ranks and the ratio to the all-business rate are our calculation. None of the construction estimates in Table 2 carries the ABS caution marker for a relative standard error of 10% to less than 25%.

Industry	Used AI, year ended 30 June 2025
Information Media and Telecommunications	38.1%
Professional, Scientific and Technical Services	24.1%
Financial and Insurance Services	24.1%
Manufacturing	9.3%
Construction	6.4%
Accommodation and Food Services	5.2%
Transport, Postal and Warehousing	1.0%

What held back technology use generally: Table 4 of the same cube asks about information and communication technology (ICT) as a whole, not AI. In the year ended 30 June 2025, 16.5% of construction businesses reported insufficient staff skills and/or capability as a limiting factor, and 13.0% uncertainty around cost/benefit.
Training was less common in construction: Table 5 reports that 8.7% of construction businesses made any investment in upskilling or reskilling of employees for ICT, against 10.3% of all businesses, in the same year.

A second official measure,
counted another way.

The NAIC runs an AI adoption tracker for small and medium enterprises (SMEs). Its method page states that an independent research provider collects and analyses 400 completed surveys each month, that the surveys have run since May 2024, and that the data is weighted by industry, location and number of employees.

Two official rates for AI use in construction, if both are stored, need their question, population and period stored with them. They are not two readings of one number.

A different question: Respondents choose one statement, from "We currently use AI broadly across our business" through intending to implement AI within 12 months to "We are not aware of how to use AI within our business". The ABS question is a tick-box on a list of technologies.
A different population and period: The tracker covers SMEs, monthly, with industry data shown quarterly. The ABS covers all employing businesses, with exclusions, for one financial year.
We did not read a construction figure from it: The tracker publishes industry results in interactive dashboards, which we could not read as text on 9 October 2026. Figures attributed to it in secondary articles were not used.

None of the official sources we read says
which workflow a builder gives to AI first.

Jobs and Skills Australia's Our Gen AI Transition report (14 August 2025) states plainly: "There is limited evidence in Australia to show the extent to which Gen AI is being applied to specific tasks." The ABS reports whether AI was used, not what for. A business choosing where to start is working from guidance, not from a published ranking of construction use cases.

Use often starts with individuals: The report describes "shadow use", adoption by workers "without employers' involvement or awareness", and notes that it "results in responsibility for governance and risk management being transferred to the worker".
Why that matters for records: If a site supervisor drafts client emails in a personal tool, the business has a use with no record of what went in or who checked what came out. Choosing a first workflow deliberately is also choosing where the record starts.

What the guidance treats
as a place to start.

The NAIC's Choose a solution page describes four ways businesses bring in AI: AI built into software, standalone AI tools, AI integrations, and customised AI solutions. It says: "Start with the first option and move to later options if you need more control or capability."

Choose by impact: "A helpful way to choose the right solution is to think about the impact if something goes wrong. As potential impact increases, so does the need for oversight, accountability and ongoing management."
Lower risk, as the guidance describes it: Standalone tools are "best for testing AI, low-risk tasks, or use by a small number of staff", with examples such as drafting emails, reports or marketing content, and research or summarising information.
Higher impact, as the guidance describes it: Customised solutions are "best for complex, higher impact uses where control, data handling and accountability matter", with examples including "supporting decision-making" and "optimising complex processes, such as logistics or scheduling".
Check what you already have: The page notes that AI features added quietly to familiar software can skip the checks a new tool would get: "Teams may miss the need to run these checks if AI features are implemented quietly within familiar software."

Residential building workflows,
read against the guidance's own test.

The NAIC's Explore ways to use AI page gives worked examples; the ones nearest to building office work each have a person checking. Ordinary residential building work maps onto them as follows; the mapping is ours, not the guidance's, and is not a risk assessment of any business.

Workflow in a building business (our mapping)	Nearest NAIC example	What the guidance says a person does
Drafting client and subcontractor emails, letters and site notices	Draft emails and documents	"A person always checks and edits before sending."
Summarising site meeting notes and long email threads	Summarise meeting notes, emails and reports	"Confirm the output with participants before distributing."
Extracting supplier, date, totals and line items from invoices, quotes and delivery dockets	Extract information from documents, PDFs and receipts	"Low-confidence extractions are flagged for a staff member to check and correct."
Drafting a quote or proposal from a brief and a template	Draft proposals and quotes	"Check every fact, figure and pricing detail before sending to a client."
Programme sequencing, subcontractor selection, payment approval, compliance sign-off	Customised solutions: supporting decision-making, scheduling	Higher impact; more oversight, accountability and management

The guidance's starting instruction for extraction: "Start with one document type such as supplier invoices. Define the fields to extract and set validation rules. Test on a sample of real documents and measure accuracy before rolling out."
The six governance practices: What the NAIC's Guidance for AI adoption asks a business to keep for every use is covered in our earlier piece on the guidance and the records it asks for.

The guidance and its records

What we would put in a system

For the first workflow a building business gives to AI, we would test for five fields.

Where agents help is narrow. An agent can draft a letter, summarise notes, extract fields from an invoice or draft a quote from a brief, and a person confirms each output before it is sent or relied on. No agent decides which subcontractor is engaged, whether a payment is approved, or whether work complies. Those stay with the business and its advisers.

Field	Why it is load-bearing
Workflow and document type	One named use, for example supplier invoices, so accuracy and errors are measured against one kind of input.
Fields extracted or drafted	The guidance's instruction is to define the fields first; a field list is what a later check compares against.
Sample test result	Number of real documents tested, fields correct, errors found, and the date. The guidance says to measure accuracy before rolling out.
Who confirmed each output, and when	A named person and a timestamp per document, so a drafted figure that reached a client can be traced to its check.
Impact if wrong	Stated in plain words for the workflow. The guidance ties oversight to impact, so the next workflow's record can show why it needs more.

Questions worth asking
of your own business

For a builder: which AI tools are your staff already using, and for what?

Jobs and Skills Australia describes worker-led use without employer awareness as a common pattern. A first workflow chosen by the business replaces uses nobody recorded with one that has a record.

For a building group: which document arrives in the highest volume across your entities?

The NAIC's extraction example starts with one document type, such as supplier invoices. Volume is what makes a measured sample meaningful and the accuracy result worth storing.

For an operations or finance manager: if a drafted figure were wrong, who would see it before a client or supplier did?

That is the guidance's own test, the impact if something goes wrong. If the answer is nobody, the workflow is not yet a starting point.

For anyone switching on an AI feature in software you already use: did it get the checks a new tool would?

The NAIC notes that teams may miss those checks when AI features arrive quietly inside familiar software. The feature is a use like any other and needs the same record.

Questions people ask
about AI in construction

How many construction businesses use AI in Australia?

The ABS reports that 6.4% of construction businesses used AI in the year ended 30 June 2025, against 12.1% of all businesses (Characteristics of Australian Business, 2024-25). The question asked whether AI was used at all, not how much.

How should a construction business start using AI?

The National AI Centre's guidance says to choose by the impact if something goes wrong and to start with the simplest option, such as AI built into existing software or standalone tools for low-risk tasks. Its examples are drafting, summarising and extracting fields from documents, with a person checking.

Which AI use cases in construction are lower risk?

None of the official sources we read ranks construction use cases. The NAIC treats drafting emails and documents, summarising notes and extracting information from documents as worked examples, each with a person checking, and places decision support and scheduling optimisation among higher impact uses.

Is AI use in construction growing?

The ABS release compares 12% of all businesses in 2024-25 with 1% in an earlier survey, citing it once as 2022-23 and once as 2021-22. It does not publish a construction trend in the same table, so a construction growth rate cannot be read from it.

What this analysis
does and does not show.

Evidence note

What it shows: The ABS figure for AI use in construction (6.4% of businesses, year ended 30 June 2025), what that question does and does not measure, a second official measure counted differently, and the workflows the National AI Centre's guidance treats as places to start, read against ordinary residential building work.
Key facts quoted: 6.4% of construction businesses and 12.1% of all businesses used AI in the year ended 30 June 2025; 10.7%, 22.3% and 34.7% for businesses with 0 to 4, 20 to 199 and 200 or more persons employed; 16.5% of construction businesses limited by staff skills for ICT; response rate 87.9%; NAIC tracker 400 surveys a month since May 2024; JSA report of 14 August 2025.
Derived: Construction's rank of 14th of 17 industries and the ratio 6.4 / 12.1 = 0.53 are our calculations from Table 2.

The mapping of building workflows to NAIC examples is ours. The guidance contains no residential construction examples and does not classify any construction task.
We could not read the NAIC tracker's industry dashboards as text and quote no construction figure from it.
The ABS does not publish AI use for construction by business size, so the rate for a mid-size building group is not known from this release.
The ABS question records whether AI was used, not how much or for what. A business counted as using AI may have one person using one tool.
Figures are national. The release reports AI use by state for all industries, not for construction by state.
Nothing here is legal, technology procurement or WHS advice, and it does not assess the risk of any particular use.

Bring us the one document
your team re-keys most.

Tell us the document, how many arrive in a month, which fields your team types out, and who checks them. We will show you what a measured sample on your own documents would record, and where a person still confirms.

General information about published Australian statistics and government guidance on AI adoption. It is not legal, technology procurement or WHS advice and does not assess any particular use of AI. Figures and quotations are as read on 9 October 2026 from the sources listed.

Send us the document type · Construction AI pilot

Sources

Suggested citation: ABS, Characteristics of Australian Business, 2024-25 (data cube 6, Table 2); National AI Centre planning guidance; Jobs and Skills Australia, Our Gen AI Transition; as read 9 October 2026. Derived: construction's rank of 14th of 17 and the ratio 0.53, our calculations from Table 2.

In 2024-25, 6.4% of construction businesses used AI. Start where an error does least harm.

Key takeaways

What the ABS measured, and what it did not.

Construction against the other sixteen industries.

A second official measure, counted another way.

None of the official sources we read says which workflow a builder gives to AI first.

What the guidance treats as a place to start.

Residential building workflows, read against the guidance's own test.