Case studies

What happened to the ones who spent the money

Five sets of figures, all of them checkable. Some bought back real time. Some returned nothing at all. The ratio is probably not the one you would guess.

Where Australian businesses actually are

All industries Β· Australia Β· official statistics

What was measured

How many Australian businesses are genuinely using AI, split by size and by industry β€” the baseline everything else should be read against.

How

Australian Bureau of Statistics, Business Characteristics Survey, 2024–25.

The numbers

12%
Of all Australian businesses use AI
3%22%
Medium businesses, in three years
9%35%
Large businesses, same period
11%
Small and micro businesses

By industry: information, media and telecommunications 38% Β· professional, scientific and technical services 24% Β· financial and insurance services 24%, up 24-fold from 1% Β· arts and recreation up 16-fold.

The part that gets left out

The line that matters most is easy to miss: among small businesses that were innovation-active, adoption was 19% β€” almost five times the rate of those doing no innovation at all.

Why it matters to you

Nine in ten small businesses haven't started. That's not a reason to panic-buy β€” it means the window is still open and you are not behind. It also means the gap isn't about size or budget: the small firms that got there were the ones already in the habit of changing how they work.

Source: Australian Bureau of Statistics, 2024–25

The pilots that went nowhere

Cross-industry Β· the base rate nobody quotes

What was measured

What share of generative AI pilots actually reach the profit and loss statement, across companies that had already spent the money.

How

MIT's NANDA project, *The GenAI Divide: State of AI in Business 2025* β€” a systematic review of 300+ publicly disclosed AI initiatives, structured interviews with 52 organisations, and survey responses from 153 senior leaders, January–June 2025.

The numbers

95%
Of pilots produced no measurable P&L impact
5%
Reached real value
US$30–40bn
Enterprise spend behind those results

The report labels itself preliminary findings. The failures weren't blamed on model quality β€” they were tools that never entered the workflow they were bought to change.

The part that gets left out

More than half of all budgets went to sales and marketing tools β€” while the study found the largest returns in unglamorous back-office work.

Why it matters to you

This is the number to hold every pitch against. When a vendor shows you their one success, the honest question is what happened to the other nineteen. And note where the returns actually were: not the exciting front-of-house tool, the boring internal one.

Source: MIT NANDA, State of AI in Business 2025

A trial run properly, and what still didn't land

Office and knowledge work Β· Australian Government

What was measured

What happens when an organisation does the rollout well β€” licences, training, support β€” and then measures honestly instead of announcing a win.

How

The Australian Government's whole-of-government evaluation of Microsoft 365 Copilot, published on digital.gov.au.

The numbers

69%
Agreed it improved the speed of their tasks
61%
Agreed it lifted the quality of their work
40%
Could reallocate the time to higher-value work
up to 7%
Said it ADDED time, because outputs had to be verified

About 65% of managers saw a positive impact on their team's quality and efficiency. On speed, 16% of respondents disagreed outright and another 16% sat neutral.

The part that gets left out

Read the distribution rather than the headline: even with training and support in place, roughly a third of people did not agree it made them faster, and a slice were slower because they had to check the output.

Why it matters to you

This is a well-run rollout, and it still lands unevenly. Which is the argument for picking one task and measuring it, rather than buying a licence for every desk and assuming. Note where the time went for the 40% β€” reallocated, not removed.

Source: Australian Government, Microsoft 365 Copilot evaluation

Accounting and bookkeeping practices

Professional services Β· Australia Β· small firms

What was measured

Where the time actually goes in small practices using AI in ordinary daily work, rather than as a transformation programme.

How

The Access Group surveyed 434 Australian accountants and bookkeepers, 9 September – 13 October 2025.

The numbers

37%
Save at least 30 minutes a day β€” 125+ hours a year
26%
Save 15–30 minutes a day
19%
Save an hour a day or more

The report values 30 minutes a day at over three working weeks a year, roughly A$13,750.

The part that gets left out

No single practice reported anything dramatic. The result is broad and unspectacular: a lot of firms getting back half an hour a day.

Why it matters to you

This is what realistic looks like at your size. Half an hour a day never makes a headline, but it compounds β€” and it's the honest benchmark to measure a vendor against when they promise a transformed firm.

Source: The Access Group, State of AI in Accounting 2026

The search results your customers see

Every business with a website

What was measured

Whether AI summaries sitting above search results change what people click β€” measured from real browsing, not asked in a survey.

How

Pew Research Center tracked 900 US adults across 68,879 Google searches during March 2025.

The numbers

15%8%
Visits where the user clicked any result
16%26%
Sessions that simply ended
1%
Clicks on links inside the AI summary itself

About 18% of searches in the sample triggered an AI summary; 12,593 of the 68,879 searches produced one.

The part that gets left out

For most businesses this arrives as flat or falling traffic with no ranking drop to explain it β€” which is why it usually gets blamed on the wrong thing.

Why it matters to you

You can hold your ranking and still lose the visit. The useful question is no longer only where you rank, but whether the answer sitting above the results mentions you, and whether what it says about you is right.

Source: Pew Research Center, July 2025

See also Β· our own client work

An AI phone service we built and delivered: calls to stores on a schedule, the support line answered and triaged, every call recorded and traceable. Anonymised at the client's request.

Read that engagement

Where does your business sit against these?

Ten questions, two minutes. You'll get which of the four steps you're actually on and the single next action for it.