Skip to main content
Neon Light / Perspectives

What are businesses actually getting back?

Matt Merryfull | Neon Light
Toy-like blocks representing data, process, people and measurement connect along a gold path towards a shared business goal.

An AI budget can be easy to explain in terms of tools, licences and people trained. Explaining what the business has received in return takes more work.

That is where I think the conversation needs to move. The rush to onboard AI capability has left little room for the changes to make the investment worthwhile. Businesses still need to decide how work will change, where people need support and what improvement would justify the cost.

In Getting from AI capability to meaningful outcomes, I dived into the transition from individual capability to dependable ways of working. There is a further question for anyone approving the budget: once those improvements are in place, what return can the business reasonably claim?

01

What the numbers actually show

There is evidence of useful gains.

Reported value

KPMG’s Global AI Pulse Q3 2026 reports that 89% of surveyed organisations see measurable value from AI, while only 12% consistently assess that value against its cost across the organisation.

These are leaders’ reported assessments, drawn from businesses with minimum revenue thresholds of US$50 million or more depending on the market. They describe benefits and a gap in measurement, rather than audited returns across the whole business population. KPMG report.

Modelled opportunity

For Australian SMEs, the Tech Council’s October report estimates that broader AI adoption could generate up to A$93 billion in additional annual economic value over the next five years. Its A$136,000 annual productivity figure for an average SME is a modelled opportunity from moving from no AI use to broad adoption.

A business owner should read it as potential to investigate, rather than an amount to slap into next year’s savings forecast. Tech Council announcement.

Measured improvement

Measured operational improvements are also available. A study published in the Quarterly Journal of Economics followed the introduction of an AI assistant across 5,172 customer-support agents and found a 15% average increase in issues resolved per hour.

Those gains varied substantially depending on the level of experience and skill. It provides evidence for a particular application in a particular setting, without establishing a universal return on AI spending. Generative AI at Work.

Together, these findings give leaders a reason to invest and a reason to be precise about the claim. A projected economic opportunity, a faster operation and a financial return answer very different questions.

02

Work out where the benefit goes

Time saved is often the first improvement people notice. It also needs careful treatment when it enters a business case.

Consider this … A team spends 40 hours each month preparing and checking a recurring report. With AI, preparation takes 10 hours and review and correction take another 20. The team has released 10 hours, assuming the final report meets the same standard.

At an assumed staff cost of A$75 an hour, that is A$750 of monthly capacity, before the additional technology and implementation costs. The payroll bill may end up being exactly the same.

What happens to those hours determines the next part of the return. The team might clear a backlog, take on more work without overtime, improve analysis or stop working late. Each one could justify the investment. They should be described and measured for what they are.

If overtime or external spending falls, there is a cash saving to verify. If the team serves more customers, look at the additional contribution after the cost of serving them. If an organisation gains breathing room, track workload and service quality rather than trying to convert every hour into a claimed financial saving.

The same discipline applies to growth. Producing more proposals does not establish additional revenue. Look for changes in conversion, delivery capacity and margin, and consider what else changed during the period. Avoid counting both the value of released hours and the output produced in those hours as separate benefits when they describe the same situational gain.

03

Include the cost of making it work

The AI subscription is only part of the investment. Depending on the application, the business may also need to pay for integration, data preparation, training, security controls and ongoing support. People need time to learn the new process, check results and handle exceptions.

Those costs belong in the decision from the beginning. Leaving them unfunded can make an initiative look inexpensive while making the expected return harder to achieve.

Compare the full cost of the existing way of working with the full cost of the proposed one. Review and correction may already happen today, so measure how they change rather than charging all checking time to AI. Separate one-off implementation costs from continuing costs, and use the same period when comparing costs and benefits.

Quality belongs in that comparison too. A cheaper output can create extra work later through incorrect information, missed requirements or decisions that need revisiting. The relevant cost is what it takes to produce work the business can use with appropriate confidence.

This is where investment in transformation becomes practical. It may mean connecting two systems, changing an approval step or giving experienced staff time to develop and teach a reliable way of working. The scale of that investment should follow the value available and the consequences of getting it wrong.

Granted, this may all sound like common sense and to a large extent, it is. Up until the very recent past, digital transformation has lived behind a curtain of skills, technology and execution. Luckily for everyone, that’s changing.

04

Make the next spending decision easier

A small business should be able to test a modest improvement without creating an elaborate measurement programme. A short, bounded trial may be enough. A critical workflow affecting many customers needs stronger evidence and controls.

For the initiative being considered, a useful starting point is to agree:

  • The business result expected, with the current position recorded.
  • The costs and quality measures that will show whether the improvement holds up.
  • Who will use any released capacity and how the resulting benefit will be checked.
  • When the evidence will be reviewed and what would justify continuing, changing or stopping the investment.

Use comparable work in the trial and account for changes in demand or complexity. A result from a few easy cases should not become the forecast for every case. Keep estimates visible as estimates until there is enough evidence to replace them.

There is also room to invest in learning. People need practical experience to understand where AI helps, where it fails and what they can safely rely on. Give that work a proportionate budget and a clear question to answer. It can be worthwhile before there is an immediate financial return, provided the business is clear about what it is funding.

My advice – start small. Just like heading to the gym, you don’t go in on day one trying to break a world record, you start small, learn boundaries and constraints, and continually improve and expand.

The question I would bring to the next AI budget discussion is straightforward: what has improved, what did that improvement cost, and what should we do with that evidence?

A credible answer makes it easier to fund the capabilities that are helping, make the changes still needed and stop paying for activity that has not earned its place.