AI is making work faster but proving that faster work is making the business better is another story.
Our latest AI research found 85% of AI users report some efficiency gains, including 53% who say AI has saved them meaningful time. Yet just 8% of organisations are measuring AI impact through clear metrics or KPIs.
On the surface, that looks like a measurement problem. Dig a little deeper and there’s a bigger question: are organisations measuring the right things in the first place?
For John Valastro, Director – New Business Development, Innovation & Growth at Avec, the starting point shouldn’t be AI at all, but the business problem you’re trying to solve.
Time saved isn’t the same as value created
Time saved is one of the easiest benefits of AI to understand. An hour-long task is reduced to less than 20 minutes. A report that took a day can be produced in an afternoon. A process becomes faster.
But what happens to the time you get back?
As John explains:
“Time saved is a useful metric, but I see it as an input to value rather than the value itself. If someone completes a task 30 minutes faster, the question is what happens to those 30 minutes.”
If that capacity allows the business to process more work, reduce a backlog, improve service levels, avoid additional costs or focus people on higher-value work, there is a clearer path to value.
If nothing changes downstream, the productivity gain can remain largely theoretical.
“This is why some of the productivity claims around AI can be technically true but don't actually hit the P&L or a customer outcome,” says John.
It's an important distinction when 53% of respondents to our survey say AI has saved them meaningful time. The individual benefit may be real, but organisations still need to determine what that time is worth to the business.
How should organisations measure AI ROI?
John’s answer is simple: start with the measures you already have.
He explains:
“I feel like a broken record on this topic, but this is a well-worn path that existed long before AI came onto the scene. Start with the business outcome the process already exists to deliver.”
Before introducing AI into a process, establish its current state and ask:
- How long does it take?
- What does it cost?
- Where do errors and rework occur?
- What level of manual intervention is required?
- How does the customer experience it?
- Where does risk sit?
Then ask whether those measures change.
For example, if AI is introduced into a customer service process, the useful measures might be cycle time, resolution rates, quality, rework and customer experience. A metric such as the number of AI interactions would be unlikely to tell you whether the process has actually improved.
This is also why adoption metrics need context. Licence activation, usage rates and prompt volumes can tell you whether people are using a technology, but they can't tell you whether it's making the business better.
Faster isn’t always better
Our research found time saved is the most common measure organisations are using to assess AI impact, followed by productivity or output and speed of delivery.
All useful measures, but none tell the whole story.
As John explains:
“A process can become faster while producing more errors. An employee can become more 'productive' while simply being given more work. An AI-enabled service can reduce handling time while making the customer experience worse.”
Efficiency needs to be considered alongside quality, cost, risk, employee experience and the actual business or customer outcome.
The balance will look different for every use case. A customer service team might prioritise resolution and satisfaction. A software delivery team might look at cycle time alongside defects and rework. A risk function may place far greater weight on accuracy and assurance.
There is no such thing as a perfect universal AI KPI. The point is to understand what good performance already looks like for the work AI is being introduced to improve.
Before asking about ROI, ask: why AI?
There's another problem with starting the conversation at ROI. Sometimes AI wasn't the right solution in the first place.
Almost half (48%) of respondents to our recent survey believe hype is still influencing organisational decisions about AI. With pressure to move quickly, it's easy for the technology to become the starting point rather than the problem.
John suggests a simple test: If AI disappeared tomorrow, would this still be a problem worth solving?
He says, “If the answer is no, there's a reasonable chance you're starting with the technology rather than the business problem.”
Strong AI opportunities tend to have a clearly defined problem, enough volume or economic value to matter, a measurable current state and a credible way for AI to materially improve the outcome.
Then comes the second question: Why AI?
Sometimes AI will be the answer. Sometimes conventional automation, cleaner data, process redesign or fixing an existing system will be cheaper, simpler and more effective.
There lies the nuance. An impressive AI solution applied to the wrong problem is unlikely to become a compelling business case.
The hardest part of AI ROI often isn’t the AI
Even with the right use case, moving from a promising pilot to measurable value isn't guaranteed.
“From my experience, the hardest part isn't the model itself. It's everything around it,” says John.
Poor data, no meaningful baseline, a solution that sits outside the actual workflow, governance introduced too late, weak adoption or change management, or processes that haven't been redesigned around the new capability.
Any one of these can prevent a technically successful AI implementation from producing a measurable business result.
This is where the AI ROI conversation needs to move beyond the technology itself. The business case needs to account for what it takes to make the technology work in practice: the data, integration, process change, governance, capability, adoption and ongoing oversight around it.
Agentic AI could change the value equation again
That calculation becomes even more important as organisations move towards agentic AI.
While AI can generate content, analysis or recommendations, it still leaves a real person responsible for taking the action. More autonomous systems can increasingly take actions within workflows and systems themselves.
That creates new opportunities for productivity and scale, but it can also introduce new costs around control, assurance, security and oversight.
For organisations building a business case, those costs belong in the value equation too.
It means AI ROI may increasingly need to be considered through a total cost of ownership lens: not simply what the technology costs and how much time it saves, but what the organisation needs to invest to deploy, control and operate it safely and effectively.
Stop measuring AI. Start measuring the business.
If AI is already being used across an organisation but nobody can confidently say whether it's improving performance, adding more AI metrics probably isn't the answer.
Go back to the major use cases, define the underlying business problem, and establish the baseline. Then ask what has actually changed.
Where there is measurable improvement, invest and scale. Where there isn't, understand why and be prepared to stop initiatives that don't have a credible path to value.
As John puts it:
“The objective shouldn't be maximum AI adoption, but better business performance, with AI used where it genuinely helps achieve it.”
Our 2026 AI Report: How is AI really changing the way we work? explores what 1,505 business leaders and technology professionals across Australia and New Zealand told us about AI adoption, productivity, measurement, governance and the future of work.
See where the gap between AI use and business impact sits.
