.png)
AI is saving time. Why isn’t it making money?
The CEO Institute in conversation with Ross Sardi, Managing Director and Chief Innovation Officer, First Focus
For the past 18 months, return on AI investment has remained one of the leading issues raised by members of The CEO Institute. The market has plenty of new products, promises and predictions, yet far less economic certainty.
During a national roadshow presented by The CEO Institute and First Focus, Ross Sardi returned to the question every CEO and board should ask: where, precisely, will the value appear?
Answering it requires leaders to connect AI investment to the operating model and, eventually, the P&L.
The adoption story is concealing an execution problem
McKinsey research presented by Ross indicates that close to 90% of organisations use AI regularly in at least one function. Around 80% say it improves personal productivity.
Then the commercial evidence falls away. Only 37% report any positive EBIT impact, and just 6% qualify as high performers by generating an EBIT improvement greater than 5%.
This is the AI ROI gap. Organisations are failing to convert individual usefulness into enterprise performance.
“AI” has become a catch-all term for capabilities at different stages of maturity, Ross explained. Productivity tools and knowledge assistants are becoming routine. Automation and analytics need greater process integration. AI-assisted development is changing software creation. Agentic workflows sit further along the curve and depend on the earlier layers working together.
Treating these as one investment category leads to poor decisions. A meeting assistant cannot be assessed like an automated service operation. A business should not pursue autonomous agents when its people barely use existing tools and its data cannot be trusted.
Sequencing matters. Ambition without organisational readiness creates expensive theatre.
Time saved has no value until leadership allocates it
Suppose AI saves an account manager 20 minutes each day by preparing quotes, summarising meetings or updating records. The business has created capacity, but nothing has reached the bottom line. Return appears when management decides how that capacity will be used.
Will that person carry more customers, pursue a higher target or lift conversion? Will the business avoid its next hire? Unless that consequence is designed into the operating model, the saved minutes will be absorbed by the day.
Many business cases multiply minutes saved by an employee’s hourly cost and declare a benefit. The salary has still been paid. The calculation measures theoretical capacity, not realised value.
Ross identifies five ROI drivers:
- cost savings,
- revenue growth,
- productivity,
- quality and
- risk mitigation.
Each still needs an economic destination. Ultimately, AI investment shows up through lower cost or higher revenue.
Quality can reduce rework or customer loss. Risk detection can prevent fraud or downtime. Better analytics can improve pricing. These are legitimate returns when the causal chain is measured. “Better quality” without a baseline and commercial consequence remains a description, not a result.
The real investment is larger than the licence
Vendor ROI models often assume perfect adoption and zero friction. Real organisations have neither.
The full cost includes technology, implementation, data preparation, integration, security, training, workflow redesign, support and consumption. Excluding these inflates the projected return and leaves the organisation unprepared to achieve it.
Ross estimates that a supported productivity tool can cost closer to $100 per person each month. A knowledge worker may need to recover only two useful hours to cover it. But the return comes from active, capable use. An unused licence creates nothing.
Restricting tools to enthusiasts can be false economy. It creates a two-speed workforce and encourages unapproved use elsewhere. Governed access, practical enablement and clear expectations give the investment a chance to perform.
Efficiency is the entry point. Growth is the prize
Most organisations approach AI as a way to reduce labour cost. Efficiency has a ceiling. Competitors can recover similar hours, turning today’s advantage into tomorrow’s minimum standard.
Growth has a different ceiling.
MYOB data shared during the roadshow found that Australian SMEs using AI were growing 2.8 times faster than those that were not. PwC research showed AI leaders were 2.6 times more likely to transform their business model, 2.5 times more likely to increase speed to market and 2.4 times more likely to create new or enhanced offerings.
QuickBooks research reinforces the pattern. Among Australian SMEs using AI, 79% reported productivity gains and 43% reported increased revenue. More had increased headcount, 19%, than reduced it, 6%.
Successful organisations are extending the reach of scarce, capable people. Senior talent can examine more opportunities, decide faster and bring products to market sooner. Revenue rises without the cost base increasing at the same rate.
For Australian and New Zealand businesses, the strategic opportunity is operating leverage, with revenue and output growing faster than cost.
Start smaller, but demand proof
Large transformation programmes are one route to meaningful value. Ross’s evidence also supports a faster approach: fast proof beats big plans.
A repeatable task consuming 20 minutes every working day may justify an investment of around $10,000 and repay it within 12 months. That could mean one employee for 20 minutes or ten for two minutes each. The recurring economic burden is the right unit of analysis.
Some solutions observed by First Focus took less than four hours to create and now save several hours each week. At scale, First Focus has automated around 900 service tickets a week. With approximately 20 minutes released per ticket, the value can be built into workforce planning.
Live fund and wealth management applications have generated estimated first-year returns of around six times and six to seven times respectively. A construction solution is projected to return around 16 times its investment over two years by avoiding up to $500,000 in hiring. A manufacturer achieved an 80% productivity uplift, reducing work from days to hours.
The method matters more than the headline. Each case identifies the problem, cost, source of value and whether the outcome is realised, projected or still being piloted. That prevents hope from being reported as ROI.
Adoption velocity predicts whether the return will arrive
Financial return is a lagging measure. Usage is the leading one.
If an automation saves 20 minutes each time it runs, first ask how often people use it. Low usage reveals early that the return will not arrive. Waiting for an annual EBIT review wastes time and capital.
Adoption depends on usefulness and ease. One unreliable answer can destroy trust. Clean, connected and permissioned data therefore belongs in the economic case.
Ross calls the principle “connect once, use everywhere”. Application data, documents, emails, conversations and operational events are cleansed, linked and governed. This trusted layer supports chat, dashboards, applications and automation. It can also reduce consumption costs by removing repeated reconciliation.
Data readiness creates compound value. Poor data multiplies friction and makes advanced automation, particularly agentic workflows, unsafe and unreliable.
ROI is a leadership system
AI is often assigned to IT because it involves technology. That is a category error.
Technology leaders should own architecture, security, governance and reliability. Business leaders must own process and performance. AI spend often competes with money traditionally spent on people, making executive accountability essential.
Ross’s challenge is intentionally blunt: “If AI ROI is unclear, leadership has failed. Not IT.”
Before approving an initiative, a CEO should answer three questions:
- Who owns the outcome?
- Which P&L line, operating measure or material risk will move?
- What evidence will cause us to scale, change or stop?
Pilots cannot survive indefinitely because their creators are invested in them. Proof earns the right to scale. Weak evidence earns a redesign or an end.
Ross also warns that AI magnifies a poor process. Leaders should ask how the work would be designed on a blank sheet today. Automating yesterday’s process may preserve the constraints the investment was meant to remove.
ROI also has a longer time horizon. Some repetitive work has helped junior employees develop expertise. Leaders must redesign how judgement, context and customer understanding are learned, or a short-term gain may create a future capability deficit.
The question has changed
The era of approving AI because the organisation should “be doing something” needs to end.
Build a portfolio of defined business problems. Establish the baseline and true cost. Assign an executive owner. Measure usage early, connect capacity to a commercial outcome and review the economics. Then scale decisively or stop.
AI has proved that it can save time. The organisations that lead from here will be those that decide, with discipline and intent, what that time is worth.