Hidden AI costs catch out corporate budgets as indirect expenses bite

More than half of business executives say their spending on artificial intelligence exceeded expectations or initial budgets over the past twelve months, according to research published on 12 November by the Capgemini Research Institute.

The global survey of 500 senior corporate executives found that 59 per cent experienced higher spending than anticipated. Among the organisations that faced budget blowouts, costs ran an average of 24 per cent over target, while roughly one in five exceeded their budget by more than 30 per cent.

Direct technology costs, such as software subscriptions and computing tokens, were rarely the main culprit. Instead, the research found that 53 per cent of executives are now concerned about rising AI expenses because of indirect operational costs. Only one in three leaders reported being able to estimate indirect expenses reliably.

The primary drain on finances came from human oversight, correcting mistakes made by automated tools, and governance requirements. Half of all surveyed executives admitted that their organisations approved AI projects without a clear understanding of the full operating expenses involved. Consequently, just 56 per cent of executives felt confident they could accurately forecast how their AI expenditure would look as they expanded usage.

The study also noted that only one in five companies actively measures verified business returns after subtracting the total safe-delivery cost of an AI project. In response to the mounting expenses, many organisations are re-evaluating their setups, tightening budgetary controls, and moving towards multi-model approaches rather than relying on a single large vendor.

What this means for Australian businesses

For Australian small and medium business owners, the Capgemini findings show why simply buying AI software subscriptions does not automatically save money.

• Factor staff time into the true cost of any tool, because hours spent reviewing text, checking facts, or fixing coding errors quickly eat into projected productivity gains.
• Track outcomes rather than activity by setting clear targets to see if a tool actually reduces working hours, increases sales, or lowers running costs.
• Start small before rolling out tools across an entire team, testing software on one repetitive workflow first to see the real balance between time saved and time spent checking work.
• Review software subscriptions and usage limits regularly, as API fees, automatic upgrades, and unused licences quietly add to monthly overheads.