Martin Filz, the CEO of Pureprofile, has seen firsthand the game-changing potential of AI in software development.
His team recently used AI to build a prototype that perfectly matched their business needs in just three days - a process that would have normally taken up to a year. However, Filz has also experienced the pitfalls of relying on AI-generated code, which can sometimes fail unexpectedly and end up taking engineers more time to fix than if they had started from scratch.
Pureprofile is an ASX-listed data and consumer insights company, and like most technology businesses, Filz says it has embraced AI across its operations. "We've seen first-hand how quickly it can accelerate innovation, improve quality and reduce costs," he writes. "But we've also learnt that AI isn't a silver bullet. Some of its limitations and hidden costs only become apparent once a solution moves from experimentation into production."
Since the release of ChatGPT 3.5, Filz notes that innovation itself has sped up, with boards and investors now expecting a clear AI strategy, employees experimenting independently, and vendors promising transformation across every part of the business. "The pressure to adopt AI is real," he writes. "But is it always the answer? For CEOs, the dilemma is not whether to deploy and use AI, but where and when to use it."
The advantage of automation
Filz says organisations have historically had to choose between being faster, cheaper or better, but argues AI offers the chance to achieve all three at once. "Skill shortages, escalating hiring and wage costs, as well as uncertainty are a triple challenge for businesses," he writes. "AI helps lower business costs by automating a lot of repetitive, manual tasks, so existing staff can focus on higher-value work."
He points to AI's ability to speed up internal analysis as one of its most powerful benefits for business leaders. "AI is dramatically reducing the time between asking a business question, understanding what the data is telling you and acting on the answer," Filz says.
At Pureprofile, he says AI also helps combine previously unrelated data sources to surface insights. "It's able to near-instantly identify trends from large volumes of data and understand how different groups of people respond to research requests, enabling better targeting," he writes. The company has also started using AI as an added layer of quality control, with automated alerts flagging research data that may need further review. "The objective isn't to remove human oversight, but to use AI to make it more targeted and effective," Filz says.
In practice, he says AI-powered workflows have cut the time needed to analyse raw data and test product scenarios by around 80 per cent, while in another case, AI streamlined a data entry and formatting task, eliminating 22 manual clicks and saving nearly 10 minutes of processing time.
Knowing when not to automate
Despite those gains, Filz cautions that AI token costs can sometimes exceed the cost of a human doing the same task. "It's important to check the overall cost of an AI-driven system once live," he writes, adding that even large technology companies are still getting this wrong.
He cites two recent, high-profile examples. Amazon's project to automatically match author names to product listings, built on Anthropic's Claude Sonnet model, ran 860 per cent over its original budget, ultimately costing around US$1.8 million before the deployment was abandoned, according to a Financial Times report. The overspend reportedly went undetected internally for five months. Separately, Uber placed monthly spending caps of US$1,500 per employee on agentic AI coding tools such as Claude Code and Cursor, after the company burned through its entire 2026 AI budget in just four months, driven by adoption jumping from roughly a third to more than 80 per cent of its engineering workforce.
Filz says AI can also be inconsistent without clear direction. "When testing AI-assisted language checks on questionnaires, we found that the same questionnaire could produce different recommendations depending on how the AI was prompted," he writes. "It reinforced the importance of clear instructions and human oversight when using AI." He describes similar problems in data analysis, where unclear naming conventions led AI to link the wrong fields or edit the wrong columns, reinforcing that "human review remains critical."
"The lesson is this - don't just jump on the AI bandwagon for the sake of keeping up," Filz writes. He argues the key is empowering the whole organisation, not just developers, to identify where AI can produce tangible results. "Some of the best ideas come from users who understand a process intimately rather than the developers themselves," he says.
"Ultimately, the solution to the CEO AI dilemma comes down to knowing where AI can genuinely make a business faster, cheaper or better," Filz concludes. "The winners will be the ones that leverage AI with discipline and understand where AI creates real value instead of using AI for the sake of it."