Behind the big promises on AI in business lies... nothing

9 October 2025

Business School, Staff news, Business and economy, AI

Opinion by Dr Dulani Jayasuriya (finance and accounting) and Professor Alex Sims (commercial law), University of Auckland Business School.

A recent MIT study highlights an awkward truth about corporate AI, suggesting 95 percent of generative AI pilots deliver zero return on investment.

While many executives are celebrating their "digital transformation" and investing billions of dollars into implementing AI solutions, some employees are using $20 ChatGPT subscriptions that work better than million-dollar enterprise systems, according to the MIT report.

Despite companies investing a staggering USD$30-40b in enterprise AI systems, 95 percent are getting absolutely nothing back. Not small returns. Not disappointing results. Zero.

This isn't just another tech bubble story. It's a wake-up call that shows something profound about how modern businesses work, and why they're failing at their biggest technology bet in decades.

A real story from the MIT study captures the problem: A corporate lawyer at a mid-sized firm watched her company spend $50,000 on specialised contract analysis software. The sales pitch was impressive. The demos looked great. The IT department was thrilled. But the lawyer tried it once and never touched it again.

Instead, she uses her personal $20 monthly ChatGPT subscription, because in her words, it "actually works". The expensive AI tool provided "rigid summaries with limited customisation", while ChatGPT let her refine and iterate until providing exactly what she needed.

This isn't an isolated incident. The MIT researchers found that 90 percent of employees are using personal AI tools for work; ChatGPT, Claude, and various consumer apps, often violating company policy and raising confidentiality and privacy issues.

Why banks score an F in the AI revolution

Financial services should be leading the AI revolution. After all, banks essentially run on data, algorithms, and automated decisions. They invented complex derivatives, high-frequency trading, and can calculate the risk of obscure events years into the future. Yet, MIT's AI Market Disruption Index gives financial services a low score of just 0.5 out of 5, barely registering any transformation at all.

Your favourite news website has probably changed more from AI than your bank.

There's another phenomenon costing companies millions that nobody wants to discuss: "workslop", the avalanche of mediocre, AI-generated content now flooding corporate communications.

Many meetings now start with AI-generated summaries of AI-generated reports based on AI-generated research. Teams wade through perfectly formatted documents that look and sound impressive but say nothing. It's like everyone hired the same mediocre ghostwriter who doesn't understand the business.

The irony is painful. The same technology meant to boost productivity is creating a new form of inefficiency; polished nonsense that takes time to read, process and ultimately ignore.

Follow the money (hint: it's going to the wrong places)

Want to know why AI isn't working? Look at where companies spend their AI budgets. According to MIT's research, 50% of AI spending goes to sales and marketing, the flashy, visible functions that impress boards and investors.

Meanwhile, the boring stuff that actually saves money gets ignored. The few companies seeing real returns from AI aren't the ones with fancy chatbots. They're quietly automating their back offices and seeing results.

But here's the problem, beautifully captured by a Fortune 1000 procurement executive: "How do I quantify faster workflows? How do I justify it to my CEO when it won't directly move revenue?" This is the same industry that can price the probability of a hurricane hitting Miami on a specific Tuesday in 2027, but apparently can't measure the value of their team working 40 percent faster.

The trust problem that changes everything

When MIT researchers asked executives what matters most in choosing AI vendors, "a vendor we trust" topped the list. However, sticking to trusted vendors creates a dilemma because banks' trusted existing vendors can't innovate; meanwhile innovative startups, who could transform their business aren't trusted. According to the Financial Stability Board, this conservative approach reflects legitimate concerns about AI risks in financial services, but it's also keeping institutions stuck with outdated technology.

Companies that partner with external AI partnerships succeed 67% of the time. Those that try to build AI internally? Only 33% succeed, according to the MIT report.

Why? Most internal teams build what one executive called "science projects", impressive demos that collapse under real-world pressure.

External partners who understand specific industries are not trying to reinvent banking; they're trying to make banking work better.

MIT's research contains a warning. Several companies will lock in AI vendor relationships over the next 18 months that will be "nearly impossible to unwind". Once an AI tool learns your company's workflows, ingests your data and integrates with your operations, switching becomes prohibitively expensive. You're not just changing software; you're retraining an entire digital workforce.

The uncomfortable truth about jobs

While CEOs publicly claim AI won't impact employment, the data tells a different story. Organisations successfully using AI report 5-20% reductions in customer support and administrative processing.

But here's the twist: It's not employees being replaced, it's outsourced functions. That $10 million contract with an overseas call centre? Replaced by a $500,000 AI system and two analysts. Those consultants charging $5000 per day? Superseded by AI working 24/7 for the cost of a junior employee.

After analysing hundreds of implementations and interviewing executives across industries, MIT's conclusion is that most AI failures aren't about technology, they're about approach.

Companies pouring millions into AI that doesn't learn, adapt, and remember, are essentially buying very expensive calculators in the age of smartphones. Meanwhile, their employees, using $20 consumer tools, are having better results.

The solution isn't complicated:

- Stop building internally when partners do it better.

- Focus on boring processes that actually save money.

- Buy AI that learns and improves, not static tools.

- Listen to your employees who know what's needed.

This article reflects the opinion of the authors and not necessarily the views of Waipapa Taumata Rau University of Auckland.

It was first published by Stuff

Media contact:

Sophie Boladeras, media adviser M: 022 4600 388 E: [email protected]