Across All AI Deployments, ROI Has To Come First
AI excitement was built on the promise of ROI— lots of ROI, and soon. McKinsey predicted that AI would increase corporate profits by $4.4 trillion every year. It’s easy to see why business leaders got the impression that AI would cut costs and generate revenue.
The reality is a little starker. Nearly 30% of AI projects will be abandoned after their initial proof of concept. There are any number of reasons why that might be the case: not every AI use case delivers value for money, even if the AI delivers on the concept. Similarly, not every organization has the AI expertise to deliver on potential projects. With the competition for talent so fierce, it’s no surprise that AI experts are hard to come by.
The problem is that C-suite executives are often outcomes-oriented. They have to be. They can’t afford to sink hundreds of thousands into projects that look nice, without delivering meaningful cost savings. That means that many AI projects are cut before they reach full maturity. It’s not that the technology isn’t game-changing; it’s that the game isn’t immediately changing.
Some business leaders are more patient than others. Some are prepared to take risks. But those who do are in danger of getting burnt.
A Lack of AI ROI can Lead to Buyer’s Remorse
2024 saw a rush to adopt AI as quickly as possible. Organizations bought into the promises of AI providers, but failed to see any significant ROI. The result was decisions made in haste, which business leaders have to regret.
55% of businesses that made AI-backed redundancies regret their decisions, and 38% of business leaders still don’t understand the impact that AI has had on their business. In essence, there was a knowledge gap. Businesses rushed into AI adoption without understanding what they needed or how AI would resolve those needs.
A telling example is that of Klarna, the buy-now-pay-later company. CEO Sebastian Siemiatkowski had claimed that the business would freeze hiring for customer service roles and begin the process of handing over all customer service to AI. Now, he’s backtracked on that decision, stating that Klarna will recruit a new batch of customer service professionals.
The age of human CX isn’t over yet.
Even those CEOs who haven’t made such radical proclamations are having second thoughts. The speed of adoption is likely to slow significantly, now that major flagship deployments have faltered. It’s possible that AI leaders will be less public about their achievements, and AI laggards will feel more comfortable in their hesitation.
How do we build AI ROI? Keep a steady focus and moderate expectations. The conversation tends to veer towards extremes; people pick out the headlines and leave the rest. But, a critical eye and a steady focus on genuine ROI will keep you on track toward your AI goals.
In the Future, LLMs Will Recalibrate to Drive ROI
Large Language Models took the world by storm in 2023, when the first widely adopted generative AI chatbot, ChatGPT, reached over a million users in just five days. The LLM is an impressive technology, but it’s not perfect. It is, after all, not a true intelligence, but a simulation of language. It still hallucinates, and its broadly unstructured approach to problem solving sometimes leads it to get tangled up in its own verbal rigging.
ChatGPT’s latest release, GPT-4.5, has drawn criticism for being excessively verbose and sycophantic. Its responses are longer, more repetitive, and often flatter the user at the expense of clarity or accuracy. For users applying GPT-4.5 to practical problems, like coding and research, this can prove to be a major blocker. GPT-4.5 will generate far more text than is necessary, discover problems within its response that aren’t relevant to the initial question. It tends to agree with and flatter the user, even to the point of inventing facts that agree with the user’s original statement, rather than offering a correction.
Beyond the performance quirks, a more pressing issue looms: monetization. OpenAI continues to make massive losses, with new subscription plans not nearly enough to cover the shortfall. And almost every other provider is in the same situation. New technologies, like Agentic AI, could provide new sources of revenue, but they also create new costs. Providers need ways to more efficiently monetize their technology, and fast.
One possibility is paid ads. An increasing number of people use LLMs as research tools, using them as search engines. Google has made generative AI summaries central to its search functionality in an attempt to shore up its market share. For providers that aren’t Google, it might be useful to follow Google’s monetization model. There’s only so long that you can differentiate on quality and functionality; if LLM providers want to become profitable, they may be forced to bring in advertisers to cover the shortfall in their revenues, taking advantage of their new market share. As with services like Spotify, providers may even charge users to remove ads, forcing enterprises to invest in paid alternatives to the tools their employees are currently using for free.
The dust has far from settled on the LLM. Monetization is probably the largest unanswered question hovering around the technology, and however it’s answered, it’s going to dramatically change the way we with interact with LLMs, which may undermine user trust and restrict functionality in the long run.
