The promise of artificial intelligence to catapult businesses to new heights in productivity, decision-making, and customer experience is very real. Unfortunately, a lot of the information about AI that enterprise leaders are using to guide decisions right now is not. For companies to achieve the true potential of emerging AI-driven technologies, we need to set the record straight in a few areas. 

According to a recent global survey conducted by Censuswide for my company, involving 2,000 business leaders, almost three-quarters (73%) of business leaders globally believe in the potential of AI and large language models (LLMs) and are excited by the prospect of AI to drive business expansion. Enthusiasm for AI and LLMs is even higher in the U.S., where 93% of leaders are excited about these technologies when it comes to expanding their businesses. 

We must not lose this enthusiasm due to disappointments sparked by misinformation and improper expectations. Here’s where we need to reset business leaders’ understandings surrounding the power and state of AI. 

Infrastructure and Data Quality

While AI solutions – and, more specifically, LLMs like GPT-4 or BERT – are often celebrated for their potential to transform businesses, one of the greatest breakthroughs of these models, arguably, is their intuitive ease of use. However, this also creates the misconception that LLM models are plug-and-play solutions that drive value on their own and can solve problems in isolation. That’s not the case. 

There are many AI-driven solutions being touted as capable of solving enterprise-level challenges, but they fail to discuss the infrastructure and data requirements for success. This needs to change. LLMs can only deliver value when supported by robust infrastructure and high-quality, well-structured data. This is particularly true when thinking about leveraging LLMs to drive new levels of accuracy or personalization, such as tailoring content to match a brand’s tone of voice or specialized industry terminology. In addition, common enterprise use cases, like translating user-generated reviews, aren’t possible with LLMs on their own. Such tasks require an infrastructure that can handle large volumes of incoming content and turn around translations in near-real-time. 

All of this requires an infrastructure where data flows seamlessly. Even more crucially, it demands sophisticated data repositories, such as translation memories, that are up to date and clean to fuel AI applications. Common approaches rely heavily on in-house, domain-specific data. The data feeding these AI systems must not only be of high quality but also properly formatted and organized to enable effective use and outputs.

“End-to-End” Claims

Building on the need for proper infrastructure, let’s talk about the concept of “end-to-end automation.” This is the promise that so many AI-driven solutions promise today, and it’s one where many fall short. The idea of end-to-end automation suggests that the information input into a system would be able to travel all the way through to a successful output without any intervention. However, most enterprise-grade AI solutions today aren’t able to achieve this promise because they are, in essence, a number of point solutions that have been cobbled together. 

It’s like the difference between a system of country roads and a freeway. Country roads meander and loosely intersect, often with the help of roundabouts and various connectors. That’s what a lot of today’s AI “platforms” are – a loose cobbling of systems via the use of plug-ins and workarounds. The ability of data to seamlessly flow through such systems is limited. What’s needed is a data highway – one that’s been purpose-built for high volumes of traffic and in a way where data flows freely without the need to slow down or take a side street. 

In other words: Is an enterprise’s “end-to-end” AI platform a well-integrated system? Or is it a patchwork of disconnected tools where data integrity and outputs are degraded with each handoff? These are key questions to ask when assessing the value of an AI platform. 

The Cost vs. ROI of AI

Finally, let’s talk about the underlying financials of AI, which is being touted chiefly for its ability to reduce costs while improving scale. Industry watchers continue to observe a big gap between the revenue expectations for AI investments and actual revenue growth in the AI ecosystem. This macro industry trend can be observed on a micro level within a lot of today’s enterprises, where outlays for AI-driven technology have yet to produce the expected return on investment. 

AI is an expensive endeavor, and its cost-effectiveness is increasingly being questioned. That’s largely because of the imprecise and all-encompassing way in which AI is being approached. To fix this, enterprises need to focus on more-conscientious applications of AI, using it where it genuinely adds value, rather than applying it universally. AI is not efficient or scalable in every area. When it’s forced upon the wrong use cases, overall ROI suffers. 

To truly harness the transformative potential of AI, business leaders must go beyond the hype and take a strategic, grounded approach. This means investing in robust infrastructure, prioritizing data quality, and applying AI where it truly adds value. The future of AI is bright, but only for those who are willing to build the right foundations today. 

Business leaders need to take a hard look at their organizations’ current AI initiatives: Are they built on solid ground, or are they merely patchworks of disconnected solutions? The choices executives make now will determine whether AI becomes a powerful driver of growth or just another unrealized promise.