Gartner: Avoid AI project money pit

Generative artificial intelligence (GenAI) is past its peak in the Gartner hype cycle but has not met expectations, analysts warned at the company’s European conference in Barcelona.

In the opening keynote at the European Gartner Symposium, Alicia Mullery, a vice-president for research at the analyst firm, discussed two AI races: the first being the tech providers’ race; the second being to deliver AI outcomes safely and securely. “This is your race,” she told the audience of IT executives.

One of the takeaways from the opening session is that it is easy to waste money with GenAI. “You need to understand the bill and monitor it at all times,” warned Mullery and co-presenter Daryl Plummer, chief research analyst at Gartner.

Plummer noted that the majority of organisations Gartner has spoken to are not ready for AI. “They’re not ready for it emotionally, technologically, organisationally or management-wise,” he said.

To minimise failure. Gartner recommended two approaches: one for organisations looking primarily at using AI for improving productivity; the second focused on using AI to drive transformational change.

Data from Gartner shows that running a proof of concept project can cost anything from $300,000 to well over $2m. While IT and business leaders may appreciate the large costs associated with training AI models on expensive graphics processing unit (GPU) hardware, Plummer said the costs associated with AI inference can quickly get out of hand.

“Processing is very expensive because the AI models have to use something called matrix multiplication to process all the parameters they use to get to a prediction. This requires GPUs, which you either buy and put in your own datacentre, or lease from a cloud provider. Both of these are very expensive,” he said.

Plummer warned that tech providers were too focused on looking at the advancement of AI from their perspective, without taking customers on the journey to achieve the objectives of these advanced AI systems. “Microsoft, Google, Amazon, Oracle, Meta and OpenAI have made one major mistake – they’re showing us what we can do [but] they’re not showing us what we should do,” he said.

Since many organisations are not ready to adopt the advanced AI available to them from the major providers, Plummer said many are finding that 75% of their budget is being spent on IT consulting to understand how the new technology can benefit their organisation.

“To get to the proof of concept stage requires more budget,” he said, adding that costs will continue to rise until IT leaders start putting enterprise AI systems into production, at which point they should be able to gain a better understanding of how to manage ongoing costs.

The analysts explained that IT leaders need to consider the outcomes they want to achieve. Those looking at deploying AI to achieve business efficiency improvements – referred to by Gartner as “AI-steady” organisations – are likely to be running 10 or fewer pilots or AI initiatives. In this scenario, people can be tasked with monitoring and checking to ensure the AI systems are working correctly.

Those organisations where GenAI is seen as an industry-transforming technology are likely to run many more pilots. Gartner categorises these organisations as “AI-accelerated”. The analyst firm does not believe it is humanly possible to manage the AI systems that AI-accelerated organisations are looking to deploy.

As such, it predicted the rise of technology dubbed TRiSM (trust, risk and security management), which it said would play a significant role in ensuring AI systems remain compliant.