Gartner: almost a third of generative AI projects will fail soon

Analyst firm Gartner believes 30% of generative AI projects currently on the go in businesses won’t make it past the proof of concept stage and be abandoned by 2025.

Given the sudden explosion in generative AI, that 30% failure rate may even feel low — and, of course, the whole point of a proof of concept is to see if something works. If an idea doesn’t pan out, there’s no shame in kicking it to the curb.

But Gartner’s prediction suggests generative AI projects will be ditched because of poor data quality, insufficient risk controls, high costs or unclear business value.

“After last year’s hype, executives are impatient to see returns on GenAI investments, yet organisations are struggling to prove and realize value,” said Rita Sallam, Distinguished VP Analyst at Gartner, speaking at a conference this week.

“As the scope of initiatives widen, the financial burden of developing and deploying GenAI models is increasingly felt.”

The mad rush to use generative AI isn’t cheap — and the more complex the problem addressed by the technology, the more it costs.

Cost of generative AI projects to business

Indeed, Gartner notes that one of the challenges of deploying generative AI is the high costs. With up-front costs hitting many millions, it’s hard to see an immediate return on investment.

Of course, the cost depends on the nature and size of the project, the industry and how AI is being used.

Gartner estimates that giving staff a coding assistant via a ready-to-go app could massively boost productivity, but it comes with two costs. First, an upfront cost between $100,000 and $200,000 to set up a system; second, recurring costs in the hundreds of dollars per user.

Building an AI-powered sales app via an API could cost up to $1 million, with recurring costs of $1,000 per user annually. AI-assisted data retrieval such as document search would cost about the same.

Think that’s a lot? A virtual assistant running a model tweaked to work for your company could cost as much as $6.5 million to set up, and several thousand per user annually. And building a custom model from scratch has an upfront bill between $8 million and $20 million, with annual per user costs in the thousands of dollars.

“Unfortunately, there is no one size fits all with GenAI, and costs aren’t as predictable as other technologies,” said Sallam.

“What you spend, the use cases you invest in and the deployment approaches you take, all determine the costs. Whether you’re a market disruptor and want to infuse AI everywhere, or you have a more conservative focus on productivity gains or extending existing processes, each has different levels of cost, risk, variability and strategic impact.”

Generative AI projects may deliver benefits slowly

Generative AI may be an easy sell to trend-focused executives at the moment, but Sallam warned that benefits may be slow to come — so be prepared to wait for the payoff.

But she also noted that it’s difficult to predict exactly how this nascent technology will impact different companies, across myriad industries, depending on their use cases.

Despite so many generative AI projects failing to move beyond a proof of concept, Gartner research suggests that early results from early adopters suggest gains in revenue, cost savings and productivity.

“This data serves as a valuable reference point for assessing the business value derived from GenAI business model innovation,” said Sallam.

“But it’s important to acknowledge the challenges in estimating that value, as benefits are very company, use case, role and workforce specific. Often, the impact may not be immediately evident and may materialise over time. However, this delay doesn’t diminish the potential benefits.”

In other words, the long-term potential for generative AI remains, but figuring out exactly how — and whether — it will benefit businesses may be costly in the short run.