Despite the promise of artificial intelligence transforming industries, rising costs and mounting risks are causing many AI projects to falter, as highlighted by several recent reports.
At least 30% of generative AI projects will be abandoned after the proof-of-concept stage by the end of 2025, according to a new Gartner report. Companies are “struggling to prove and realise value” in their endeavours, which are costing from $5 million to $20 million in upfront investments.
A separate report from Deloitte provided a similar result. Of the 2,770 companies surveyed, 70% said they have only moved 30% or fewer of their GenAI experiments into the production stage. Lack of preparation and data-related issues attributed to this low success rate.
The overall outlook for AI projects is not rosy. Research from the think tank RAND found that despite private-sector investments in AI increasing 18-fold from 2013 to 2022, over 80% of AI projects fail — twice the rate of failure in corporate IT projects that don’t involve AI.
The disparity in financial backing and completion is a likely contributor to the “Magnificent Seven” tech companies — NVIDIA, Meta, Alphabet, Microsoft, Amazon, Tesla, and Apple — all losing a combined $1.3 trillion in shares over five days last month.
SEE: Nearly 1 in 10 Businesses to Spend Over $25 Million on AI Initiatives in 2024, Searce Report Finds
Using a GenAI API — an interface that allows developers to integrate GenAI models into their applications — might cost up to $200,000 upfront and an additional $550 per user per year, Gartner estimates. Additionally, building or fine tuning a custom model can cost between $5 million and $20 million, plus $8,000 to $21,000 per user per year.
The average AI investment of global IT leaders was $879,000 in the last year, according to a report by automation software provider ABBYY. Almost all (96%) of respondents to that survey said they would increase these investments in the next year, despite a third claiming they have concerns about these high costs.
Gartner analysts wrote that GenAI “requires a higher tolerance for indirect, future financial investment criteria versus immediate return on investment,” which “many CFOs have not been comfortable with”.
But it’s not just the CFOs that have concerns about the ROI of AI endeavours. Investors in the world’s biggest tech companies have recently expressed doubt as to when, or if, their backing will pay off. Jim Covello, a Goldman Sachs stock analyst, wrote in a June report: “Despite its expensive price tag, the technology is nowhere near where it needs to be in order to be useful.”
SEE: New UK Tech Startups See First Decline Since 2022, Down 11% This Quarter
Furthermore, market values in Alphabet and Google declined in August as their revenue did not offset their investments in AI infrastructure.
A primary reason for the failure in launching enterprise GenAI projects? A lack of preparation.
Fewer than half of the respondents to the Deloitte survey felt their organisations were highly prepared across the areas of technology infrastructure and data management — both basic elements required for scaling up AI projects to a level where benefits can be realised. The RAND study also found that organisations often do not have the “adequate infrastructure to manage their data and deploy completed AI models.”
Only about 1 in 5 Deloitte respondents indicated preparedness in the areas of “talent” and “risk and governance,” and many organisations are actively hiring or upskilling for AI ethics roles as a result.
SEE: 83% of U.K. Businesses Increasing Wages for AI Skills
The quality of data represents an additional hurdle in seeing GenAI projects to completion.
The Deloitte study found that 55% of businesses have avoided certain GenAI use cases because of data-related issues, such as data being sensitive or concerns about its privacy and security. The RAND research also stressed that many organisations don’t have the necessary data to train an effective model.
Through interviews with 65 data scientists and engineers, the RAND analysts found that the root cause of AI project failure involves a lack of clarity on the problem that it promises to solve. Industry stakeholders often misunderstand or miscommunicate this problem, or choose one that is too complicated to solve with the technology. The organisation may also be more focused on employing the “latest and greatest technology” than actually solving the problem at hand.
Other concerns that may contribute to GenAI project failure cited by Deloitte include the inherent risk of AI — hallucinations, bias, privacy concerns — and keeping up with new regulations like the E.U. AI Act.
Despite poor success rates, 66% of U.S.-based CIOs are in the process of deploying GenAI copilots, compared with 32% in December, according to a Bloomberg report. The main use case cited was chatbot agents, such as for customer service applications.
The percentage of respondents that stated they were currently training foundation models also rose from 26% to 40% in the same period.
The RAND report provided evidence that businesses were not reducing their GenAI endeavours as a result of challenges in getting them over the line. According to one survey, 58% of mid-sized corporations have already deployed at least one AI model to production.
Driving this continued perseverance in GenAI are some tangible impacts on revenue savings and productivity, according to Gartner. Meanwhile, two-thirds of the organisations surveyed by Deloitte said they are increasing their investments because they have seen strong early value.
However, the ABBYY research found that 63% of global IT leaders are worried their company will be left behind if they don’t use it.
There is even evidence that GenAI is becoming a distraction. According to IBM, 47% of tech leaders feel their company’s IT function is effective in delivering basic services, a decrease of 22% since 2013. Researchers suggest this is linked to them turning their attention to GenAI, as 43% of technology executives say it has increased their infrastructure concerns in the last six months.
Rita Sallam, VP analyst of Gartner, said: “This data serves as a valuable reference point for assessing the business value derived from GenAI business model innovation.
“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 materialize over time. However, this delay doesn’t diminish the potential benefits.”
