- Some enterprises are struggling to measure ROI on AI investments, with Gartner predicting a 30% abandonment rate for generative AI projects by 2025
- The outcomes of AI projects are so far varied, leaving experts divided on whether they're worth the investment
- Alternative ways to measure the value of AI investments could include focusing on non-financial metrics
As enterprises pour millions into AI technology, many are left questioning if the hefty investments will ever deliver on their promises. So far, the results have been varied and at times, elusive.
Gartner estimates that organizations may spend between $5 million to $20 million each year on embedding, customizing or building generative AI (GenAI) applications. However, the firm also predicts that at least 30% of these projects will be abandoned after proof of concept by the end of 2025.
As many enterprises invest substantially in AI tools from major vendors like Microsoft, Salesforce and Oracle, the ROI for these tools is often difficult to measure, Frances Karamouzis, Gartner’s distinguished VP analyst, noted.
Calculating ROI in traditional financial terms for AI projects is inherently challenging due to the ubiquitous nature of applications that touch across departments within an organization. Around 70% CIOs believe that even trying to predict the ROI is a "finger in the air" exercise, a study from software firm Ardoq found.
“When people buy Microsoft Copilot and a CFO asks, ‘so 20,000 seats is going to be somewhere between $11 million and $16 million. What's the ROI on that?’ First of all, you're not going to get a uniform ROI. It becomes very, very difficult to calculate ROI,” Karamouzis told Fierce Network.
The ROI debate
AI's high development and operational costs mean that it should solve complex and important problems to justify the investment, Jim Covello, Head of Global Equity Research at Goldman Sachs said in a June letter.
Even if AI costs decline (which Covello added, they might not), they would need to drop dramatically to make automation truly affordable. Many businesses could be setting up for losses with rushed AI strategies or over-investment, but "sustained corporate profitability" will mean continued experimentation with negative ROI projects, he said. "As long as corporate profits remain robust, these experiments will keep running."
Some enterprises are beginning to withdraw from AI projects that are becoming “black holes” of spending without delivering the anticipated returns, Karamouzis pointed out.
But results have been varied, and the GenAI gamble could pay off for some companies or projects.
New research from Google Cloud showed 86% of respondents reported estimated revenue gains of more than 6% due to GenAI implementations. Google, a large investor in AI technology, argued that AI's potential for boosting productivity is already being realized, with nearly half of respondents noting that AI tools have more than doubled employee productivity.
Productivity returns
For some AI projects, Gartner is recommending not to even try calculating ROI, according to Karamouzis. Alternative ways to measure the value of AI investments could include focusing on non-financial metrics such as return on investment in employees (ROIE).
An ROIE would consider improvements in employee productivity and engagement, and perhaps acknowledge that AI’s value may lie in enhancing the overall efficiency and morale of the workforce, even if it doesn’t directly translate into immediate financial gains.
Business leaders have high expectations for AI’s ability to boost productivity. According to an Upwork study, 96% of C-suite leaders expect AI tools to increase their company’s overall productivity, with many companies already mandating or encouraging the use of these tools.
There are reasons for optimism in the face of those expectations. Kash Rangan, senior equity research analyst at Goldman Sachs, said “rays of hope” have already emerged across several domains that demonstrate AI’s productivity benefits.
Although AI technology is expensive, and “the human brain is 10,000x more effective per unit of power in performing cognitive tasks vs. generative AI,” Rangan said he believes that the cost equation will eventually shift, leading to new, unforeseen applications.