NEA’s Tiffany Luck says enterprises are still figuring out their AI ROI

The recent surge of AI enthusiasm, exemplified by the now-waning trend of "tokenmaxxing," highlights a critical inflection point in enterprise adoption. The initial rush to leverage generative AI, fueled by CEO mandates to push usage to its limits, has quickly collided with the stark reality of cost and, more importantly, demonstrable return on investment. Reports of companies like Uber exceeding their AI budgets within months, alongside internal license cuts and scrapped initiatives like Meta's leaderboard, paint a picture of premature exuberance and a lack of strategic grounding. This isn't to say AI lacks value; rather, it underscores the need for a more measured and pragmatic approach. We've previously explored the pitfalls of over-investment in hardware, as seen with Snap’s recent AR glasses launch After unveiling ridiculously expensive AR glasses, Snap’s stock takes a dive, demonstrating that even compelling technology needs a solid business case. The current AI ROI reckoning reinforces this lesson, suggesting a broader shift away from chasing shiny objects towards building focused, practical applications.

The core issue isn't the technology itself, but the lack of clear objectives and robust measurement frameworks. Many organizations jumped into AI implementation without fully understanding *what* problems they were trying to solve, or *how* they would quantify success beyond simply increasing token consumption. The "build it and they will use it" mentality, common in the early days of the internet, has resurfaced in the AI space. As our recent piece on agent frameworks pointed out You Probably Don’t Need an Agent Framework, complex architectures aren't always necessary – often, a well-defined workflow built with simpler tools delivers superior results. This resonates strongly with the current situation: rather than sprawling, resource-intensive AI experiments, enterprises are likely to refocus on targeted applications that directly address specific business needs, and where ROI can be readily tracked. The messy, unglamorous reality of data collection for physical AI, as detailed in our piece on XDOF Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it further highlights the operational groundwork required for successful AI deployment – a grounding in practical reality that often gets overlooked in the hype cycle.

This correction isn’t necessarily negative. It’s a crucial evolution, forcing businesses to move beyond the initial excitement and develop more sustainable AI strategies. It signals a maturation of the market, where vendors will be increasingly held accountable for delivering tangible value, and where users will demand greater transparency around costs and performance. We're likely to see a rise in specialized AI solutions tailored to specific industries and use cases, rather than generic, all-purpose models. Furthermore, the focus will shift from simply deploying AI to effectively integrating it into existing workflows and ensuring that it empowers employees, rather than replacing them wholesale. This requires a deeper understanding of human-AI collaboration and a willingness to invest in training and change management.

Ultimately, the current AI ROI reassessment compels a critical question: how can organizations move beyond the experimentation phase and build a framework for demonstrating the true, long-term value of AI? The answers likely lie in a combination of clear strategic objectives, rigorous performance measurement, and a willingness to prioritize practical applications over flashy demonstrations. The future of AI in the enterprise isn’t about maximizing token usage; it's about intelligently leveraging AI to unlock tangible business outcomes, and that requires a more nuanced and disciplined approach.