The Tokenmaxxing Hangover: Why Enterprises Are Still Searching for AI ROI
Not long ago, Silicon Valley was in the grip of a phenomenon known as tokenmaxxing — a philosophy that encouraged CEOs and executives to push their organizations' AI usage as far as it could possibly go. More prompts, more queries, more automation. The idea was simple: flood the zone with AI and let the productivity gains pay for themselves. Then the invoices arrived.
Uber reportedly exhausted its entire annual AI budget within a matter of months. Some companies quietly trimmed or eliminated Claude licenses for entire departments. Meta shut down an internal AI usage leaderboard that had been gamifying consumption. The message from the market was unmistakable — unbounded AI experimentation is expensive, and the returns are far from guaranteed.
Now, as the dust begins to settle, one of the most pressing questions in enterprise technology is the one that should have come first: What is AI actually worth to a business? Tiffany Luck, a partner at prominent venture capital firm NEA, argues that most enterprises are still a long way from having a satisfying answer.
The Gap Between Enthusiasm and Accountability
The early days of the generative AI wave were defined by speed and experimentation. Companies raced to deploy tools, sign enterprise agreements, and announce AI initiatives to investors and boards. The competitive pressure to "do something with AI" was enormous, and for many organizations, that urgency came at the expense of rigorous measurement.
What got lost in the rush was a clear framework for evaluating success. Unlike traditional software investments, where metrics like uptime, cost-per-seat, or transactions-per-second offer relatively clean benchmarks, AI investments introduce a far murkier set of variables. How do you quantify the value of a sales rep who drafts emails faster? How do you isolate the ROI of a customer service chatbot that deflects some tickets but frustrates others?
These are not rhetorical questions. They are the exact challenges that finance teams and CIOs are wrestling with right now, often without good tooling or precedent to guide them. The enterprise AI stack has matured quickly, but the measurement layer has not kept pace.
What Tiffany Luck and NEA Are Watching
From her vantage point at NEA, Luck has a broad view of how enterprises across sectors are navigating this challenge. Her perspective reflects what many in the venture and enterprise software world are beginning to acknowledge: the first wave of AI adoption was about access, and the second wave will be about accountability.
Companies that deployed AI broadly in 2023 and 2024 are now entering a maturation phase. Budget owners are asking harder questions. Procurement teams are scrutinizing renewals. And the vendors who will win in this environment are those who can demonstrate measurable, repeatable value — not just impressive demos.
This shift has significant implications for the AI tools market. Products that integrate seamlessly into existing workflows, that surface usage analytics, and that tie outputs to business outcomes will have a meaningful advantage. Novelty is no longer a differentiator. Impact is.
Why Measuring AI ROI Is Uniquely Difficult
Part of what makes enterprise AI ROI so elusive is the nature of the technology itself. Traditional software automates defined processes with predictable outputs. AI, particularly large language models, operates probabilistically. Results vary. Outputs require human review. Errors compound in ways that are sometimes invisible until they cause real damage.
There are at least three structural reasons why enterprises continue to struggle with AI ROI measurement:
- Attribution is complex. AI tools often augment human work rather than replacing it outright, making it genuinely difficult to isolate the contribution of the technology versus the employee. A lawyer who uses AI to research case law faster is more productive — but by how much, and how do you measure the quality delta?
- Adoption is uneven. Even when licenses are purchased at scale, actual usage within organizations varies dramatically. A tool that is theoretically available to five hundred employees may be actively used by fifty. Average cost-per-user calculations can look alarming when adoption is factored in.
- Benefits are often soft. Many of the genuine gains from AI — reduced cognitive load, faster onboarding, improved morale through less tedious work — are real but notoriously hard to convert into a dollar figure that satisfies a CFO.
The Path Forward: From Experimentation to Value Architecture
The enterprises that will emerge from this transitional period in the strongest position are those that approach AI investment the same way they approach any other major capital allocation decision — with defined objectives, measurable outcomes, and honest post-mortems.
That means moving away from sprawling, company-wide rollouts in favor of targeted deployments where the value hypothesis is clearer. A legal team automating contract review, a finance team accelerating month-end close, a support organization reducing average handle time — these are use cases where before-and-after comparisons are achievable and defensible.
It also means investing in the infrastructure to track AI usage and outcomes systematically. Organizations that build internal AI governance functions now, complete with usage dashboards and value tracking, will be better positioned to make confident investment decisions in the next cycle of the technology.
The Bottom Line
The tokenmaxxing era exposed a fundamental truth about enterprise technology adoption: enthusiasm without accountability is just expensive optimism. As Tiffany Luck and the broader investment community watch the AI market mature, the enterprises that will define the next chapter are those willing to slow down long enough to ask — and answer — the hard question of what their AI investment is actually delivering. The companies that get this right will not just survive the ROI reckoning. They will be positioned to pull decisively ahead of those that do not.