The ROI Illusion: Why Enterprise AI is Stalling, and the Execution Layer That Will Fix It
Why we're backing Modo - the AI success layer for the enterprise: translating cross-tool workflows into measurable productivity
The enterprise AI honeymoon is over, and the market is entering a phase of ruthless arithmetic.
Macro industry projections show enterprise AI investment scaling at an unprecedented trajectory over the next few years. Yet, beneath this massive capital deployment lies a structural disconnect. Despite historic spending, many organizations report seeing no meaningful, bottom-line impact from their software investments.
According to tracking from McKinsey, a staggering percentage of enterprises are struggling to realize direct business value from early AI pilots.
The standard enterprise playbook for the generative AI era has failed.
The cycle is predictably flawed: organizations hire Chief AI Officers (CAIOs) who are forced to operate blind, pay legacy consultants millions to identify theoretical use cases, and blindly buy premium enterprise licenses per employee without any real adoption strategy.
We are witnessing a massive misallocation of capital. The defining question for the next era of enterprise software is no longer whether to adopt AI, but how to force actual behavior change, measure the ROI, and integrate models into deeply entrenched, legacy workflows.
The Silent Bottlenecks of Enterprise AI
Industry research reveals that the hardest part of enterprise AI deployment is not the underlying technology—it is the human element. Pervasive tool abandonment rates and lagging adoption curve plateaus are hitting organizations within weeks of deployment.
As we evaluate the enterprise landscape, three distinct bottlenecks are choking value creation:
1. Identity Threat and the “Rewriting Paradox”
Employees whose professional identities are tied to specific expertise often experience the introduction of AI as an existential threat, resulting in active resistance or what researchers term “frozen expertise”.
In striking empirical examples, human workers have been observed consistently rewriting perfectly accurate AI suggestions out of habit or a need for control. You cannot fix this with a better model; it requires psychological safety and deep behavioral retraining.
2. The Middle Management Plateau
Middle management remains one of the most underestimated obstacles to systemic AI change. Department leaders often stall adoption to protect their team budgets, defend headcount, and maintain ownership of legacy processes.
Even when the C-suite mandates transformation, this layer acts as a hidden bottleneck, slowing procurement and suffocating pilot velocity.
3. The Data Readiness Ceiling
You cannot build reliable AI agents on top of inconsistent, unmapped processes. A massive volume of enterprise data remains completely unstructured, siloed, or stuck on legacy infrastructure.
Furthermore, large organizations built through M&A suffer from severe fragmentation, functioning as entirely separate sub-companies across different geographies. Without a clean data foundation and standardized processes, the ceiling for AI utility is remarkably low.
The Missing Layer: Visibility for the Non-Technical Enterprise
Management has long obsessed over productivity metrics, which is why engineering teams rely on massive visibility platforms. Yet, for the non-technical workers who make up the vast majority of the modern enterprise, there is absolutely no infrastructure to guide or measure their AI usage.
Employees are lost in a sea of fragmented point tools—unable to determine which model to use, where it fits into their specific workflow, or how to prompt it effectively. Meanwhile, leadership is flying blind, unable to calculate the exact ROI of their license spend or identify where to scale investment next.
Enterprises do not want their workflows entirely captured by a single walled garden, nor can they afford to keep burning token costs on disconnected applications.
The market desperately needs a neutral, cross-platform orchestration layer. It needs a system that can observe human workflows, guide behavior in real-time, and provide executives with concrete productivity math.
That is why Bouken is excited to back Modo.
Enter Modo: Behavior Change as the Precursor to Orchestration
Modo is building the AI success layer for the enterprise. Rather than selling a theoretical future of autonomous agents to companies that aren’t ready, Modo focuses on a highly surgical, immediate go-to-market wedge: visibility and behavior change.
Modo sits directly inside the employee’s workflow, capturing actions across daily tools . By analyzing user intent, Modo recommends the ideal purchased AI tool, suggests workflows, and enhances prompts precisely when the employee needs them.
“When we evaluated the enterprise market during our due diligence, it became obvious that purchasing software itself was not the hard part for businesses—driving real adoption was where the value was leaking,” notes Yasmine Zhu, Partner at Bouken Capital (Ex-McKinsey)
“Modo doesn’t just diagnose the adoption gap; it actively bridges it at scale. By embedding directly into the user’s natural workflow, it intercepts the moment of intent and turns passive software seats into active, measurable productivity.”
Furthermore, Modo solves the human adoption problem by operationalizing the “champion” model. Peer-led champions are the single most effective driver of sustained organizational adoption. Modo naturally identifies internal super-prompters and scales their best practices across the organization, turning local wins into company-wide defaults.
The Agentic Horizon and the “Smart Power”
At Bouken, our “Smart Power” thesis is built on the belief that the winners of this era will deliver practical, industrial utility that forces actual behavior change. Modo embodies this exact philosophy.
Modo’s observation layer is simply the beginning. Their context layer is the exact prerequisite needed for the agentic horizon. Executing a vision of this magnitude requires founders with immense grit, intellectual honesty, and high-velocity shipping capabilities. Johnny Chang and Angelina You represent a rare combination of technical depth and disciplined GTM execution. Johnny previously led AI research at Stanford and founded a global AI think tank for educators, while Angelina brings deep experience building large-scale ML models at Meta.
The enterprise software landscape is undergoing a permanent architectural shift. The era of buying fragmented point solutions and hoping for employee compliance is over; long-term productivity will belong to the orchestration platforms that bridge human intent with machine execution.
We are thrilled to participate in their journey and support Modo’s vision, alongside other institutional investors led by Pear VC and One Way Ventures. As organizations work to unlock the latent value of their technology investments, building a neutral, behavioral connective tissue is no longer optional—it is the prerequisite for operational success.
References
McKinsey & Company: The State of AI — Broad industry benchmarks on enterprise adoption, commercial impact, and ROI stagnation.
Gartner: Maverick Research on Generative AI Pitfalls — Macro perspectives on capital misallocation and enterprise procurement barriers in legacy sectors.
Harvard Business Review: Overcoming Human Resistance to Automation — Behavioral frameworks on identity threat, workforce change management, and the middle management bottleneck.