Crossing the GenAI Divide: Lessons from MIT's 2025 State of AI in Business Report

The latest MIT report on Generative AI in enterprise settings has generated headlines for one striking statistic:

95% of enterprise GenAI projects show no measurable impact on P&L.

While this headline captures attention, it risks obscuring the deeper insights. The report doesn't simply tell us that "AI isn't working." It explains why most projects fail, and what differentiates the few that succeed.

At Medusa, we see this study as both a reality check and a roadmap. Here are the key lessons that matter most for boutique investment firms.

1. The Real Barrier Is the Learning Gap

The report makes clear that the main obstacle is not models, regulation, or budgets. It's the learning gap.

Most AI tools:

• •Don't adapt to context
• •Don't integrate into workflows
• •Don't improve from feedback

Not surprisingly, they stall once the pilot excitement fades.

A common example is generic chatbots. They demo well because they're flexible and familiar. But when it comes to mission-critical workflows, they fall short: lacking the memory, adaptability, and guardrails needed for production environments.

2. Why Pilots Stall, and How to Avoid It

MIT highlights a steep "pilot-to-production chasm." Enterprises are quick to experiment, but only 5% of task-specific AI tools reach production.

The reasons: brittle integration, lack of memory, poor fit with day-to-day operations.

The lesson? Success requires systems that:

• ✓Learn from user interactions
• ✓Integrate directly with existing processes
• ✓Improve over time

For boutique firms, this means avoiding "AI experiments" that look good in isolation but don't reduce real operational bottlenecks.

3. Where the ROI Really Is

Another overlooked insight: the highest ROI often comes from back-office and oversight functions, not front-office experiments.

• •Risk management: continuous monitoring, stress-testing, anomaly detection
• •Reporting: automated client reports with institutional polish
• •Operations: reducing reliance on outsourced BPO or agency work

These may not sound glamorous, but they deliver measurable savings and free up scarce team bandwidth.

By contrast, most budgets today still over-index on sales and marketing use cases. Precisely where impact is hardest to prove.

4. Why Mid-Sized Firms May Have the Advantage

Interestingly, MIT finds that mid-market companies often outperform large enterprises in crossing the divide. While corporates get stuck in nine-month pilot cycles, mid-sized firms move from pilot to deployment in ~90 days.

For boutique hedge funds and family offices, this is an encouraging signal: being lean is an advantage. Smaller teams can adopt, adapt, and scale practical AI faster than global incumbents.

5. The Shift from Static Tools to Agentic Systems

Finally, the report underscores the growing importance of Agentic AI: systems with persistent memory, contextual learning, and autonomous workflow orchestration.

These are not chatbots or wrappers. They are embedded co-pilots that can:

• •Monitor portfolios 24/7
• •Generate risk diagnostics instantly
• •Automate reporting with built-in guardrails
• •Orchestrate multiple processes across systems

In other words: the kind of learning-capable, workflow-native AI agents that will define the next generation of investment operations.

Closing Thoughts

The MIT report is a reminder that adoption does not equal transformation. Most AI projects will fail to move the needle, unless they are designed to learn, integrate, and deliver measurable outcomes.

For boutique investors, this presents a unique opportunity. By focusing on practical, learning-capable systems in high-ROI workflows, smaller firms can cross the GenAI divide faster than their larger peers, and level the playing field with institutional giants.