The $700 Billion AI Productivity Problem No One's Talking About
Russ Fradin, founder of Larridin, draws parallels between the early days of ad tech and the current AI boom, arguing that a robust measurement infrastructure is the missing piece to unlock AI's true enterprise value. He discusses the challenges of measuring AI ROI, the gap between AI spending and actual usage, and how to overcome employee anxiety to foster productive adoption. The key is to move beyond simple surveys and marry behavioral data with real-world outcomes to understand if AI tools are actually making companies more productive.
Russ Fradin, drawing from his experience in the early days of web advertising, posits that the current AI landscape is missing a critical component that enabled the ad tech boom: a robust measurement and attribution infrastructure. Just as companies like Comscore and Nielsen built the tools to validate ad spend on the internet, a similar stack is needed for AI to prove its value and accelerate enterprise adoption. Without it, companies are flying blind.
The Enterprise AI Dilemma: Spending Without Seeing
Enterprises are caught in a perfect storm of opportunity and anxiety. A survey of 350 IT heads revealed a strong belief that companies have only 18 months to become an AI leader or fall behind. This has unlocked massive budgets, with leaders admitting they are sure a significant portion—as high as 70%—of this AI spend is being wasted. The core issue is a lack of systems to measure success. As one executive noted about their board meetings, “For my other four metrics, I have some report of how are we doing… And on AI, all I have is the amount of stuff we bought.”
The Challenge of Measuring AI Productivity
Measuring the return on investment (ROI) for AI is complex, plagued by several core challenges:
- The Baseline Problem: When a tool like Harvey makes lawyers more productive, the immediate question is, “Compared to what?” Without a clear baseline, measuring improvement is difficult. The productivity of an individual is inherently unknowable; therefore, analysis must happen at an aggregate level.
- Goodhart’s Law: The principle that “when a measure becomes a target, it is no longer accurate as a measure” is a major pitfall. If productivity is measured by lines of code or emails sent, employees will game the metric, corrupting the data.
- Surveys vs. Behavioral Data: Traditional methods, like surveying employees on their perceived productivity, are unreliable. Employees may not actually be using the tools they claim to, or they may feel pressured to report positive results. A more effective method is to marry survey data with passive, behavioral data. This allows for a comparison between the productivity of heavy, light, and non-users of a specific AI tool.
- Real-World Metrics: A powerful, less gameable metric is interdepartmental responsiveness. If the legal team starts using an AI tool, does it lead to other departments feeling more comfortable sending them work? Are they getting responses faster? This behavioral shift is a strong indicator of a genuine productivity gain.
Fostering Adoption: From Fear to a “Safe Space”
A significant barrier to adoption is employee anxiety. Many employees, particularly those who are not early adopters, fear looking foolish or getting fired for using AI incorrectly—for example, by accidentally uploading sensitive data. This leads to a situation where the most innovative employees often hide their AI usage, preventing their valuable workflows from being shared across the organization.
To counter this, companies must create a “safe space” for AI experimentation. Larridin’s approach involves a “harness” or wrapper around LLMs that acts as a guardrail. It blocks users from submitting prompts that are illegal (e.g., in the EU, using AI for employee reviews) or violate company policy (e.g., uploading personally identifiable information). This reassures employees that they can explore AI tools without risk, which in turn drives usage and allows the company to build an internal IP of effective, tailored AI use cases.
The Future of Work: Augmentation Over Replacement
Fradin is skeptical of predictions of mass job loss. He argues from a competitive, capitalist standpoint: a company that fires half its staff to boost margins will be outcompeted by a rival that uses AI to do more with its existing workforce. As Jeff Bezos said, “Your margin is my opportunity.” Competitors will leverage AI to increase output and capture market share, not just cut costs.
Ultimately, the diffusion of AI into the workplace is a “product marketing problem.” A tool that “does everything” is a poor pitch. Success comes from identifying specific, high-value use cases—the “tip calculators” that solve a concrete problem for a specific role, such as AI for coding or for legal contract review. By focusing on these specific applications and building the infrastructure to measure their impact, enterprises can finally begin to realize the true competitive advantage of AI.