Joko Widodo
To understand this transition, it’s crucial to differentiate between an invention and an innovation. AI, in its current powerful form, is undoubtedly an invention – a new capability that opens up previously unimaginable possibilities. However, an invention does not become an innovation until it is integrated with a viable business model, consistently delivering measurable value that transforms operations or creates new markets. Last year’s flurry of experimentation was, in hindsight, a sensible first step in exploring this invention. Companies needed to understand what AI could do, how it could be applied, and what its limitations were. The challenge now is to move beyond mere capability and into concrete, demonstrable innovation.
It is becoming increasingly clear that the path to true AI innovation lies in entrusting AI systems with real, impactful decisions. This vision aligns perfectly with what management guru Peter Drucker would have termed "executives" – entities capable of making autonomous, goal-oriented decisions. In contemporary terms, these are referred to as agentic AI: intelligent systems designed to act independently to achieve specified objectives, often by breaking down complex tasks into smaller steps, reasoning through problems, and interacting with their environment. These agentic systems represent the next frontier, moving AI from a mere tool to a proactive, decision-making partner within an organization.
As we grapple with the pressing question, "Is this thing working?", we can draw invaluable insights from one of Drucker’s most influential intellectual disciples: Andy Grove. The legendary former CEO of Intel, Grove masterfully translated Drucker’s philosophical writings into a hard-nosed, pragmatic approach for managing knowledge-worker organizations. His seminal work, "High Output Management," provides the classic framework for measuring the outputs of middle managers – a task notoriously difficult but one Grove insisted could and must be done. Grove’s philosophy is particularly relevant now because agentic AI, in essence, functions as a highly sophisticated knowledge worker. Just as we demand measurable outputs from human managers, we must demand the same from our AI counterparts.
The fundamental shift required is to move our focus away from activities, anecdotes, and initiatives. These are merely inputs – the effort, the tools, the processes. Grove argues forcefully that organizations must instead fixate on outputs. If we adopt Grove’s rigorous mindset, our first step would be to clearly define the specific business outcome we aim to achieve. Only then can we measure our agentic AI’s success solely by its ability to improve that defined performance metric. This demands a discipline that many organizations have yet to apply to their AI endeavors, often contenting themselves with reports of "usage" or "engagement" rather than bottom-line impact.
This emphasis on measurable outputs resonates deeply with a mathematical approach to problem-solving. Several years ago, while working on this across our software portfolio at Strattam, I had the great fortune to encounter Dario Fanucchi, a brilliant mathematician. Dario was already using AI to solve real-world problems in a strikingly similar fashion. As co-founder and CTO of Isazi, a decade-old firm comprising over 70 mathematicians and engineers, Dario and his team have completed hundreds of projects for leading companies globally, all with a singular focus: improving core business metrics.
Isazi arrived at the same conclusion about measuring outputs, albeit from the rigorous field of mathematics rather than organizational behavior. Their approach frames AI projects as mathematical optimization problems. This involves a precise methodology:
- Define a target measure: Identify a critical business metric that needs improvement, such as increased throughput, reduced working capital, enhanced customer retention, accelerated sales conversion rates, or optimized supply chain efficiency.
- Identify influencing variables: Determine all the internal and external factors and variables that directly or indirectly impact this target measure.
- Model the mechanism: Create a mathematical or algorithmic model that precisely describes how these variables interact to influence the target measure. This model forms the basis for the AI’s decision-making process.
Once this framework is established, all AI initiatives are meticulously aligned to this single target measure. Success is then unequivocally determined by the demonstrable improvement in that metric. This approach aligns seamlessly with how AI models are inherently built and refined: benchmarks and evaluations are always the core measures of an AI system’s efficacy. Here, these technical evaluations are directly and inextricably linked to overarching business metrics, ensuring that technical prowess translates directly into commercial value.
The critical insight here is that you must begin with the output you want to measure. This isn’t an afterthought; it’s the foundational principle. Once established, you must monitor that output measurement with the vigilance of a hawk, treating it like a gauge on a critical machine. Observe how long it takes for the gauge to register changes, quantify the magnitude of those changes, note the direction of movement (positive or negative), and, crucially, assess whether the improvement is sustained over time.
The duration required to observe and sustain a material, positive movement in the target output is what we term "Time To Production" (TTP). Our theory on why so many AI pilots falter is precisely because companies often reverse this logical sequence. They tend to first select an AI tool, then define a pilot duration, and finally, at the end of that period, conduct qualitative check-ins with users. This subjective, input-focused approach rarely yields clear, quantifiable business results, leading to the widespread disappointment we see today.
While we at Strattam and Isazi certainly value experimentation and pilots, our experience shows that optimal results emerge when this process is inverted. We first articulate the specific, measurable output we intend to improve. Then, we systematically vary the AI tools, methodologies, and approaches until one demonstrably moves the dial in the desired direction. Our success is then measured by the Time To Production – the duration it takes to positively and sustainably alter that critical business output. The shorter the Time To Production, the more efficient and impactful the AI implementation.
To illustrate this principle, consider a real-world example from one of Strattam’s portfolio companies, Trax Technologies. Trax specializes in helping very large multinational corporations manage their complex global shipping logistics. A cornerstone of their service offering is ensuring that freight bills are meticulously complete, accurately match contractual agreements, are properly approved for payment, and are correctly accounted for.
Trax operates across every geography and all shipping modes, interacting with thousands of carriers. Given this immense scale and complexity, discrepancies between freight bills and shipper contracts – known as "exceptions" – are remarkably common. Historically, managing and resolving these exceptions at scale has required a substantial in-house team of human experts. In 2024, Trax astutely identified AI’s potential to automate and streamline the resolution of a significant portion of these exceptions. This led to the in-house development of their AI Audit Optimizer. The output goal for this new system was exceptionally clear and quantifiable: to increase the fraction of exceptions resolved without any human intervention.
The initial deployment of the AI Audit Optimizer showed promise. In the first quarter following its release, the system successfully resolved approximately 826,000 exceptions that would otherwise have demanded human attention. This was a good start, providing early validation, but it wasn’t yet a game-changer. However, in the second quarter, the system’s performance plateaued, remaining stuck at roughly the same resolution level. This stagnation triggered Trax’s agile response: rapid experimentation to identify what factors could improve outcomes.
The breakthrough came in Q3. The company discovered that incorporating a human prompt engineer to interact strategically with the AI system made a dramatic difference. This human-in-the-loop approach, where an expert guided and refined the AI’s understanding and decision-making, proved transformative. As a direct result, in Q4, the number of resolved exceptions tripled to an impressive 2.5 million. Now, that’s a level of impact worth celebrating.
With the output gauge firmly in mind, Trax continues to refine its AI Audit Optimizer. They are adjusting the interaction points between the prompt engineer and the system, leveraging data from both successful and unsuccessful resolutions to continuously retrain and improve the AI model. The company has also established aggressive quarterly goals, aiming to surpass previous resolution records each quarter. This iterative process, driven by clear output measurements, has allowed Trax to fine-tune and adapt its AI tooling to deliver outcomes that genuinely matter. By fixing its customers’ problems more efficiently, Trax is strengthening its market position, and its rigorous output measurements unequivocally prove the real-world value of this AI innovation.
In the midst of all the pervasive hype surrounding AI, what truly matters is that our companies genuinely adapt, consistently deliver measurable customer value, and ultimately succeed. We understand intellectually that the status quo is unsustainable, and that our future prosperity may hinge on our ability to adapt with agility. But mere understanding is not synonymous with actual adaptation.
To successfully navigate this transformative era, we must resist the seductive urge to simply acquire new tools, run vague pilots, share anecdotal successes, and report on mere activities. These are, fundamentally, just inputs. Instead, we must rigorously determine the precise outcome measurement that holds the most significance for our business. Then, we must watch that output like a hawk, constantly scrutinizing whether our AI initiatives are delivering cold, hard business results. If the dial isn’t moving, or isn’t moving enough, we must be prepared to change our AI approach, our tools, or our strategy until that gauge unequivocally registers positive and sustained improvement. By drawing upon the time-tested wisdom of management titans like Peter Drucker and Andy Grove, and by adopting a disciplined, output-focused methodology, we can ensure that AI truly earns its keep at our firms, transforming from a mere invention into a powerful engine of sustained innovation and competitive advantage.
Bob Morse co-founded Strattam Capital in 2014 and is managing partner. He has served on numerous private and public technology company boards, and currently is a director of CloudHesive, Contegix, Daxtra Technologies, Green Security, Resource Navigation and Trax Group. Previously, he was a partner and member of the investment committee at Oak Hill Capital Partners. He also worked at GCC Investments and Morgan Stanley. Morse serves on the board of directors of Austin PBS and as member of the advisory board for the HMTF Center for Private Equity Finance at The University of Texas at Austin McCombs School of Business. He attended Princeton University, graduating summa cum laude with a B.S.E., and Stanford Graduate School of Business, where he earned his MBA and was an Arjay Miller Scholar. Morse lives in Austin.
Dario Fanucchi contributed to this article. He is chief technology officer at Isazi, a Johannesburg-based applied artificial intelligence firm purpose-built to deliver production-grade AI software solutions for clients. Fanucchi has excelled academically in the fields of computer science, mathematics and physics throughout his career.
Illustration: Dom Guzman
