In 1997, I received a copy of The Goal by Eliyahu Goldratt. The book presented a foundational principle: a system can only move as fast as its slowest constraint.
Twenty-eight years later, I built a company around that idea.
Optimize the Constraint, Not Everything
Most organizations fail at optimization by trying to improve everything simultaneously. This is wrong. Identify the single bottleneck limiting system throughput. Focus there.
Modern AI tools frequently violate this principle through universal optimization approaches. They optimize everywhere because they can, not because they should.
The Constraint Keeps Moving
Constraints are dynamic. Resolving one bottleneck inevitably reveals another. Organizations typically treat constraint identification as a one-time exercise rather than a continuous process.
Demand shifts. Suppliers adjust capacity. Markets evolve. The constraint that limited you last quarter may not be the constraint limiting you now.
Twenty Years of Circling the Same Question
Despite decades in supply chain and enterprise systems across industries (Pfizer, Gilead, AstraZeneca, BASF, Syngenta), I repeatedly encountered the same problem: organizations possessed constraint data but lacked real-time systems to identify current bottlenecks.
Instead, they relied on tribal knowledge held by experienced operators. When those operators left, the knowledge walked out with them.
The Gap Between Knowing and Doing
Theory of Constraints principles remain valid forty years after publication. Implementation has not evolved significantly. Traditional TOC relies on manual identification, static analysis, and periodic reviews rather than continuous, real-time constraint detection.
The theory is sound. The tooling is not.
What If the Constraint Could Find You?
In 2025, I started building ChainAlign: continuous, probabilistic, and proactive constraint identification using live ERP and MES data.
The evolution progressed through phases:
- Classical constraint logic from Goldratt’s original framework
- Probabilistic modeling via Monte Carlo simulation
- Bayesian reasoning for causal understanding
- AI-enhanced optimization with counterfactual analysis
Learning From Disagreement
The system incorporates a Socratic Inquiry Engine that learns from user disagreements. When an operator overrides a recommendation, we ask why. Their reasoning becomes institutional knowledge that persists beyond any individual tenure.
Human judgment is not a bug to be engineered out. It is signal to be captured.
After thirty years, I remain engaged with Goldratt’s original insight. The difference: now it runs continuously in software, surfacing constraints before they become crises.
Originally published on pramod.ch