Most optimization projects never deliver their promised value. After years of working on these initiatives across railways, energy grids, and production systems, I've identified the patterns that separate successful projects from expensive failures.
The typical optimization project starts with excitement. Someone reads about a competitor saving millions through mathematical optimization. A consulting firm presents a compelling business case. Management approves the budget. And then, 18 months later, the project quietly dies—or worse, delivers a system nobody uses.
Why does this keep happening?
Problem 1: Wrong Problem Definition
The most common failure mode is solving the wrong problem. Teams optimize what's easy to model rather than what actually matters. They build sophisticated algorithms for scheduling decisions that represent 5% of costs while ignoring the messy, hard-to-formalize decisions that drive 80% of value.
The fix: Start with the business problem, not the math. Spend time understanding where decisions actually create or destroy value. Often the most impactful opportunities aren't the most mathematically elegant.
Problem 2: Data Quality Delusions
Every optimization project assumes the data will be clean. It never is. Historical records are incomplete. Real-time feeds have gaps. Master data is inconsistent across systems. The algorithm needs inputs that don't exist yet.
The fix: Do a data audit before committing to the project. Assume data quality will be worse than expected. Build data pipelines as a first-class project deliverable, not an afterthought.
Problem 3: Ignoring the Human Element
Optimization systems don't operate in a vacuum. They produce recommendations that humans must understand, trust, and act on. Too often, projects focus entirely on algorithmic performance while ignoring user experience.
If operators don't understand why the system recommends a particular action, they won't follow it—especially when the recommendation contradicts their intuition. And sometimes their intuition is right, because the model is missing context.
The fix: Design for explainability from day one. Involve end users throughout development. Build systems that augment human judgment rather than trying to replace it entirely.
Problem 4: Scope Creep
Projects that try to optimize everything end up optimizing nothing. The temptation to add "just one more constraint" or "extend to another use case" is constant. Each addition seems small, but together they transform a focused initiative into an unwieldy monster.
The fix: Define a minimal viable optimization scope. Launch something that works for one specific use case. Expand only after proving value in production.
What Successful Projects Do Differently
The projects that work share common traits: they start small, they obsess over data quality, they involve operators from the beginning, and they measure success by adoption, not algorithmic sophistication.
Mathematical optimization is powerful. But the math is rarely the hard part. The hard part is everything around it—the change management, the data infrastructure, the trust-building with users who've seen similar projects fail before.
Get those things right, and the optimization almost takes care of itself.
Written by
Jonasz Staszek