Other Mental Models · OM-20
A system that only ever makes small, incremental improvements can get permanently stuck at a decent-but-not-best solution — because every small step away from it looks like a step backward, even though a genuinely better solution exists further away.
A concept from optimization theory distinguishing a local optimum — a solution that is better than all its immediate neighboring alternatives, but not necessarily the best solution available overall — from a global optimum, the actual best solution across the entire space of possibilities. A system that only ever makes small, incremental improvements can become permanently stuck at a local optimum, unable to reach a superior global optimum that requires temporarily moving to a worse position first.
A foundational concept in mathematical optimization and computer science, with the underlying hill-climbing/gradient-based search problem formally studied since at least the mid-20th century, and widely applied metaphorically across business strategy, evolutionary biology, and machine learning.
The Mechanism
Every direction from here looks worse — but a better peak exists, reachable only by first descending
A hiker trying to reach the tallest peak in a mountain range, using only the simple rule 'always take a step upward,' will reliably get stuck at the top of whichever hill they started nearest to — even if a taller peak exists elsewhere in the range, reaching it would require first walking downhill into a valley, a move the simple always-improve rule never permits, illustrating exactly how an incremental-only optimization process gets trapped.
01 · PURELY INCREMENTAL, GREEDY IMPROVEMENT PROCESSES ARE STRUCTURALLY VULNERABLE TO GETTING STUCK
This is a mathematical property of the search method, not a flaw specific to any one domain
Any optimization process that only ever accepts moves which immediately improve the current position (a 'greedy' or 'hill-climbing' approach) is mathematically guaranteed to be vulnerable to stopping at a local optimum whenever the solution space isn't perfectly smooth — this is a structural property of the search method itself, which is why more sophisticated optimization techniques deliberately build in some tolerance for temporarily worse moves.
02 · ESCAPING A LOCAL OPTIMUM OFTEN REQUIRES DELIBERATELY ACCEPTING A TEMPORARY STEP BACKWARD
This is precisely the move a purely incremental process cannot make on its own
Techniques designed to escape local optima — simulated annealing, genetic algorithms with mutation, deliberately exploring 'worse' options in strategy or R&D — all share the same core mechanism: deliberately tolerating some temporary decline in performance in order to explore a wider region of the solution space that a purely improvement-only process would never reach.
03 · IT EXPLAINS WHY SUCCESSFUL ORGANIZATIONS CAN GET STUCK OPTIMIZING A GOOD-BUT-NOT-BEST BUSINESS MODEL
A well-documented pattern in business strategy and organizational behavior
An organization that only ever makes incremental improvements to its existing, already-successful business model or product can become permanently stuck at a local optimum — every small deviation from the current model looks like a step backward in the short term, even when a fundamentally different model would eventually reach a substantially better outcome, a pattern closely related to why disruptive innovation frequently comes from new entrants rather than established incumbents.
Where It Fails / Inversion
Where it fails / inversion
Not every local optimum is worth abandoning — the effort, risk, and temporary decline required to search for a global optimum has a real cost, and for many practical purposes, a good local optimum reached efficiently and reliably is a better actual outcome than gambling significant resources chasing an uncertain, possibly nonexistent, better global optimum.
How To Use It
Worked example · deciding whether to overhaul a working but suboptimal business process
A team that has incrementally optimized an existing process to a genuinely solid but plateaued level of performance should periodically ask whether a fundamentally different approach — one that would require accepting worse short-term performance during a transition — might reach a substantially better long-run outcome, rather than assuming the current plateau represents the best achievable result simply because no small incremental change currently improves it.
How to use it
When a system or process seems to have plateaued despite continued incremental effort, consider whether it has reached merely a local optimum rather than the actual best achievable outcome — escaping a local optimum typically requires deliberately tolerating a temporary decline in performance, a move that pure incremental improvement will never make on its own.
See Also