Global & Local Maxima Mental Model… | Faster Than Normal
Mathematics & Probability
Global & Local Maxima
The distinction between the best possible outcome and merely the best nearby outcome — getting trapped at a local maximum prevents reaching the global one.
Model #0090Category: Mathematics & ProbabilityDepth to apply:
Imagine you're hiking in dense fog. You climb until every direction slopes downward. You've reached the highest point you can find — but you can't see the mountain range. You might be standing on the tallest peak for miles, or you might be on a foothill while the true summit towers somewhere beyond the cloud line. This is the difference between a local maximum and the global maximum, and it is one of the most consequential distinctions in decision-making.
A local maximum is the best outcome reachable through incremental improvements from your current position. A global maximum is the best outcome available across the entire landscape of possibilities. The two are almost never the same place. The tragedy of most careers, companies, and strategies is that they optimise relentlessly for the local peak — making things better within the current framework — without ever asking whether an entirely different framework would yield a fundamentally higher ceiling.
The concept originates in mathematical optimisation. Given a function with multiple peaks and valleys — what mathematicians call a non-convex landscape — gradient ascent algorithms climb by always moving in the direction of steepest improvement. They are guaranteed to find a local maximum: a point where no small step in any direction produces a better result. They are not guaranteed to find the global maximum, because reaching it may require descending first — getting worse before you can get better. The algorithm has no mechanism for that descent. Neither do most humans.
The formal mathematics traces to the calculus of variations developed by Euler and Lagrange in the eighteenth century, and to the optimisation theory that emerged from operations research during World War II. George Dantzig's simplex method (1947) could find global optima for linear problems, but for nonlinear, non-convex problems — which describe nearly every interesting real-world situation — no deterministic algorithm guarantees finding the global maximum in polynomial time. The problem is NP-hard in general. This means that in complex landscapes with many local peaks, finding the absolute best solution is computationally intractable. You can find a good solution. You can rarely prove it's the best.
The breakthrough insight came from the physical sciences. In 1983, Kirkpatrick, Gelatt, and Vecchi published "Optimization by Simulated Annealing" in Science, borrowing a metallurgical technique to solve combinatorial optimisation problems. The idea: heat a metal to high temperature (where atoms move freely and explore many configurations), then cool it gradually (allowing the system to settle into a low-energy crystalline structure). Translated into optimisation: start with high randomness, accepting worse solutions early to escape local traps, then progressively reduce randomness as the search converges on the global optimum. The algorithm works because it trades short-term degradation for long-term discovery. It gets worse on purpose so it can eventually get better.
This is precisely what humans and organisations find hardest to do. Loss aversion — Kahneman and Tversky's foundational finding from prospect theory — means the psychological pain of descending from a local peak is felt roughly twice as intensely as the potential gain from reaching a higher one. The certain loss of abandoning a working strategy looms larger than the uncertain gain of a potentially superior one. The result is systematic entrapment at local maxima: individuals stay in careers that are good but not great, companies iterate on products that are profitable but not transformative, investors hold positions that are adequate but not optimal. Each incremental improvement feels like progress. The landscape beyond the valley remains unexplored.
The concept maps cleanly onto fitness landscapes — a framework introduced by Sewall Wright in 1932 to describe evolutionary dynamics. Wright imagined a multidimensional landscape where each point represents a combination of genetic traits and the height represents the organism's fitness. Evolution, like gradient ascent, climbs toward local peaks through incremental mutation and selection. But the fittest possible organism might require a combination of traits that can only be reached by passing through less-fit intermediate forms. Wright showed that small, isolated populations were more likely to discover global maxima because genetic drift — random fluctuation — could push them off local peaks and into valleys that might lead to higher ground. Large populations, by contrast, were trapped on local peaks by the sheer weight of their own optimisation. The evolutionary insight maps directly onto corporate strategy: large, successful organisations are the most likely to be trapped at local maxima, because their size and success make the descent into uncertainty feel irrational.
Stuart Kauffman extended Wright's framework in the 1990s with his NK fitness landscape model, demonstrating that as the number of interacting variables (K) increases, the landscape becomes increasingly rugged — more local peaks, steeper valleys, and greater distance between local and global optima. The more complex the system, the harder it is to find the global best. A startup with three variables (product, market, pricing) navigates a relatively smooth landscape. A multinational corporation with hundreds of interdependent variables (product lines, geographies, regulatory environments, organisational structures, partner relationships) navigates a landscape so rugged that the global maximum may be unreachable through any sequence of incremental moves. The only path to the highest peak may require a discontinuous leap — what business strategists call a pivot and what Kuhn called a paradigm shift.
The concept shows up with different vocabulary across every domain that matters. In machine learning, it's the problem of gradient descent converging to suboptimal solutions — the reason researchers use techniques like random restarts, momentum, and learning rate scheduling to shake models out of local minima. In career theory, it's the "golden handcuffs" problem — high-paying positions that are locally optimal but globally suboptimal because they prevent exploration of higher-ceiling trajectories. In product strategy, it's feature-factory syndrome — teams that optimise an existing feature set because the metrics are clear and the improvements are measurable, while the fundamentally better product architecture remains unbuilt because nobody wants to reset the dashboard to zero. In investment theory, it's the disposition effect — the tendency to hold losing positions too long and sell winning positions too early, which Hersh Shefrin and Meir Statman documented in 1985, and which functions as a local-maximum trap in portfolio construction.
The practical power of the model is diagnostic. It forces a question that incremental thinking never asks: is the ceiling on my current trajectory high enough? You can optimise a horse-drawn carriage for decades — lighter materials, better suspension, faster horses — and never arrive at the automobile. Every improvement is real. Every improvement is progress. And every improvement takes you further from the insight that the entire framework should be abandoned. The local maximum of horse-drawn transport is infinitely below the global maximum of combustion-engine transport, and no amount of optimisation within the first paradigm reaches the second. You have to descend — abandon the working system — and cross a valley of uncertainty before the higher peak becomes accessible.
Section 2
How to See It
The signature of a local maximum is optimisation that produces diminishing returns. When you're near the top of your current peak, small improvements get harder and the gains get smaller. The landscape flattens out. This is the signal most people miss — they interpret slowing improvement as evidence that they're approaching the limit of what's possible, when they're actually approaching the limit of what's possible from where they stand.
The second signal is effort-outcome divergence: you're working harder than ever and results are flattening. Early on a peak, effort translates efficiently into improvement. Near the summit, the same effort yields marginal gains. Engineers adding the tenth performance optimisation to a mature codebase, salespeople refining the twentieth iteration of a pitch deck, investors tweaking the fifteenth parameter of a quantitative model — each is experiencing the diagnostic pattern of a local maximum: maximum effort, minimum improvement. The question is whether a fundamentally different approach would restore the effort-to-outcome ratio by placing you on the slope of a different, higher hill.
Career decisions
You're seeing Global & Local Maxima when a senior director at a large bank earns $400,000, receives annual 5–8% raises, and has been told she's "on track" for VP. She's optimising within her current trajectory — taking on higher-profile projects, building internal political capital, refining her management style. Each improvement is genuine. But the ceiling of her current peak — VP at a traditional bank — may be dramatically lower than the ceiling of an alternative trajectory: joining a fintech startup as a C-suite executive, starting her own fund, or moving into venture capital. Those paths require descending first — lower initial compensation, higher uncertainty, loss of status. The valley between her current local maximum and the potential global maximum is what keeps her climbing the same hill.
Product strategy
You're seeing Global & Local Maxima when a SaaS company has spent three years optimising its onboarding flow, reducing churn by 2% per quarter through A/B tests on email sequences, tutorial flows, and pricing page layouts. Each test produces a statistically significant improvement. Each quarter is better than the last. But the overall churn rate has plateaued at 6% monthly — because the fundamental product architecture requires users to perform a complex data migration that no amount of onboarding polish can eliminate. The global maximum requires rebuilding the data import pipeline from scratch — a six-month project that would temporarily break the existing onboarding flow. The team keeps A/B testing because the incremental wins are measurable and the rebuild is terrifying.
Investing
You're seeing Global & Local Maxima when a portfolio manager has refined a quantitative value strategy over eight years, consistently generating 200–300 basis points of alpha in mid-cap equities. The strategy is mature, well-understood, and operates near its theoretical ceiling. Meanwhile, the manager's quantitative skills would transfer directly to systematic strategies in alternative data, credit markets, or machine learning–driven approaches — domains where the alpha potential is multiples higher but requires abandoning the proven approach, accepting a period of negative returns during the learning curve, and building entirely new infrastructure. The 200-basis-point local maximum is comfortable. The potential 800-basis-point global maximum requires crossing a valley of incompetence.
Technology architecture
You're seeing Global & Local Maxima when a company's engineering team has spent four years optimising a monolithic Ruby on Rails application. Response times are excellent, the deployment pipeline is polished, and the team can ship features in days. But the architecture cannot support real-time collaboration, offline mode, or the event-driven processing that the product roadmap demands. Each optimisation within the monolith yields diminishing returns while the ceiling remains fixed. The global maximum — a microservices or event-sourced architecture — requires months of migration, temporary feature freezes, and a period where the system is objectively worse than the polished monolith. The team keeps optimising what exists because the descent feels like regression.
Section 3
How to Use It
Decision filter
"Am I making things better within my current framework, or should I be questioning the framework itself? If I could start from scratch with everything I know now, would I end up where I am — or somewhere fundamentally different?"
As a founder
The most dangerous phase for a startup is the moment after product-market fit, when the company has found a local maximum and the entire organisation begins optimising for it. Revenue is growing. Customers are satisfied. The team is executing. Every metric says "more of this." And in many cases, more of this is exactly right — you've found a genuine global maximum and should exploit it ruthlessly.
But in some cases, the local maximum you've found has a low ceiling. The market segment is smaller than you thought. The unit economics plateau at a level that can't support venture-scale returns. The technology you've built solves the problem well enough that customers won't pay significantly more for incremental improvements. In these situations, the hardest and most important decision a founder can make is to deliberately descend — to pivot, rebuild, or reposition — knowing that the transition will make everything worse before it gets better.
The decision filter is ceiling height. Ask: if we execute perfectly for five years on our current trajectory, what's the maximum outcome? If the answer is a $50 million business and you need a $500 million business, you're on the wrong peak. No amount of optimisation will close that gap. You need to descend and find a higher hill.
As an investor
Local maxima are the hidden risk in portfolio construction. A fund that has found a profitable strategy in one domain — say, growth equity in enterprise SaaS — faces constant pressure to optimise that strategy rather than explore whether a different approach (earlier stage, different sector, different geography) might yield dramatically higher returns.
The diagnostic: when your best analysts spend most of their time on incremental improvements to existing models rather than developing new theses, you're optimising for a local peak. When deal flow comes exclusively from your established network rather than from systematic exploration of adjacent spaces, you've stopped searching for higher ground.
The Gittins index from multi-armed bandit theory provides the formal framework: uncertain options with high potential deserve exploration credits that exceed their expected value, precisely because their ceilings are unknown. A new strategy you haven't tested has an unknown ceiling — it might be lower than your current approach, but it might be dramatically higher. The local maximum trap is treating the known ceiling of your current strategy as the ceiling of all possible strategies.
As a leader
Organisations are local-maximum machines. Every process, incentive structure, and cultural norm is optimised for the current peak. Promotion criteria reward people who improve existing systems. Budgets flow to proven initiatives. Performance reviews measure incremental progress. The entire organisational apparatus is designed to climb the nearest hill more efficiently — and to resist any suggestion that the hill itself is the wrong one.
The leader's role is to maintain dual awareness: operational excellence on the current peak (because abandoning a good position for a speculative one is often foolish) and strategic awareness of whether higher peaks exist. The practical mechanism is what Andy Grove called "strategic inflection points" — moments when the landscape shifts and a previously unreachable global maximum becomes accessible, or when your current local maximum begins eroding beneath you.
The hardest version of this decision is when the current peak is still rising. Abandoning a growing business to pursue a speculative alternative feels irrational — and often is. The discipline is distinguishing between a peak that's still growing toward a high ceiling and a peak that's growing toward a low one. If the ceiling is high, keep climbing. If the ceiling is low, the best time to start descending is while you still have the resources and momentum to survive the valley.
Common misapplication: The most frequent mistake is treating every plateau as a local maximum that should be abandoned. Sometimes diminishing returns simply mean you're approaching the natural ceiling of a genuinely excellent strategy — and the right move is to keep optimising, not to blow everything up. Not every valley leads to a higher peak. Many lead to lower ones. The framework is diagnostic, not prescriptive: it helps you identify when you might be trapped, not what to do about it. The decision to descend should be based on evidence that a higher peak exists, not on restlessness with the current one.
A second failure pattern: romanticising the descent. In startup culture, pivots are celebrated. Disruption is valorised. "Burning the boats" is treated as inherently courageous. But most pivots fail. Most disruption destroys value rather than creating it. The framework says that reaching a global maximum requires crossing valleys — it does not say that crossing valleys is inherently good. A pivot without a credible thesis about where the higher peak lies is not strategic exploration. It's random motion disguised as boldness.
A third failure pattern: confusing a rising peak with a high peak. A trajectory that's currently improving feels like progress toward the global maximum, but the rate of improvement says nothing about the ceiling. A company growing revenue at 50% per year on a peak that tops out at $20 million ARR is still on a low peak, no matter how exciting the growth rate feels. The diagnostic question is ceiling height, not slope. A slow climb toward a $10 billion summit is strategically superior to a rapid climb toward a $50 million plateau, even though the dashboard metrics look worse in the short term.
Section 4
The Mechanism
Section 5
Founders & Leaders in Action
The leaders below didn't just recognise that higher peaks existed beyond their current position. They made the far harder decision to descend — to accept worse outcomes in the short term, endure the valley of uncertainty, and bet that the trajectory toward a new peak would justify the cost. Each case illustrates a different version of the descent: forced by competitive pressure, chosen from a position of strength, or driven by a first-principles conviction that the current peak had a low ceiling regardless of how profitable it appeared.
What connects them isn't a single strategic playbook. It's a shared willingness to distinguish between the peak they were on and the peak they could reach — and to pay the real, immediate, often painful cost of crossing the valley between the two. In each case, the market, the board, or the press called the descent foolish while it was happening. In each case, the altitude of the new peak eventually made the criticism irrelevant.
When Jobs returned to Apple in 1997, the company was stranded on multiple low local maxima simultaneously — over a dozen product lines, each optimised within its own narrow niche, none reaching a meaningful height. Jobs' first act was to destroy all of them. He killed the Newton, the printers, the dozens of Macintosh variants, and collapsed the entire product line into a 2×2 grid: consumer/professional, desktop/laptop. Four products. Everything else was eliminated.
This was not optimisation. It was deliberate descent — a wrecking of the existing landscape to create the conditions for a fundamentally different climb. The iMac (1998) was the first ascent on the new peak, selling 800,000 units in five months and restoring financial viability. But Jobs' true genius for navigating maxima came later.
By 2001, the Macintosh business was a comfortable local maximum — profitable, loyal customer base, strong brand. Jobs launched the iPod, which cannibalised Mac accessory revenue and diverted engineering resources. By 2007, the iPod was itself a local maximum — dominant in portable music, generating billions in revenue through iTunes. Jobs cannibalised it with the iPhone, which subsumed the iPod's function entirely. Each transition required descending from a profitable peak to pursue a higher one. Each descent looked reckless in real time and obvious in retrospect. Jobs didn't just find global maxima. He was willing to destroy his own local maxima to reach them — a willingness that is psychologically rare and strategically invaluable.
Netflix's DVD-by-mail business in 2007 was a textbook local maximum: 7.5 million subscribers, dominant market share, polished logistics, and a recommendation algorithm that kept churn low. Every metric said "optimise this." The team could have spent a decade squeezing incremental improvements from the DVD operation — better warehouse routing, faster delivery, broader disc library.
Hastings saw the ceiling. DVD-by-mail had a maximum addressable market constrained by physical logistics, postal infrastructure, and the declining relevance of physical media. The local peak was profitable but finite. Streaming — uncertain, technically unreliable, with a terrible content library — had a ceiling orders of magnitude higher: anyone with an internet connection, anywhere in the world, watching anything on demand.
The descent was brutal. The Qwikster debacle of 2011, where Hastings attempted to split the businesses, cost 800,000 subscribers in a single quarter and cratered the stock 77%. The valley between the DVD peak and the streaming peak was deep, public, and humiliating. But Hastings had done the calculation: the DVD peak's ceiling was maybe $5 billion in enterprise value. The streaming peak's ceiling — visible only through the fog of bandwidth cost curves and broadband adoption rates — was north of $100 billion. He accepted years of pain to reach a fundamentally higher summit. By 2024, Netflix had over 260 million subscribers and a market capitalisation exceeding $150 billion. The DVD local maximum, had he stayed on it, would have been worth approximately nothing.
Intel's memory chip business in the early 1980s was a local maximum that was actively eroding. Japanese manufacturers — NEC, Hitachi, Toshiba — had achieved superior yields and lower costs through manufacturing discipline and government-subsidised capital investment. Intel's market share in DRAMs fell from over 80% in the 1970s to under 3% by 1985. The company was losing $173 million per year, and every incremental improvement in memory manufacturing merely slowed the descent.
Grove's insight was that Intel was trapped on a shrinking peak while an adjacent, far higher peak — microprocessors — was available but required abandoning the company's founding identity. Intel had invented the DRAM. Memory chips were not just a product line; they were the company's self-concept. Exiting memory felt like corporate suicide.
Grove's famous question to Gordon Moore — "If we got kicked out and the board brought in a new CEO, what would he do?" — was a psychological technique for detaching identity from position. The answer was obvious: a new CEO would exit memory and bet on microprocessors. So Grove and Moore did exactly that, walking out of their own office, walking back in, and acting as though they were the new management team.
The descent was painful. Intel closed memory fabrication plants, laid off thousands of employees, and wrote off hundreds of millions in assets. The valley between the memory peak and the microprocessor peak was wide and expensive. But the microprocessor peak had a ceiling that memory never approached: by the mid-1990s, Intel's x86 processors powered over 80% of the world's personal computers, and the company's market capitalisation exceeded $100 billion. Grove's willingness to abandon a local maximum that defined the company's identity — and to endure the valley of transition — is one of the clearest examples of global-maximum thinking in corporate history.
Bezos built Amazon as a machine for escaping local maxima. The key mechanism was his insistence on optimising for long-term value rather than current profitability — which meant he was willing to accept short-term descents that would terrify any operator focused on the nearest peak.
The online bookstore was a local maximum. By 2000, Amazon dominated book e-commerce. Optimising that peak would have meant deeper book inventory, better recommendation algorithms for books, and publisher partnerships. Instead, Bezos expanded into music, electronics, and eventually everything — descending from a profitable niche to enter categories where Amazon had no advantage, losing money for years as it built the infrastructure for a general marketplace.
AWS was an even more dramatic descent. In the early 2000s, Amazon's retail business was the clear local maximum — growing, profitable enough to reinvest, and well-understood. Diverting engineering resources to build a cloud computing platform for external developers meant descending from the retail peak to explore a landscape nobody could see clearly. For years, AWS generated negligible revenue while consuming significant investment. Wall Street punished the stock. Analysts questioned the strategy.
By 2024, AWS generated over $90 billion in annual revenue and accounted for the majority of Amazon's operating profit. The cloud computing peak was vastly higher than the retail peak Bezos had descended from — and it was only reachable because he was willing to cross a valley of uncertainty that every traditional retailer avoided. Bezos' framework — "disagree and commit," "two-way doors," optimising for regret minimisation over decades rather than quarters — was a systematic methodology for escaping local maxima. He institutionalised the descent.
Section 6
Visual Explanation
Section 7
Connected Models
Global and local maxima sit at the centre of a web of strategic frameworks — some that explain why the trap exists, some that provide the mechanism for escape, and some that describe what happens after the transition. Understanding these connections reveals that the local-maximum problem isn't an isolated analytical curiosity. It's the structural reason why most people and organisations plateau, and why the rare ones that break through tend to share specific strategic behaviours.
The connected models below form two clusters. The first cluster (Explore-exploit Tradeoff, Creative Destruction) describes the forces that enable escape from local maxima — the mechanisms by which search and market pressure drive actors toward higher peaks. The second cluster ([Sunk Cost Fallacy](/mental-models/sunk-cost-fallacy), Compounding) describes the forces that trap actors at local maxima — the psychological and economic gravity that makes descent feel irrational. The third pair (Paradigm Shift, Disruptive Innovation) describes the outcomes that emerge when transitions between peaks succeed or fail at systemic scale:
Reinforces
Explore-exploit Tradeoff
Exploration is the mechanism for discovering whether higher peaks exist beyond your current position. The explore-exploit tradeoff provides the operational framework for the local-global maxima problem: exploitation optimises your current peak, while exploration searches for higher ones. The Gittins index — which assigns an information bonus to uncertain options — is the mathematical formalisation of why descending from a known local maximum to explore uncertain territory can be optimal. Without exploration, you're guaranteed to stay on your current peak. The explore-exploit framework tells you how much exploratory effort is justified given your time horizon and current position.
Reinforces
Creative Destruction
Joseph Schumpeter's creative destruction is what happens to local maxima at the market level. Entrepreneurs and innovators don't just find higher peaks — they destroy the lower ones. The automobile didn't coexist with the horse-drawn carriage. Digital photography didn't coexist with film. Streaming didn't coexist with DVD rental. Creative destruction is the market-level mechanism that forces transitions between local and global maxima, punishing those who cling to lower peaks and rewarding those who find higher ones. The reinforcement is mutual: understanding local maxima explains why creative destruction is painful (descent is required), and creative destruction explains why staying on a local maximum is eventually fatal (someone else will find the higher peak and destroy yours).
Tension
Sunk Cost Fallacy
Sunk costs are the gravitational force that keeps you on a local maximum. Every year invested in your current strategy, every dollar spent building your current infrastructure, every relationship formed around your current approach adds psychological weight to the peak you're standing on. The sunk cost fallacy tells you that abandoning this position "wastes" everything you've built — even though the mathematically correct calculation is entirely forward-looking. The tension is destructive: sunk cost reasoning makes the valley between peaks feel deeper than it actually is, because it adds the phantom cost of "wasted" past investment to the real cost of transition. The antidote is the replacement cost test — what would it cost to rebuild your current position from scratch? If the answer is "not much," the sunk costs are an illusion and the descent is cheaper than it feels.
Section 8
One Key Quote
"When a strategic inflection point sweeps through the industry, the more successful a participant was in the old industry structure, the more threatened it is by change and the more reluctant it is to adapt."
— Andrew Grove, Only the Paranoid Survive (1996)
The deeper insight in Grove's observation: success itself creates the trap. The companies most optimised for the current landscape are the ones least able to navigate to a new one. Their very competence — the precision of their processes, the depth of their expertise, the alignment of their incentives — becomes the force that holds them on a shrinking peak while the global maximum shifts to higher ground elsewhere.
Section 9
Analyst's Take
Faster Than Normal — Editorial View
Global and local maxima is the mental model I reach for most often when diagnosing strategic stagnation — the specific condition where everything is improving, everyone is executing well, and the outcome is still mediocre. It's the model that explains why competence can be a trap and why the most dangerous moment in a company's life is often the moment of peak operational excellence on the wrong trajectory.
The core insight is uncomfortable: optimisation and strategy are sometimes opposites. Optimisation takes the current framework as given and makes the best of it. Strategy asks whether the framework itself is right. The local-maximum model reveals that you can be world-class at optimisation — the best in your industry at refining, iterating, and executing — and still end up in a mediocre position, because you've been climbing the wrong hill with extraordinary skill. The most operationally excellent horse-carriage manufacturer in 1905 was probably doing exceptional work. It didn't matter.
The pattern I see most often in founders is premature convergence. A company finds something that works — a product, a channel, a customer segment — and immediately shifts into optimisation mode. The team runs A/B tests, tightens the funnel, reduces CAC, improves retention. Every metric moves in the right direction. But nobody asks the harder question: is the ceiling on this thing high enough? Because the ceiling question requires stepping back from the dashboard, abandoning the dopamine of incremental wins, and confronting the possibility that the thing you've built and optimised and measured and improved is simply not on the right peak. That's psychologically brutal, and most founders avoid it until the plateau forces the question.
The second pattern is what I'd call valley blindness. People can see their current peak clearly. They can sometimes see the higher peak in the distance. What they can't see — and what they dramatically overestimate — is the depth and width of the valley between the two. The valley is where the descent happens: the period of negative returns, organisational chaos, identity crisis, and external scepticism that accompanies any transition between peaks. Reed Hastings crossed the valley from DVD to streaming and nearly destroyed Netflix in the process. Andy Grove crossed the valley from memory to microprocessors and had to lay off thousands. Steve Jobs crossed multiple valleys and was fired from his own company during one of them. The valley is real, and it's worse than it looks from the peak.
What separates the leaders who successfully navigate the descent from those who don't isn't courage or vision — it's a specific analytical discipline. They estimate three things before committing: (1) the ceiling of the current peak, (2) the ceiling of the target peak, and (3) the cost of the valley between them. If the target ceiling is dramatically higher than the current ceiling and the valley cost is survivable, the descent is rational. If the ceilings are similar or the valley cost is existential, staying put is rational. Most failed pivots result from either overestimating the target ceiling (wishful thinking about the new opportunity) or underestimating the valley cost (naivety about the difficulty of transition).
Section 10
Test Yourself
These scenarios test your ability to distinguish between genuine local-maximum traps — where incremental optimisation cannot reach the best outcome — and situations where the current trajectory is sound and patience is the right strategy. The hardest cases are the ones where the local maximum is profitable, growing, and psychologically comfortable — because those are precisely the conditions that make the trap invisible.
Is this mental model at work here?
Scenario 1
A direct-to-consumer skincare brand has optimised its Shopify store obsessively: conversion rate is 4.2%, email flows generate 35% of revenue, and customer acquisition cost has dropped 40% over two years. Growth has plateaued at $8 million ARR. The founder considers wholesale distribution through Sephora and Ulta, which would require reformulating products for retail shelf life, building a field sales team, and accepting 50% lower margins per unit.
Scenario 2
A machine learning engineer at a FAANG company earns $450,000 annually and has been promoted twice in four years. She's considering leaving to join a Series A startup as VP of Engineering at $180,000 plus 1.5% equity. Her manager warns her: 'You're giving up a guaranteed trajectory for a lottery ticket.'
Scenario 3
A B2B software company serving small accounting firms considers rebuilding its product from scratch to serve enterprise clients. The current product generates $30 million in ARR with 90% gross margins and 110% net revenue retention. The enterprise rebuild would take 18 months, during which the existing product would receive minimal updates.
Section 11
Top Resources
The literature on local and global maxima spans mathematical optimisation, evolutionary biology, organisational theory, and strategic management. The strongest resources combine formal treatment of the landscape metaphor with practical frameworks for recognising and escaping the trap. Start with Kirkpatrick for the algorithmic foundation, read Kauffman for the complexity science, and finish with Grove and Christensen for the corporate reality of navigating between peaks.
The paper that introduced simulated annealing as an optimisation technique, published in Science. It provides the clearest mathematical treatment of why accepting worse outcomes temporarily is necessary to escape local optima in complex landscapes. The metallurgical metaphor — heating to explore, cooling to converge — translates directly into strategic thinking about when to tolerate short-term degradation in pursuit of long-term improvement. Essential reading for anyone who wants to understand the formal basis for "getting worse before getting better."
Kauffman's magnum opus on self-organisation and complexity, including the NK fitness landscape model that formalises how system complexity creates rugged landscapes with many local optima. The chapters on fitness landscapes provide the mathematical framework for understanding why more complex organisations face more local-maximum traps — and why the distance between local and global optima increases with complexity. Dense but transformative for anyone building strategy in complex adaptive systems.
Grove's account of Intel's pivot from memory chips to microprocessors is the definitive case study of a company escaping a local maximum. His concept of "strategic inflection points" — moments when the landscape shifts and the current peak begins to erode — provides a practical diagnostic framework for recognising when you're trapped. The book doesn't use optimisation terminology, but every chapter describes the psychology, politics, and operational mechanics of descending from a known peak and climbing toward an unknown one.
Christensen's theory of disruptive innovation is the local-maximum trap analysed at industry scale. Every disrupted incumbent in the book — disk drive manufacturers, steel minimills, excavator companies — was trapped on a local peak by its own success. The book provides the most detailed empirical evidence for why large, well-managed companies fail to transition between peaks, and why the structural forces that make them excellent at optimisation simultaneously make them incapable of the descent required to reach a higher peak.
Holland's foundational work on genetic algorithms — the first systematic approach to escaping local optima through population-based search with recombination and mutation. The book bridges evolutionary biology and computer science, showing how maintaining diversity in a population of solutions prevents premature convergence on local peaks. Essential for understanding the algorithmic mechanisms that parallel the strategic challenge of navigating rugged landscapes, and for anyone designing organisations or processes that need to avoid the local-maximum trap.
The Local-Global Maxima Landscape — Why the highest achievable outcome often requires getting worse before getting better
Tension
Compounding
Compounding rewards staying on the same peak. Each year you invest in a skill, a relationship, a strategy, or a market, the accumulated base generates increasing returns. Abandoning a compounding position to search for a higher peak means resetting the base to zero and restarting the compounding clock. The tension is real and often favours the local maximum: a modestly high peak with twenty years of compounding frequently outperforms a higher peak with five years of compounding. The resolution requires estimating ceiling heights — if your current peak compounds toward a ceiling of $10 million and the alternative peak compounds toward a ceiling of $1 billion, the reset cost is justified. If the ceilings are similar, compounding on the current peak dominates.
Leads-to
Paradigm Shift
Thomas Kuhn's paradigm shifts are transitions from one local maximum to a higher one at the level of scientific or strategic understanding. Normal science optimises within the current paradigm — the scientific equivalent of climbing a local peak. Anomalies accumulate until the paradigm can no longer accommodate them, triggering a crisis and eventually a shift to a new paradigm that resolves the anomalies and opens a higher peak. The transition is never smooth: it requires abandoning the conceptual framework that produced all your previous successes, passing through a period of confusion and conflict, and rebuilding from a new foundation. Every paradigm shift is a descent from one local maximum through a valley of uncertainty toward a global maximum that was invisible from the old peak.
Leads-to
Disruptive Innovation
Clayton Christensen's disruptive innovation is what happens when a new entrant discovers a global maximum that the incumbent can't reach from its current position. The incumbent is trapped on a local peak — optimised for its best customers, its existing technology, its proven business model. The disruptor starts on a lower peak (worse product, smaller market, lower margins) but one with a higher ceiling. As the disruptor improves, its peak rises until it overtakes the incumbent's. The incumbent can see this happening but cannot respond, because responding would require descending from its own profitable peak — cannibalising its core business, alienating its best customers, and accepting worse performance during the transition. Disruptive innovation is the local-maximum trap viewed from the perspective of the company that fails to escape it.
The model also explains one of the most counterintuitive patterns in technology: why startups beat incumbents. Incumbents are trapped at local maxima by their own success. Their revenue, their customers, their employees, their investors — all of these are optimised for the current peak and resist any descent. A startup, by contrast, starts in the valley. It has no peak to defend, no sunk costs to mourn, no identity to protect. The startup's disadvantage — no resources, no customers, no proven strategy — is simultaneously its advantage: it can search the landscape freely, unencumbered by the gravitational pull of an existing position. This is why Clayton Christensen's disrupted incumbents always saw the threat coming and couldn't respond: they could see the higher peak from their local maximum, but the valley between the two was too deep for a $10 billion organisation to cross.
One nuance that most discussions miss: the landscape itself is dynamic. Peaks rise and fall. New peaks emerge. The global maximum in 2020 may be a local maximum in 2025 and a valley in 2030. This means that even if you successfully reach the global maximum, you can't stay there by standing still. The discipline isn't just "find the highest peak once." It's "continuously reassess whether the peak you're on is still the highest one" — which requires maintaining some exploratory capacity even while exploiting your current position. The companies that do this best — Amazon, NVIDIA, Apple under Jobs — institutionalise the search for new peaks while ruthlessly exploiting the current one. They operate on two timescales simultaneously: short-term exploitation of the known peak and long-term exploration for higher ones.
There's a useful heuristic I've come to rely on: the "clean-slate test." Ask yourself or your team: if we were starting from scratch today, with everything we know now, would we build what we currently have? If the answer is an immediate and emphatic yes, you're probably on a high peak and should keep climbing. If the answer requires rationalisation — "well, we've already built it" or "it would be too expensive to change" or "the team is used to this approach" — you're hearing sunk-cost reasoning, and the honest answer is probably no. The gap between "we would build this again from scratch" and "we're keeping this because changing is hard" is the gap between a global maximum and a local one.
The bottom line: the local-maximum trap is the default outcome of competent execution in a complex landscape. If you're good at what you do, you will find a local peak. If you're great at what you do, you'll reach the top of it. And if you stop there — if you mistake the top of your hill for the top of the range — you'll spend years optimising a position that was never going to be good enough. The model doesn't tell you when to descend or where to go. But it forces the question that incremental thinking never asks: is this the highest peak I can reach, or just the nearest one?
Scenario 4
Nokia's smartphone division in 2008 held 40% global market share with Symbian OS. Internal teams proposed a complete platform rewrite to compete with the iPhone's touch-based interface. Senior leadership chose to optimise Symbian instead, investing in incremental touch features, better app performance, and carrier partnerships. By 2013, Nokia's smartphone market share had fallen below 3%.
