Optimisation is a usual objective. No CEO claims in the Financial Times: “we plan and commit to reduce our efficiency in the next fiscal year”. However, sometimes optimisation is a trap, and the opposite strategy wins: being deliberately suboptimal.

When optimising is a problem.

In most situations, optimising pays off, but in three situations, it can become a liability.

When optimising makes your moves more predictable

Optimisation regularly depends on shared information, well-known sources or playbooks. This means competitors or algorithms having the same data and logic can easily forecast your next moves. Here the threat is a rival exploiting you. The countermove is paradoxical: deviate from the optimal.

It’s a situation observed in chess, for example. Now that players train with machines that remember all the games played in competition for decades and define optimal moves, they can more easily understand their opponent’s strategy. This is why chess grandmasters are deliberately playing “suboptimal” moves to escape AI-memorized lines.

An advantage a rival can perfectly anticipate and replicate is neither rare nor inimitable, so it has low value. Being unpredictable, by contrast, is a source of advantage. However, unpredictability is not irrationality; it’s a deliberate, costly investment in illegibility.

When optimising leads to commoditisation

Another issue with shared benchmarks and best practices is that they often homogenise offers and practices in a given market. Products look the same, and offers are interchangeable. In this race to the bottom, the only way to differentiate is to reduce the price, which erodes margins for all competing companies. Here, the threat is the market erasing you.

The airline industry is a case in point. Decades of everyone adopting the same best practices (yield management, hub-and-spoke, then low-cost imitation) made carriers more alike. They drove industry profitability to near zero for most of its history.

A whole generation of Direct To Consumers brands handed the same problem — “acquire customers profitably” — to the same optimisation engines (Meta and Google’s ad auctions) and all bid against each other for the same eyeballs. Because they optimised identically, they bid up the price of the input: social CPMs rose 22% and search CPCs 23% year-on-year into late 2021, with Facebook CPMs up 37% in early 2022. Customer-acquisition cost climbed for everyone at once, unit economics broke across the cohort, and “growth-at-all-costs” became “prohibitively expensive”.

When everyone competes using the same best practices for operational effectiveness (the “competitive convergence” described by Porter), firms become more alike, competition turns into mutually destructive attrition, and returns are pushed toward zero.

When optimising makes you sound less valuable

Generative AI democratises optimisation for written text. It helps produce highly optimised content with little or no initial ability. With an adequate prompt, machines can turn a high school student into an experienced content producer, ticking all the boxes for the perfect social media post. As usage of Generative AI for content production rises, what used to signal effort and competence- a well-written text- now signals the opposite: no one was really here.

The recent backlash against “AI slop” (low-quality, mass-produced digital content generated by artificial intelligence with little to no human effort) has turned “Made by Humans” into an actual selling point. Some brands are reintroducing visible human authorship (imperfections, handmade marks, named creators) as a premium signal, precisely because machine-perfect content reads as cheap and generic.

GenAI drives the cost of the polished signal toward zero, so polish loses its separating power; the credible signal becomes whatever is still expensive to fake (evident human effort, verifiable track record, a name that takes accountability). Deliberate “imperfection” is a costly, and therefore honest, signal.

Designing for strategic suboptimality

Being deliberately suboptimal is no silver bullet, quite the contrary. Optimisation means improvement, which is a natural way of outperforming the market. Deliberate suboptimality is a departure from mastery, not a substitute for it. The chess grandmaster knows the engine line cold before choosing to leave it. Deviation without command of the optimum is just underperformance.

So the question is: when is it relevant to be deliberately suboptimal?

Situations when being deliberately suboptimal pays off

Variety, slack and agility are the first victims of optimisation because prioritising short-term gains reduces the resources available for options, resilience and discovery. This means that situations when being suboptimal pays off are the ones where those features matter more than immediate gains.

For example, during Covid, companies that had optimised their supply chain the best (single source, zero inventory, just in time) were the ones that broke, and many of them reintroduced suboptimal practices (buffer inventory, for example) to continue operating.

In innovation, the discovery process consumes resources all along the way with no direct payoff, and a company allocating those resources strictly on expected outcome would fund almost nothing: no test, no experiment, because the business case and the ROI are never proven in advance. Some firms go further and deliberately budget resources for exploratory work, accepting the majority will bring no return. They do it precisely because most bets return nothing, and a rare one returns everything. Optimising that portfolio on expected value would quietly kill the very projects that justify the whole exercise.

This is the distinction between exploration and exploitation. Organisations that over-invest in improving their current model (exploitation) earn faster returns now. Still, as they under-invest in developing new models (exploration), they often fail to adapt when the situation changes. Survival requires deliberately protecting exploration, which is locally suboptimal by construction.

Conditions for suboptimality to pay off

Deliberate suboptimality creates value only in specific circumstances. The first is the presence of an adversary optimising against you: unpredictability is worth something only when someone is trying to read and exploit you, and in a non-adversarial setting being suboptimal simply leaves money on the table. The second is an uncertain or shifting environment, where slack and redundancy pay off precisely because the future differs from the data the optimiser was trained on; in a stable world, the optimum is safe, and any deviation is pure cost. The third is a domain that is becoming “solved” and legible to machines: the more computable and transparent the optimum, the more everyone crowds onto it, and the more the residual advantage sits off it. The fourth is a convex, long-tailed payoff, where a rare hit dwarfs the routine cost of being suboptimal; when payoffs are linear and capped, disciplined optimisation wins.

Absent these conditions, being suboptimal is just underperformance. You must be able to hit the optimum before choosing to leave it. The chess grandmaster deciding to play suboptimal moves is first and foremost … a grandmaster.

Conclusion

None of this is an argument against optimising. Optimisation is still the right default mode almost everywhere, every time. However, now that machines compute the optimum instantly, from the same historical data available to everyone, the optimum has become the baseline. So, the edge migrates to the unstable move the solver never memorised; to the costly, human signal that stays credible because it’s still expensive to fake; to the deliberately different position that refuses to converge on the industry mean.

This is not new; strategy was always about being different, not so much about being better. Porter said it more than thirty years ago. As AI democratises “better” to everyone at once, “different” is critical and deliberate suboptimality is one way to achieve it.

In a world of perfect optimisers, the last real edge is the willingness to be, on purpose, a little bit wrong.

Photo de Filip Bunkenssur Unsplash