Essay
When Optimization Makes Performance Worse
The paradox of improving parts while degrading the whole.
Optimization is the work of making things better. Find the inefficiencies, measure the performance, improve the metrics. The logic is straightforward. The practice is treacherous.
Optimization often makes performance worse. Not despite the optimization, but because of it. The mechanisms are subtle but predictable. Understanding them is essential for anyone serious about results.
Optimizing the Wrong Thing
The most common failure: optimizing a proxy instead of the goal. Marketing metrics drift from reality, but optimization continues against the drifted metrics.
Click-through rate is optimized while conversion suffers. Cost per lead is optimized while lead quality declines. Form completions are optimized while actual inquiries decrease. The metric improves. The outcome does not.
Attribution models lie, and optimization against lying metrics produces systematically wrong decisions. Budget flows to channels that capture credit rather than create value.
Local vs. Global Optima
Optimizing one part of a system can degrade the whole system. This is the local versus global optimization problem.
Funnel optimization illustrates this well. Optimizing lead form conversion might improve that stage while reducing total customers. The form rejects people who would have converted with different handling. The local metric improved. The global outcome worsened.
Or optimizing Google Ads cost per click might reduce total conversions. Lower bids win fewer auctions. Cost per click improves. Total volume drops. Cost per acquisition might actually increase because you are no longer competitive for the high-value searches.
The Short-Term Long-Term Trade-off
Optimization typically operates on short time horizons. A/B tests that run for weeks. Campaign optimization that updates daily. Performance reviews that happen quarterly.
But many valuable activities only pay off over longer horizons. Brand building compounds over years. Mental availability builds gradually. Customer relationships develop over time.
Short-term optimization can cannibalize long-term value. Cutting brand investment improves quarterly numbers while eroding the brand equity that drives future demand capture. The optimization worked; the strategy failed.
The Narrowing Problem
Optimization tends toward narrowing. Find what works, do more of it. Stop what does not work. This sounds sensible. The problem: it reduces exploration.
Narrow targeting optimizes toward proven segments while excluding potential new segments. You optimize toward current best customers while missing the light buyers who drive market share growth.
Narrow creative optimization converges toward proven messages while stopping experimentation. You optimize toward diminishing returns while missing breakthrough approaches.
Narrow channel optimization concentrates budget in proven channels while starving emerging ones. You optimize toward a shrinking opportunity set.
The Competitor Problem
Optimization often assumes a static competitive environment. But competitors also optimize. When everyone optimizes against the same signals, advantages disappear.
Everyone optimizes for the same high-value keywords. Costs rise. Margins compress. The optimization worked for early adopters; late adopters find only crowded, expensive markets.
Everyone optimizes for the same best practices. Differentiation disappears. Distinctive assets converge toward category norms. The optimization produced sameness.
The Complexity Tax
Optimization adds complexity. More segments. More variants. More rules. More exceptions. Each optimization is a new thing to maintain.
Systems fail when complexity exceeds capacity to manage. The optimized system becomes so complex that no one fully understands it. Bugs accumulate. Interactions cause unexpected behaviors. The optimization tax exceeds the optimization benefit.
When Optimization Works
Optimization is not always bad. It works when:
The metric connects to outcomes. When the thing you are optimizing genuinely drives business results, improving it helps. The key is verifying the connection, not assuming it.
The time horizon matches. When optimization operates on an appropriate time scale for the activity being optimized. Brand building cannot be A/B tested weekly.
Exploration is preserved. When optimization includes space for experimentation. Some portion of budget reserved for learning, not just exploiting.
Complexity is managed. When the optimization benefits exceed the complexity costs. Simple optimizations that do not require elaborate systems to maintain.
Global is considered. When local optimization is checked against global outcomes. Does improving this part actually improve the whole?
The Operator Perspective
Systems scale judgment. Optimization encodes judgment into algorithms and processes. When the judgment is flawed, the system produces flawed results at scale.
Good operators are skeptical of optimization. They ask: what is actually being optimized? Over what time horizon? With what trade-offs? Against what competitive environment? At what complexity cost?
Sometimes the best move is to stop optimizing. To accept good enough rather than pursuing perfect. To preserve simplicity rather than adding sophistication. To maintain slack rather than maximizing efficiency.