Why Improvement Feels Like Getting Worse

You've been driving for fifteen years. You don't think about it — mirrors, lane position, braking distance, all of it automatic. Then you sign up for an IAM RoadSmart advanced driving assessment. Within the first twenty minutes, your observer asks you to commentate aloud on everything you're seeing and doing. Suddenly you're terrible. You miss a signal check you'd never have missed on your commute. You brake too early, then too late. Everything that was fluid has become effortful, and everything effortful has become strained.

This is not a sign that the assessment is failing you. It is a sign that it's working.

What's Actually Happening

Every skill you've stabilised — driving, writing emails, having difficult conversations, estimating project timelines — is a local minimum. The term comes from mathematics: a point where every small move in any direction would make things worse. Local minima are stable. They're resistant to small perturbations. And they're not the same as the global minimum — the best achievable state.

The problem with local minima is that reaching a better one requires climbing out of the current one first. You cannot move from one stable state to a better stable state without passing through a period of instability. Performance drops before it rises. The dip between the two states is real, measurable, and the reason most improvement efforts get abandoned halfway through. This is the Competence Dip — the predictable performance valley between two stable states.

This is not a failure of effort or ability. It is not a sign that the new approach is wrong. It is the mechanism of improvement itself — and not knowing that is what stops most people.

Why the Dip Happens

In the 1970s, researchers at Gordon Training International identified four stages in the development of any competency. You begin in unconscious incompetence — you don't know what you don't know. You move to conscious incompetence when someone names the gap. Then conscious competence as you effortfully apply what you've learned. Finally, unconscious competence: the skill becomes automatic, fast, and invisible to you again.

What's rarely explained is what happens when you try to improve a skill you've already automated. You don't start at stage one. You're at stage four — unconscious competence — and the moment you bring deliberate attention to a pattern you were executing implicitly, you drop to stage two. You become consciously incompetent again. Not because you've forgotten what you knew, but because knowing and doing are handled by different systems, and the conscious system is much slower than the automatic one.

Tim Gallwey documented this precisely in The Inner Game of Tennis (1974). When players were asked to observe their racket swing consciously — just to notice it, not to change it — their performance degraded predictably. Explicit attention interferes with automated skill execution. This isn't a gap in focus. It's a structural feature of how cognition works.

The neurological split is literal. Automated skills run through the basal ganglia — a subcortical system that executes learned sequences at millisecond precision without conscious oversight. Conscious deliberate execution routes through the prefrontal cortex, which is more flexible but operates at substantially slower timescales. When you redirect an automated movement back through deliberate cortical control, you are not adding effort to the same system. You are switching to a different, slower one.

Flow state is the clearest expression of this. Flow depends on a skill being sufficiently automated that the gap between intention and execution disappears. Bringing conscious attention to an automated skill collapses that gap — you're now executing consciously what was happening automatically, and conscious execution is slower, less precise, and more error-prone. The disruption is real and measurable. It is also temporary. The point of the dip is not to stay in it.

The Same Pattern, Three Times Over

Andre Agassi, at age 29, was ranked 141st in the world — too old, many assumed, to compete at the highest level again. He spent the following year rebuilding his physical conditioning and mental approach to the game from the ground up, as he later documented in his autobiography Open. He did not try to refine the game he had. He dismantled it and rebuilt it. He went through a sustained dip before winning the French Open in 1999 at 29, and the Australian Open at 32. The rebuilt game outlasted the original by a decade.

The same pattern appears at the professional scale. Annie Duke, the professional poker player and decision strategist, described in Thinking in Bets (2018) how learning about cognitive biases initially made her slower and less confident at the table — suddenly she was consciously interrogating reads that had previously been automatic. The new knowledge had arrived; the automaticity of applying it had not. For a period, the player who knew less made faster calls.

At the organisational scale, consider Amazon in 2002. Jeff Bezos issued a mandate: every team had to expose its data and functionality through service interfaces. No exceptions. The transition created real friction — productivity dropped, timelines slipped, engineers spent months rebuilding systems rather than shipping features. The outcome was the internal infrastructure that eventually became Amazon Web Services. The dip was not the problem. It was the price of the upgrade.

This is why organisational training programs so often get cancelled at exactly the wrong moment: the performance dip that follows genuine learning looks identical, from the outside, to the performance dip that follows wasted investment. Without a name for the difference, the rational decision is to stop funding what appears to be making things worse.

The Other Side

The dip ends. The new stable state is better than the old one.

Agassi's rebuilt game lasted longer and reached higher. The manager who works through the bias-awareness dip eventually makes faster, less biased decisions — not by ignoring what they've learned, but by integrating it until it becomes the new automatic. Iterative processes applied to skill development work exactly this way: each cycle of destabilisation, conscious adjustment, and re-integration produces a lower minimum than the last. Feedback loops during the dip — from a coach, a reviewer, or a structured practice regimen — are what guide the new pattern toward correctness rather than leaving the learner correcting against nothing.

Flow returns. When it does, it operates at a higher level than before. The driver who passes the IAM RoadSmart assessment doesn't just drive more safely — they see more, respond to more, and do so without additional effort, because the richer pattern has been automated.

Why Most People Stop

Pain-driven change is the reason improvement begins. The current state becomes painful enough — limiting, frustrating, visibly insufficient — that the transition cost starts to look worth it. But pain-driven change is also the reason improvement stalls. The dip produces its own pain signal: the experience of performing worse at something you previously did well. Without knowing that this signal is the mechanism and not a verdict, the rational response is to stop.

This is not a weakness. It is what the signal usually means. Decline typically does indicate a problem. The dip is one of the rare situations where it doesn't. The people who continue through it are not more talented or more resilient by nature. They've usually been told what to expect. They know the dip is a predictable, nameable phase — the Competence Dip — with a known structure and a known exit. Naming it doesn't make it painless. It makes it legible.

When the Dip Isn't Worth It

Deliberate destabilisation is not always warranted. Five situations where the dip costs more than it delivers:

Near-optimal current state. If you're already performing near the ceiling for this skill in your context, the marginal gain from further development may not justify the transition cost. Not every local minimum is worth leaving.

High-stakes timing. The dip requires time to complete, and the timing of when you enter it matters. A musician shouldn't rebuild their technique the week before a major performance. Choose the dip deliberately or not at all.

Wrong level of analysis. The dip pays off when you're destabilising a genuine limitation. If the real problem is environment, tools, or incentives, technical skill improvement won't resolve it, and the dip produces cost without gain.

No feedback available. The dip requires a mechanism to know whether the new pattern is forming correctly. Without feedback, you can destabilise without ever recovering. This is why deliberate practice with a coach outperforms solo practice at equal volume: the feedback closes the loop.

Chronic destabilisation. Continuously disrupting stable states without allowing integration time produces anxiety, inconsistency, and skill fragmentation. The dip needs to end. If you're always entering a new one before completing the last, you accumulate disruption without accumulating improvement.

The Dip Diagnostic

When performance drops during a deliberate learning effort, one question clarifies what's happening: Am I worse because something broke, or because I'm watching myself for the first time?

If you've recently named a pattern that was previously invisible — read about a cognitive bias, learned a technique, identified a habit — and your performance in that area has declined, you are almost certainly in stage two of four. The paralysis, the over-analysis, the sense of regression are not signals that the learning is wrong. They are signals that the learning arrived.

Once the diagnostic confirms it: reduce the stakes of your practice environment, not the volume. Find a feedback mechanism — a coach, a structured review, anything that closes the loop between attempt and correction. And name the dip to anyone tracking your performance. Unexplained decline looks like failure from the outside. Explained decline — "I'm restructuring how I do X, and I'm in the slow phase" — is a milestone.

The trigger situation to watch for: you encounter a named concept — sunk cost fallacy, availability bias, any pattern with a name — and you begin noticing it everywhere, including in decisions you thought were sound. The confidence you had before the name existed was not expertise. It was unconscious incompetence. The discomfort after the name arrives is not failure. It is stage two. Stage four is available on the other side, and the only way there is through.

Back to the Car

The driver who passes the IAM RoadSmart advanced assessment is not the one who never felt uncertain during training. Every candidate goes through the commentating phase — the effortful articulation of what they're seeing, which disrupts the fluency they arrived with. The ones who pass are the ones who understood what the disruption was for.

The dip is not the obstacle between you and improvement. It is the first evidence that improvement has begun.