The Tyranny of Small Decisions

The concept comes from economist Alfred E. Kahn, who described how a series of individually rational, short-term choices can lead to a collectively suboptimal outcome over time.

No single decision is wrong.

The outcome still is.

It doesn't break. It accumulates.

In marketing, failure rarely announces itself as a mistake.

A budget shifts toward higher-ROAS channels. Inefficient reach is reduced. Frequency increases where conversion is visible. Spend follows what can be measured cleanly.

Each move is justified. Often backed by data.

And over time:

Demand creation weakens. Audience breadth narrows. The system becomes increasingly dependent on existing demand. Growth slows, then stalls.

No one decision caused it.

But every decision contributed.

The system learns what you reward

The tyranny is not simply a consequence of media choices. It is embedded in how decisions are structured.

Attribution windows shorten. KPIs prioritise efficiency over contribution. Models are trained on historic allocations. Platform data becomes the dominant signal.

Each choice improves clarity locally.

Collectively, they redefine the system.

Not around growth.

Around what is easiest to measure and defend.

Everyone is right.

The outcome is wrong.

The effect strengthens in organisations

Performance teams optimise conversions. Brand teams defend positioning. Finance pushes efficiency. Agencies optimise to the brief.

Each function does its job.

But no one owns the accumulation of decisions across time.

So the accumulation happens without friction.

And without accountability.

Why it's hard to see

Because at every step:

The metric improves.

The logic holds.

The data supports the move.

You never see the full sentence.

Only the clause:

“This improved CPA by 12%.”

What you don't see is:

“...while reducing future demand.”

The signal is immediate.

The cost is delayed.

Where this shows up in practice

You see it in patterns that feel familiar.

Declining effectiveness despite better efficiency.

Channels that appear weak because they have been systematically underfunded.

Models reinforcing the very allocations they were meant to test.

Digital channels capturing demand that other channels created.

When measurability becomes a narrative

There is a version of this shift that gets explained too easily.

Budgets move from TV to digital.

The rationale sounds obvious.

Digital is measurable.

TV is broad and opaque.

Performance is easier to track.

But this is largely a framing choice.

TV reporting is often delayed and audience-level.

Digital reporting is often immediate and impression-level.

That's different.

TV can be reported with considerable structure: audience by age, gender and region, programme context and placement, reach and frequency curves, and time-of-day dynamics.

Digital appears more granular because that is how platforms choose to report it. Impression-level logs. User identifiers. Real-time dashboards.

An IP address or device ID creates the impression of precision.

But it is still a proxy.

What's really happening

Granularity is not a property of the channel.

It is a property of the measurement system.

What gets logged.

What gets exposed.

What gets surfaced in reporting.

TV can be made to look more granular.

Digital can be made to look more aggregated.

The difference is not capability.

It is convention.

The consequence

Budgets shift.

Not necessarily toward what is better understood.

But toward what is more visible, more immediate and more defensible.

And over time, investment moves from creating demand to capturing existing demand.

These are not interchangeable functions.

Then comes the familiar line:

“There’s nothing good on TV anymore.”

Part of that decline is endogenous.

Spend moves because quality declines.

Quality declines because spend moves.

The loop closes.

The decline is treated as external.

But it is partly self-created.

No single decision caused the outcome.

But the accumulation did.

Modelling is not an alibi

In many organisations, modelling appears once a year.

A model is built. Budgets are justified. Slides are presented. Decisions are signed off.

Then it disappears.

Until the next cycle.

In that role, modelling becomes retrospective.

A way to explain decisions already made.

Or to legitimise them.

What it should be

Modelling is not a checkpoint.

It is a control system.

Not there to validate the plan.

But to interrogate the trajectory while it is still forming.

Because the risk is not a single wrong allocation.

It is the accumulation of right ones.

Where it changes the game

When used properly, modelling does not ask:

“What worked?”

It pressures a different question:

“What are we systematically doing, and where does that lead?”

Are we repeatedly shifting from creation to capture?

Are we reinforcing the same allocation because it is what the model has seen before?

Are we optimising signals that reflect the system rather than the market?

That requires continuity.

Not an annual reset.

The implication

The problem is not a bad decision.

It is a series of good ones, taken in isolation.

Which shifts the role of analysis.

From evaluating individual decisions.

To understanding what repeated decisions do over time.

Because the question is not:

“Is this the right move?”

It is:

“What happens if we keep making this move?”

That is the question the model cannot ask about itself.

That is where human analysis comes in.

Not as a check.

Not as a fallback.

As the bridge between what the system optimises and what the organisation actually needs.

Growth rarely collapses from a single mistake.

It erodes through a sequence of decisions that all made sense at the time.

If modelling only appears once a year, it becomes an alibi.

If it stays present, it becomes a constraint on accumulation.

If you want to change the trajectory, you do not need better optimisation.

You need visibility over accumulation.

That is where the real work starts.