Adapt

A new offering is planned for months, fully developed and rolled out widely. Only then does it become clear that a central assumption was wrong. Because so much money, time and reputation have already been invested, changing course is difficult.

Tim Harford argues in Adaptthat complex problems are rarely solved by a perfect plan. Learning emerges more through variation, honest feedback and selection. What matters is not celebrating mistakes, but designing experiments so that errors become visible early, are contained and can be used.

1. Variation rather than a premature single solution

In complex systems, cause and effect are not fully predictable. Several distinguishable approaches increase the chance that at least one provides important information or a viable solution.

Variation does not mean arbitrary experimentation. Each approach should rest on a comprehensible assumption. What matters is that alternatives genuinely differ, rather than merely repeating the same idea in slightly different packaging.

2. Selection needs honest feedback

Experiments help only if results have consequences. Criteria are therefore needed before the start: what would support our assumption? What outcome would count against it? When will we change or end the trial?

Organisations hinder this selection when bad news is punished or invested resources are confused with quality. Projects then survive because nobody wants to name the failure. A culture capable of learning separates a person's worth from the assessment of a trial and creates ways to bring in uncomfortable observations early.

3. Keep experiments small and survivable

Not every area permits trial and error. Where people are endangered, rights violated or irreversible harm is possible, high standards of protection apply. In other areas, risks can be reduced through limited scope, duration, separate systems and clear stopping conditions.

This ‘safe to fail’ approach is no invitation to carelessness. It is deliberate risk design: a failure should yield information without damaging the entire system. Scaling follows only after a successful small test.

Your exercise: design a survivable experiment

Choose a change with an uncertain outcome:

1. Minute 1 – alternatives. Formulate two distinctly different paths and the assumption behind each.
2. Minute 2 – selection. Define an observable success signal, a counter-signal and a decision point.
3. Minute 3 – limits. Reduce duration, scope or investment so that a negative result is manageable and reversible.

Document what you want to learn before starting. This makes evaluation less dependent on how attached you later become to your favourite idea.

What we take from this

Adapt strengthens experimentation and constructive handling of uncertainty. Within WEITER7 a project becomes a bounded learning loop with criteria, protection and deliberate evaluation rather than a final bet.

Which important assumption could I test on a small scale before investing more time, money or reputation in a single path?

Context: This insight summarises selected ideas from Harford's practical book in our own words. The exercise and WEITER7 links are our own applications. Failure is not automatically instructive and can cause unequally distributed harm. In safety-critical, medical, legal or ethically sensitive areas, standards, evidence, oversight and protection take priority over open experimentation.

Sources and further reading: Tim Harford's official book page ↗