Why do top-down plans fail?

Top-down plans don’t fail because planners are stupid. They fail because planning requires simplification — and the things deleted in the simplification turn out to have been doing the actual work.

The forest that died in straight lines

James C. Scott’s Seeing Like a State opens with the cleanest case study of top-down failure ever written. Eighteenth-century Prussian foresters, optimizing for timber revenue, replaced wild old-growth forests with neat monoculture rows — same species, same age, same spacing. Everything illegible was cleared: underbrush, deadwood, “weed” species. The first harvest was a triumph, and the model spread across Europe.

The second generation collapsed. The deleted mess — fungal networks, insect diversity, soil-building litter — had been the forest’s infrastructure. The Germans had to invent a word for what happened: Waldsterben, forest death.

That’s the template. Every top-down plan is a scientific forest: a complex, evolved system replaced by a representation of it that’s optimized for the planner’s visibility — what Scott calls legibility — and the costs land later, downstream, on someone else.

The four ingredients of catastrophe

Scott is careful: simplification alone produces mediocrity, not disaster. The historic catastrophes — Soviet collectivization, Tanzania’s forced villagization, Brasília — required four things at once:

1. Legibility schemes: society reorganized so the center can see it.
2. High-modernist ideology: unbounded confidence that expert design beats evolved practice.
3. Power to impose the plan over objections.
4. A civil society too weak to resist.

The corporate translation is uncomfortable but exact: a metrics regime nobody can question, total confidence in the new architecture, an executive mandate, and teams without standing to push back. Most failed reorgs and great rewrites have all four ingredients in miniature.

What the plan deletes: mētis

Scott’s name for the deleted material is mētis — practical, local, adaptive knowledge that can’t be written into the spec. The farmer’s feel for this valley’s frost. The ops engineer’s knowledge of which “redundant” service is actually holding up production. The support agent’s sense of which complaints predict churn.

Two of Scott’s observations should be tattooed on every planning document:

• Formal order is parasitic on informal order. Work-to-rule strikes prove it: when workers follow the official procedures exactly, production halts. The org chart never described how work happened; it described how the center imagined it.
• The plan’s tidiness is aesthetic, not functional. Brasília looked perfect from the air and was unlivable at street level — it was rescued by the unplanned settlements that grew around it. Beware any plan whose chief virtue is how clean the diagram looks.

There’s a Wisdom of Crowds corollary here: Surowiecki’s case for decentralization is the same argument in reverse. Knowledge in complex systems is distributed by nature; the question is whether your decision process aggregates it or overwrites it. Top-down planning is the overwrite. The center doesn’t just lack the local knowledge — the act of imposing the plan destroys the conditions under which the knowledge was produced.

How to plan anyway

Scott isn’t anti-planning, and “bottom-up everything” is not the lesson. His closing rules are a how-to for planning in complex systems, and they transfer directly to product, infrastructure, and org design:

1. Take small steps. Prefer ten reversible moves to one grand stroke; let feedback steer.
2. Favor reversibility. Ask of every change: how do we undo this when we’re wrong?
3. Plan on surprises. Leave slack — budget, schedule, architectural headroom — for what the plan can’t foresee.
4. Plan on human inventiveness. Users and workers will adapt, repurpose, and subvert the design. Build channels that turn that adaptation into signal (pilot programs, escape hatches, feedback loops) instead of suppressing it as non-compliance.

And one addition from the audit side: for every dashboard, model, or metric you impose, keep asking “what does this representation delete, and who relied on it?” People optimize against the representation — gamed metrics are legibility’s revenge.

The forest is the test. If your plan’s success depends on the system staying as simple as the slide that describes it, you have planted rows. Wait for the second generation.