The tax is paid in re-derivation, not in meetings
The intuitive picture of coordination cost is a calendar clogged with meetings. That picture is wrong, or at least shallow. The real cost is paid at the boundary between functions, where a decision crosses from people who hold the originating context to people who do not. Marketing decides a launch date with a hundred implicit assumptions about why; the date arrives in engineering as a number on a ticket. Engineering then has to either reconstruct the reasoning or comply blindly. Both are expensive. Reconstruction burns time and is never complete; blind compliance preserves the date while quietly discarding the conditions under which the date made sense. This is the mechanism: each handoff strips the decision down to its output and forces the receiver to re-derive, or skip, the rationale that justified it.
Information theory gives the phenomenon its shape. A handoff is a lossy channel, and rationale is the most compressible and therefore the most discarded payload. What survives is the conclusion, because conclusions are short. What dies is the conditional structure — "do X, but only while Y holds" — because that structure is long, situational, and inconvenient to carry. A decision that has crossed four functions is no longer a decision; it is a fossil of one, with the living logic stripped out.
Decay compounds, and serial chains compound it fastest
If a single handoff preserves even ninety percent of a decision's fidelity, the arithmetic is unforgiving. Four sequential handoffs leave roughly two-thirds; six leave just over half. The output still moves — the date is still a date — but the reasoning that would let anyone adapt it has largely evaporated. This is why long cross-functional chains do not merely run slow; they run brittle. The decision arrives intact in form and hollow in substance, and the first unexpected event exposes the hollowness because no one in the chain still holds the conditions under which the original choice was correct.
The failure mode this produces is specific and recognizable: frozen decisions that everyone executes and no one can revise. When the rationale is gone, the conclusion calcifies into policy. People defend the date, the spec, the number, long after the world that justified it has changed, not from stubbornness but because the information needed to reopen the question no longer exists anywhere in the chain. The organization mistakes its own forgetting for stability.
The trade-off: centralizing authority cuts decay but creates a worse bottleneck
The obvious remedy is to stop handing off — route every cross-functional decision through one authority who holds full context. This works, briefly, and then fails predictably. A single decision-holder is a serial processor; throughput collapses to the rate at which one node can absorb context, and every function now waits on the same queue. You have traded fidelity loss for latency, and bought a single point of failure besides. Centralization does not abolish the coordination tax. It relocates it from the handoff boundary to the decision-maker's calendar, where it is more visible and frequently larger.
The structural answer is to attack the loss directly rather than route around it. The lever is not who decides, but what travels with the decision and how short the path is.
- Make rationale a first-class artifact. Ship the conditions and the kill-criteria alongside the conclusion, so the receiver inherits the logic, not just the output. A decision that states what would reverse it cannot silently calcify.
- Shorten the chain before widening the mandate. Each removed handoff returns compounding fidelity; collapse a four-function relay into a two-function pairing before reaching for a central authority.
- Co-locate the originating context at the boundary. A standing cross-functional owner who carries the reasoning across one boundary is cheaper than reconstructing it at three.
The decision implication for the operator
The practical instruction is to treat fidelity as the metric and authority as the last resort. Before adding an approver or a central owner, count the handoffs and ask what dies at each one. Most cross-functional decay is solved not by concentrating power but by refusing to let rationale travel separately from its conclusion — and by paying, deliberately and once, the cost of moving context to the boundary instead of paying it again, partially and badly, every time the decision is touched. The operator's job is not to own the decision. It is to ensure that whoever receives it inherits enough of its reasoning to keep it alive.