account-tier-scoring
Account tier scoring
The insight: a tier is a budget-allocation decision the user will need to re-tune, so the deliverable is a visible weight table, not a number. The most common scoring antipattern is picking the wrong engine: prompting an AI to do weighted arithmetic (pay per row for worse-than-formula math, byte-inexact, untunable), or cramming meaning-level judgment ("is this really our vertical?") into brittle keyword terms. The split that works: deterministic math for everything countable, judgment only where meaning lives — and every judgment call writes down what it saw. The second antipattern is scoring rows the data can't support: a missing dimension silently scored as 0 craters good accounts for being under-enriched. Sparse rows get renormalized over the weights actually observed; rows below the observability gate return "cannot tier," never a confident middle grade.