Most teams assume their production Claude costs will dwarf dev spend. The opposite is usually true.
A small SaaS team in the EU (case-002) cut their customer support auto-reply bill from $1,840 to $287—identical quality, verified by blind A/B tests. Their production workload, running at scale, cost less per seat than the iterative dev work that preceded it. The same pattern appears in every verified case: dev environments bleed tokens through repetition, tweaking, and abandoned branches.
An indie hacker (case-001) building a CRM UI spent $312 in a month, then $74 after switching. The difference wasn’t production efficiency—it was dev churn. A UK agency (case-004) saw a similar ratio: $2,490 down to $498 on agentic workflows, where research-draft-critique loops compound dev waste. Production runs the same prompts once; dev runs them ten times.
The 80/20 rule applies: 20% of your prompts drive 80% of the cost, and most of that 20% lives in dev. Audit your own Claude usage. Paste your last 30 days at aiusage.ai—no signup for the number.