Most teams assume doubling users means doubling Claude costs. It doesn’t.
The first curve is linear: more users, more prompts, same per-prompt cost. A small SaaS team in the EU (case-002) ran customer support auto-replies at scale. Their bill grew from $460 to $1,840 when they quadrupled users—predictable, proportional. The math is simple: 4x prompts, 4x spend.
Then the curves bend. Agentic workflows (case-004) show super-linear growth. A UK agency’s research-draft-critique loop cost $2,490 at 100 users, up from $622 at 25. The compounding effect of multi-step prompts turns linear assumptions into budget surprises. Each iteration calls Claude again, and costs stack faster than user counts.
Even everyday coding assistants (case-006) reveal hidden cliffs. A developer’s daily Cursor-style usage scaled from $36 to $145 when moving from 10 to 40 users. The prompts themselves didn’t change—just volume. But volume alone can push bills past thresholds where savings tools become necessary.
The pattern is clear: repetitive, high-volume prompts (support replies, PR bots) scale linearly. Iterative, multi-step prompts (agents, loops) scale exponentially. Neither is “wrong,” but both need different budgeting.
Audit your own Claude usage. Paste your last 30 days at aiusage.ai—no signup for the number.