Volume-shift caveat
If total token volume shifted more than 30% between the two windows, the page says so up front and de-emphasizes raw totals in favor of per-unit metrics — a volume swing can make totals misleading on their own.
Measure what changed — not what to credit
Mark when you deployed a new prompt, model, or asset. Costiva shows how your Anthropic API cost and token efficiency changed before versus after — honestly, with caveats, never as proof of causality.
See pricingDeployment markers
Record the moment you shipped something — a new CLAUDE.md, a switched model, a rewritten prompt — with a date, an optional linked asset version, and a scope (org-wide, one workspace, or one API key). Markers are cheap annotations: keep adding them on any plan, and that history becomes more valuable the moment you upgrade.
Before/after windows
Pick a 7, 14, or 28-day window. Costiva compares the period right before your marker to the period right after — excluding the marker day itself — using the exact same usage and cost data your dashboard already shows.
Low-volume badge
A comparison built on very little data is labeled anecdotal rather than presented with false confidence — every comparison shows its underlying sample size.
Chart flags
Every marker appears as a flag on your existing cost and usage charts, scoped correctly — an org-wide marker shows on the org chart; a workspace-scoped marker shows only there.
Frequently asked questions
No. Impact analytics shows what changed in your usage and cost metrics before versus after a marked date — it never claims the deployment is the reason, since other things (traffic, model updates, seasonality) can shift in the same window. It measures what changed, not what to credit for it.
Every comparison checks for a large volume shift between the before and after windows and shows a caveat when one is detected — in that case, per-unit efficiency metrics (like cost per 1M tokens) are more meaningful than raw totals, and the page says so.
You can create deployment markers and see them flagged on your charts on any plan — that history keeps building even before you upgrade. The actual before/after numbers are a Pro/Team feature.