See the impact before you make the call.
Six FinOps decision types, one simulator. Edit the inputs → live decision cards rank by impact, risk and reversibility. The math is shown — nothing hidden.
If you take all good moves
+$24K/mo
≈ +$288K/yr
Scenarios simulated
4
Top recommendation
Swap `summarise` from gpt-5.2 → claude-haiku-4.5
Saves identified
4/4
Add a scenario
Decision cards · ranked
Swap `summarise` from gpt-5.2 → claude-haiku-4.5
Monthly impact
-$12K/mo
Annual impact
-$148K/yr
Saves $12,369/mo ($148,428/yr). Reversible in seconds — flip the model name and roll back if quality drops.
Side effects
· Latency: ~faster (relative 0.18x vs gpt-5.2 baseline)
· Quality: minor quality regression on long-context tasks
Raise AWS 1y coverage 38.9% → 65.0%
Monthly impact
-$8K/mo
Annual impact
-$99K/yr
Saves ~$99,096/yr in exchange for a $353,916/yr commitment. 28.0% AWS 1Y discount applied to the delta.
Side effects
· Locks in $353,916/yr of additional commitment
· Workload-shift risk: paying for unused capacity if usage drops
Rightsize `data-staging` on cluster `prod-eks` (-35%)
Monthly impact
-$2K/mo
Annual impact
-$29K/yr
Trims 35% off `data-staging` = ~$2,450/mo ($29,400/yr). Watch p95 latency for 48h post-deploy.
Side effects
· Risk of CPU throttling if requests are aggressive
· Reversible — bump requests back up via kubectl apply
Add caching layer to `search` (hit rate 70%)
Monthly impact
-$930/mo
Annual impact
-$11K/yr
Cache deflects 70% of inference calls → saves ~$1,030/mo net of infra ($930/mo).
Side effects
· Cache hits skip the LLM = faster (-200ms typical)
· Stale-data risk for queries that need fresh context
· Cache infra cost ~$100/mo (Redis/Memcached)
Decision Simulator FAQ
Still have questions? Email hello@cartieai.com — replies within 4 business hours.
See full FAQ