Back to home
CITABLE INDUSTRY DATA · 2026-05-07

Cloud benchmarks worth quoting

The numbers we use ourselves when we walk into a room. Public sources are linked so you can verify; our own design-partner observations are clearly flagged. All third-party numbers are publicly published; click the source URL to verify. Internal numbers are flagged as such. As we hit ~50 customers we'll add an opt-in anonymised cohort comparison.

Savings unlocked

23% — median savings unlocked by mature FinOps programmes

How much of the cloud bill mature FinOps teams (established 2+ years) typically recover year over year.

If you're at <10% you're early; >25% means you're top-decile.

FinOps Foundation · State of FinOps 2024Source

Tagging hygiene

78% — typical owner-tag coverage on Day-1 of a CARTIE AI engagement
Internal · CARTIE AI

Most teams have decent surface tagging — what's missing is consistent ownership and customer attribution tags.

Aim for 95%+ on owner tag, 80%+ on customer_id within a quarter.

Internal · CARTIE AI design partner data, n=11Source

Commitment posture

47% — average Savings-Plans / RI coverage of steady-state compute

Best-in-class teams sit at 70%+ coverage with a 1-year, no-upfront ladder. Under-commitment is the #1 cause of missed savings we see at audit time.

<30% means double-digit % of bill is recoverable instantly.

Flexera · State of the Cloud 2024Source

Wasted spend

32% — share of cloud spend organisations consider wasted

Self-reported by CIOs/CFOs. The gap between this and what we typically recover (15-25%) is the room to over-promise.

If you don't know — you're not alone. ~60% of teams don't measure this.

Flexera · State of the Cloud 2024 (self-reported)Source

Anomaly response

62% — share of teams that learn about cost spikes from the invoice

Real-time anomaly detection flips this: teams catch a runaway test job in hours instead of weeks.

Same-week alerting is the cheap unlock — Slack-out-of-the-box.

Apptio Cloudability · Cost Trends Q4 2024Source

Per-customer P&L

12% — share of paying customers losing money in a typical SaaS portfolio
Internal · CARTIE AI

Most teams only spot this at the aggregate-margin level. Per-customer breakdown reveals 1-3 high-leverage repricing decisions.

Even one repriced bleeder usually pays for our annual contract.

Internal · CARTIE AI Customer P&L observations across design partnersSource

AI / GPU economics

$0.73 — median monthly AI-inference cost per active user (B2B SaaS)

Big spread by use-case: chat assistants ~$0.30, code assistants ~$2.10, agent platforms ~$8+.

>$3 per user means it's time to add prompt caching + model routing.

FinOps Foundation · AI / GPU practitioner survey 2024Source

Time-to-value

11 days — median time from CARTIE AI signup to first $$$ recovered
Internal · CARTIE AI

Time from connecting a cloud account to the first acted-upon recommendation. Includes the 7-day lookback window.

We aim to hit single digits for everyone by mid-2026.

Internal · CARTIE AI design partner data, n=8Source
Coming next: anonymised cohort lane
The network-effect roadmap, in plain language.

Around customer #50, we'll add a third "lane" to every benchmark: the median across opt-in CARTIE AI customers. It's the only way to answer "how do I compare to teams my size, with my workload mix?" — and it's what we wish existed when we were running cloud bills ourselves.

Privacy: cohort numbers are aggregated, anonymised and require a minimum of 5 contributors per cell. No customer data leaves their tenant. Customers can opt out at any time and their data is removed from the cohort within 30 days.

We value your privacy. Cookies help us improve your experience. Learn more

Install CARTIE AI

Add to your home screen for quick access and offline support