- Monthly spend
- $96k
- Active seats
- 188
- ETV delivered
- 45.7k
- Δ 12mo
- −$0.10
Cost per ETV
How much we spend on AI tools — Claude, Copilot, Cursor — for every unit of engineering throughput value produced.
AI cost trending down — still 17 % above target
We're spending $4.20 per unit of ETV delivered — 17% over the $3.60 target, but down $0.65 from a year ago. The improvement is driven by Cursor (−$1.45) and Copilot (−$0.65) — both compressing cost per ETV as adoption matures. Claude stays the cheapest line item at $2.10. Closing the remaining $0.60 gap means continuing the same trajectory and clearing idle Copilot seats.
- Monthly spend
- $58k
- Active seats
- 142
- ETV delivered
- 10.9k
- Δ 12mo
- −$0.65
- Monthly spend
- $32k
- Active seats
- 82
- ETV delivered
- 7.9k
- Δ 12mo
- −$1.45
What to act on
Biggest cost levers in the next 4 weeks.
142 seats at $5.30 / ETV — down $0.65 over 12 months, but still the highest cost per output. Redistribute 30–40 Copilot seats to Claude or Cursor where Cost / ETV is already converging on target.
Out of 412 active seats, ~58 logged under 4h of AI usage in the last 14 days. Reclaiming them is a 1-week task and drops Cost / ETV by ~$0.30 immediately.
Down $1.45 / ETV in 12 months — the steepest improvement across all tools. Concentrated in Mobile and Data Infra — both formerly bottom-quartile teams that have now closed half the gap.
If 8 points of capacity shift from Maintenance to Key-aligned objectives, Cost / ETV drops to $3.70 with no change in AI spend — the cheapest way to close the gap.
Cost per ETV over time
12 months of AI cost trajectory across all tools, vs the $3.60 target.
By team
Which teams are converting AI spend into ETV efficiently — and which aren't.