Field Guide
Your token spend is up. Do you know what it's producing?
Three outcome signals tell you whether AI is earning its cost. Eleven guardrail metrics tell you whether the program is being run well. Learn what to track, what each signal tells you, and what to do next.
This is what unmanaged token spend looks like
Faros runs diagnostics on AI-enabled SDLCs. These patterns show up consistently once the right signals are connected:
Misaligned tokens
80% of token spend falls outside top business initiatives
Wasteful tokens
30% of token spend produces no useful outcome
Mispriced tokens
Most teams are paying 3x for work a cheaper model handles just as well
The benefits
What this guide gives you
Three decisions you will be able to make with data, not gut feel.
- Where to redirect token spend toward work that matters
Find it, quantify it, and shift it toward initiatives that move the business - Where to stop paying for sessions that produce nothing
Wasteful token spend concentrates in specific repos, teams, and types of work. This guide shows you where. - Which tools and models to keep, scope, or cut
Normalized cost-per-output metrics make vendor and model decisions defensible.
The full picture
14 metrics across four categories
Based on two years of telemetry from 22,000 developers across 4,000 teams.
Outcomes
3 Metrics
Connect token spend to shipped work, strategic priorities, and defensible ROI.
- Productive vs. wasteful token spend
- Token spend by tool, normalized to output
- Alignment of spend to strategic work
Adoption
4 Metrics
Know which tools are earning their licenses and which have become shelfware.
- License utilization rate
- Usage depth distribution
- Tool preference relative to licenses
- Code acceptance rate by tool
Productivity
3 Metrics
Find where AI is accelerating output and where it is creating new bottlenecks.
- PR merge rate per developer by tool
- Cycle time by stage
- Lead time from commit to production
quality
4 Metrics
Instrument the signals that surveys miss before problems compound in production.
- Bugs per developer, trended
- Incidents per PR, trended
- PRs merged without any review
- AI risk footprint by repo
Read the guide
Get the Field Guide to Measuring Token Efficiency
14 metrics to track, what they tell you, and how to use them