Use the same scope
Match the team, repositories, and time period used for Engineering Output and AI spend. Keep incomplete or missing data visible.
AI Impact · part of Developer Intelligence
Show the board what
Show the board what
the AI spend produced.
Measure AI coding ROI by connecting tool spend with code contribution, quality, and delivery outcomes. Larridin brings these signals together so engineering and finance can review the same evidence.
Claude CodeCodexCursorGitHubGitLab
AI Impact
Example · Four complete weeks
AI costs
+6% →$28.4K
billed across four weeks, licences and API usage together
AI Code Share
+9 pts ↑54%
share of shipped lines written with or by an agent
AI quality
+4 ↑92 / 100
AI Quality Score · defect rate 2.1%
Slop Index
−6 ↓14
AI Slop Index on AI-generated code, lower is better
Token waste
−3.1 pts ↓9.1%
of model spend on abandoned turns and cache rebuilds
AI adoption
+12 pts ↑89%
weekly active engineers, 57 of 64
Computed from invoiced spend, merged PRs, CI and session telemetry. Same four-week scope, shown by team.
Trusted by AI-forward enterprises
AI costs and token waste
One AI bill, priced to the team that ran it.
Larridin brings billed AI spend and observed session usage into a team, tool, and model view. Billed charges and session estimates have different cost bases. Cache usage and interrupted work help identify opportunities to investigate.
Priced from the invoice · review billed provider charges and licence costs together; keep list-priced session estimates separately labeled
Token waste, defined · abandoned turns plus caches rebuilt after an idle gap of five minutes or more, as a share of model spend. Every dollar traces to a logged action or to the session's own usage counters
Subscriptions counted · a developer on a personal Claude or ChatGPT plan still shows up, priced at list as estimated API value
Fixed with a default · the Router sets the caching and model tier a new session starts with, per team, so the fix does not depend on a memo
See the Router
AI costs
$28,400
invoiced AI spend, seats and API
$21,480
model and API usage, at list price
9.1%
token waste, as a share of model spend
$9,240
saved by cache reads against uncached input
| Team | Engineers | Model spend | Per engineer | Waste |
|---|---|---|---|---|
| Platform | 18 | $441 | 7.8% | |
| Payments | 16 | $366 | 8.6% | |
| Infra | 14 | $337 | 10.4% | |
| Growth | 16 | $185 | 11.5% |
By tool
Claude Code 58%Cursor 24%Codex 18%
By model
Opus 5 52%Sonnet 4.6 31%GPT-5.6 Codex 17%
Where the waste isSpendShare
Abandoned turnsgenerations the engineer interrupted$1,5487.2%
Cache rebuildsafter idle gaps over five minutes, 62 sessions$4121.9%
Token waste$1,9609.1%
Billed spend uses provider and licence charges. Session cost at provider list rates is estimated API value, with cache creation and reads priced separately. Missing cost data is not a zero-cost session.
AI quality and Slop Index
Catch AI slop before it becomes next quarter's rework.
The AI Quality Score grades AI-assisted code on the same review and CI signals as human code. The AI Slop Index measures low-quality, churn-prone AI-generated code across five dimensions: signal-to-noise, unnecessary abstractions, unreviewed paste, defensive bloat and reinventing the wheel. More AI output stops quietly meaning more cleanup later.
AI quality
92 / 100
AI Quality Score, AI-assisted code
2.1%
defect rate, AI-assisted PRs
14
AI Slop Index, lower is better
6.8% vs 5.9%
30-Day Code Rework Rate, AI vs human
AI Slop Index, weekly
markers show when a routing or coaching change took effect
Jun 15Jul 6Jul 27Aug 17Sep 6
Flagged lines by dimension
ShareUnreviewed paste31%
Defensive bloat24%
Unnecessary abstractions19%
Reinventing the wheel15%
Signal-to-noise11%
Defect rate is bug-fix PRs and reverts traced to AI-assisted changes. 30-Day Code Rework Rate: recently added code rewritten or deleted within 30 days, split AI vs human.
AI Slop and missing-test signals also reduce the Engineering Output a merged change earns, so avoidable quality debt is never counted as full progress.
AI ROI
67.6 pts
Engineering Output / $1K of AI Spend, +31%
4.8×
Estimated net ROI
18h
median PR cycle, −38% vs baseline
Where the 4.8× comes fromFour weeks
Estimated value of engineering time saved$164,400
Invoiced AI spend−$28,400
Estimated net value$136,000
AI ROI4.8×
Engineering Output / $1K, weekly
Output points per $1K spent
Week 1 · 51.6 ptsWeek 4 · 67.6 pts
Estimated net value: estimated engineering capacity value minus AI spend. This is not a cash-savings claim. Velocity is median PR cycle time, AI-assisted work against the team's pre-rollout baseline.
The board answer
Walk in with the number, and the trend behind it.
Engineering Output / $1K and AI ROI fold the six metrics above into two figures a CFO can check: Engineering Output points per $1K of AI spend, and net value over that spend. Use the same scope for Output and spend. Time-savings estimates require a separate baseline and explicit assumptions.
1Connect · Connect GitHub or GitLab and the agents your team runs.Confirm capture coverage and agree on the teams and period to compare.
2Measure · First AI Impact report.Costs, code share, quality, Slop Index, token waste and adoption for every team, against the quarter before.
3Review · Board-ready impact reporting.Compare complete weeks against the same team’s baseline, with assumptions and source data available for review.
This calculation estimates engineering capacity value. Realized financial savings depend on how that capacity is used. Review the example calculation.
A closer look
How AI coding ROI is calculated
Separate delivery efficiency from estimated financial return. Engineering Output per $1K of AI spend relates scored engineering work to spend; estimated net ROI compares an explicit estimate of benefit with that spend.
Separate cost bases
Use billed cost for financial reporting. Treat session costs priced at provider list rates as estimated API value, and keep them separate from invoices.
State the benefit assumptions
An estimate of engineering time saved needs a baseline, an hourly cost, and an attribution method. Estimated capacity value does not establish realized financial savings.
Questions about AI Impact
What is the estimated net ROI formula?
Estimated net ROI = (estimated value of engineering time saved − AI spend) ÷ AI spend. In the illustrative example, ($164,400 − $28,400) ÷ $28,400 = 4.79, rounded to 4.8×. It is an estimate of capacity value.
Does higher AI Code Share mean higher ROI?
No. AI Code Share measures the AI-attributed share of eligible added lines. Read it alongside Engineering Output, quality, reliability, and spend. More AI-attributed code can also introduce rework.
How is this different from company-wide AI Impact?
This page focuses on engineering: coding tools, merged changes, agent sessions, and software delivery. The company-wide AI Impact offering covers the broader organization.
How should I interpret the product examples?
The product views on this page use illustrative data to explain the metrics and workflows. For metric definitions, sample calculations, and assumptions, see our measurement methodology.
Board-ready in 60 days.
Connected in a day, first report in two weeks. Start on your own repos or get pricing.