Is your AI investment actually working?
You're paying for Copilot, Cursor, and Claude. DXSignal tells you whether it's paying off — auto-detecting AI-assisted work and measuring it against the rest of your delivery on both speed and code quality.
Auto-detected AI co-authorship
DXSignal finds AI-assisted work from commit trailers and co-authorship metadata — Claude Code, the Copilot agent, Cursor, and aider — plus vendor usage from Anthropic, GitHub Copilot, Cursor, and OpenAI. No spreadsheets, and it works on your existing history.
Speed and quality, side by side
AI-attributed PRs are compared against the rest on cycle time, PR health score, and rework rate — so "faster" never quietly means "worse." With SonarQube connected, you also see whether AI-heavy projects accrue tech debt and duplication faster.
Difference-in-differences
For tagged teams, your DORA change is measured against non-adopting teams over the same window — isolating the AI effect from everything else changing at once. That is the difference between "we think AI helps" and evidence.
Cost vs. impact
Real AI-tool spend (synced from your vendor accounts) sits next to measured delivery lift, so you can answer the ROI question with numbers instead of vibes — and defend or rethink the investment.
A framework, not just a number
Rolling out AI tools isn't one decision, it's three stages. DXSignal scores each one and tells you what to do next.
Adoption
An AI Adoption Score — how much work is AI-assisted, how many teams have adopted, and engagement (acceptance + seat utilization) — with a maturity stage and the friction points blocking wider use.
Productivity
Where the gains are, by team and repo, on speed and quality together — so AI velocity that quietly raises rework or review load gets caught, not celebrated.
Outcomes
ROI in real spend vs measured lift, and per-tool rationalization — which of your AI tools is actually earning its seat, so you can consolidate.
The difference: most tools tell you what share of PRs used AI. That's adoption, not impact. DXSignal uses a difference-in-differences design to estimate what would have happened without AI — so you see whether it's actually working, not just whether it's being used.
Why speed alone is a trap
The seat price of an AI assistant is trivial next to a developer's salary, so any time saved looks like enormous ROI. But time saved at the keyboard isn't the same as faster delivery: AI code can take longer to review, raise rework when it's subtly wrong, and increase duplication and tech debt that slows the team later. The only honest answer comes from measuring AI-assisted code against the rest — on speed and quality together.
Honest by design
We show association, not proof, and label detection as a floor. Every headline is backed by an evidence base you can inspect — PR counts, windows, and confidence. That honesty is exactly what makes it credible in front of leadership, where overclaiming gets found out fast.
FAQ
How does DXSignal detect AI-assisted code?
It reads commit trailers and co-authorship metadata left by tools like Claude Code, the Copilot agent, Cursor, and aider, and combines that with vendor usage data from Anthropic, GitHub Copilot, Cursor, and OpenAI. Detection is treated as a floor, not a precise count.
Does it measure quality or just speed?
Both. AI-attributed work is compared to the rest on cycle time, PR health, and rework, and — with SonarQube connected — on maintainability and duplication trends. Measuring speed without quality is how AI investments look good on paper and disappoint in production.
Is this proof that AI caused the change?
No, and we say so in-product. We show association with an honest evidence base — sample sizes and confidence — and use a difference-in-differences design for tagged teams to isolate the effect as far as observational data allows. It is credible enough to put in front of a VP of Engineering, framed honestly.
Know whether your AI spend is paying off
Auto-detected AI work, measured against the rest on speed and quality, with honest evidence.
Get Started Free