What would you like to
know about milestone?

Milestone achieves better-than-90% practical accuracy. Attribution works by correlating Git activity timelines with AI vendor telemetry – Copilot, Cursor, Claude, Bedrock, and others. A PR is marked as AI-impacted when 30% or more of its commits show AI involvement. Customers can audit attribution at the commit level directly in the platform.

No. Milestone does not save customer prompts and does not train on customer data. For customers with strict data residency requirements, the on-prem deployment option keeps all raw data – including Jira fields, PR descriptions, and commit details – entirely within your own infrastructure.

This is the right question, and it’s one most tools can’t answer because they only measure adoption quantity, not quality. Milestone helps you get at this by correlating AI usage with outcomes: a developer with high AI adoption but rising post-review change rates and longer review cycles is a different signal than one with high adoption, shorter cycle times, and stable quality. The goal isn’t to score developers – it’s to identify where AI is being used effectively and replicate those patterns.

Milestone is built to produce executive-ready output. Dashboards are configurable for different audiences – developers see their own activity, team leads see their team, VPs see org-wide trends and spend. The “Vibe Metrics” feature lets you build custom metrics in natural language and combine signals (e.g. “show me AI spend vs. features shipped per team”) without writing code or exporting data. Everything is shareable directly from the UI.

Most customers have meaningful data within the first week. For a statistically credible ROI story, we recommend a 30-day window – enough to capture trend lines, compare AI vs. non-AI work across a representative sample of PRs, and give leadership something worth presenting. The first 30 days are typically about establishing a baseline. Days 30–90 are where the narrative gets compelling.

ROI on AI tooling is best measured through effectiveness of spend relative to outcomes – not raw token counts. The metrics we see engineering leaders focus on most:

• AI adoption rate – % of PRs with AI-attributed commits, by team, tool, and time period
• Cycle Time delta – are AI-impacted PRs moving faster end-to-end than non-AI work?
• Review Time trend – is reviewer burden increasing as AI adoption grows?
• Post-review change rate – a proxy for code quality; stable or improving = healthy
• AI spend per merged PR – the clearest cost-per-outcome signal
• Model-level ROI – which tools are delivering the best output per dollar?

Yes. Milestone includes an in-product benchmarking feature that lets you compare your metrics against industry norms – interactively, without needing to export data or write SQL. That said, benchmarking against your own baseline over time is almost always more actionable than cross-org comparisons. The most powerful signals come from watching your own trends: is adoption going up? Is review time coming down? Is spend per merged PR improving?

It depends – and that nuance is exactly what Milestone is built to surface. In aggregate across our customer base, AI-impacted PRs show faster coding time but longer review time. Net cycle time varies by team, tool, and PR size. The most common finding: small AI-generated PRs (under 50 lines) are the biggest drag on reviewer time. The teams getting the most out of AI are typically those who’ve set clear norms around when and how to use it – Milestone helps you identify those patterns so you can spread them.

Yes. Milestone provides per-developer profiles showing AI adoption rate, tool usage, velocity trends, and code quality signals. This is designed for managers and team leads – not to penalize developers, but to identify who might benefit from additional enablement, and to recognize where AI is genuinely accelerating output.

Milestone connects AI tool activity to your actual code. Every PR and commit is analyzed to determine whether it was AI-influenced, and that signal is layered onto your engineering metrics. You can see, for example, that Sonnet 4.5 was responsible for 52% of merged PRs last quarter, that AI-impacted PRs had a 15% shorter coding time but a 20% longer review time, and that post-review change rates stayed flat – meaning quality wasn’t the issue, reviewer bandwidth was.

A few patterns come up consistently across our customer base:

• AI adoption is uneven – a small number of teams or developers are driving most of the AI-impacted output
• Speed improves, but review time often increases – AI-generated PRs tend to take longer to review, especially small ones under 50 lines
• Spend is concentrated – a handful of models or tools account for the majority of token costs
• Code quality holds up – post-review change rates on AI-impacted PRs are typically similar to or lower than non-AI work

For VPs of Engineering, CTOs, and AI transformation leads, Milestone answers the questions the board is starting to ask:

• How much are we spending on AI coding, and what are we getting for it?
• Which teams have adopted AI effectively and which haven’t?
• Is AI coding adoption improving developer throughput and code quality?
• How do we compare to where we were 90 days ago?