> ## Documentation Index
> Fetch the complete documentation index at: https://docs.devperform.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI ROI Report

> Survey-reported time savings fused with AI-authorship telemetry — the exec-facing AI ROI number, computed honestly.

The AI ROI report (Pro, admin-only) answers the question every engineering leader gets asked: **"what are we actually getting from AI coding tools?"** It fuses two sources you already have in DevPerform:

1. **What your engineers report** — three survey questions from the AI impact set: weekly time saved by AI tools, confidence in AI-generated code quality, and the share of work that's substantially agent-written. Responses aggregate at team level with the same anonymity suppression as every other survey result.
2. **What the code shows** — AI-authored share of merged PRs and AI-vs-human rework, from commit-trailer detection on your git history. No agents or daemons on developer machines.

## The math, transparently

* Estimated savings = **median** reported hours saved per week × active contributors × a loaded hourly cost you enter (median, so one enthusiastic answer can't inflate the number; the top "16h+" bucket is deliberately capped at 20h — half a work week — in the math. Surveys sent on the older 5-bucket scale keep their original 10h cap).
* Your hourly-cost and AI-spend inputs live only in your browser — DevPerform never stores compensation data.
* Entering monthly AI spend adds a net-ROI line and cost per AI-assisted PR.
* Telemetry is shown **alongside** the self-reported figure, never multiplied into it — self-reports are perception data, telemetry is a lower bound, and the report says so on the page.

## What published research says

The report includes a collapsible summary of peer-reviewed and published studies on AI coding productivity — results genuinely range from **negative** (a 2025 randomized trial with experienced open-source maintainers found tasks took 19% *longer* with AI) to **strongly positive** (a 2023 controlled experiment measured 55.8% faster completion on a scoped task), with large field experiments landing in between (+26% across \~4,900 developers). Each study is linked with its scope caveats. That spread is exactly why the report leads with what *your* engineers report, corroborated by *your* git telemetry, instead of applying a universal multiplier.

Some tools compute AI ROI by tracking individual developers before and after AI adoption. DevPerform doesn't — adoption and impact are measured for teams only, by design. See the [AI Impact views](/metrics/ai-metrics) for team-level cohort and trend analysis.