Measure AI ROI
with Confidence

Paste any AI conversation, email, or document. Get an audit-ready report with time saved, dollar value, and defensible confidence intervals.

P10/P50/P90 ranges
Multi-model consensus
Finance-ready exports
12,306 people in controlled trials· 200,000 developers observed in the field· 25 studies calibrated / 85 analyzed for inclusion· Stronger evidence, stronger weighting

ROI Analysis

This text is sent to the AI providers you select and stored on your account, encrypted at rest. Remove anything you would not put in a third-party tool — client names, credentials, personal data — before analysing. Share links can redact it later, but the analysis itself reads what you paste.

Supports full conversations, email threads, and documents. Results include P10/P50/P90 confidence intervals with per-task role-based pricing.

Report

Submit content to generate an ROI analysis.

Calibrated against 25 studies drawn from 85 analyzed for inclusion: 12,306 people in controlled trials and experiments, plus 200,000 developers observed at work over six years. Trial and field evidence are counted separately and never pooled, because they answer different questions. Estimates are weighted by the strength of the evidence behind each task type — a well-evidenced one carries more weight than a thinly-sourced one, and the report says which it is. P10/P50/P90 are confidence intervals. Per-task values use role-based hourly rates.

Task Decomposition

Conversations decomposed into discrete tasks with individual time estimates, complexity ratings, and segment attribution.

Empirical Calibration

Per-task savings rates, role multipliers and bias corrections, drawn from a study registry that records what each source measured. Weighting follows evidence quality, so a well-studied task and a thinly-sourced one are not presented as equally certain.

Per-Task Pricing

Each task mapped to a professional role with BLS-sourced hourly rates. Coding at $105/hr, legal drafting at $130/hr.

Multi-Model Consensus

Multiple LLMs independently estimate time saved. Results aggregated via mean/median for higher confidence.

Audit-Ready Reports

Full methodology documentation, assumption tracking, confidence scoring, and exportable PDF for finance review.

Feedback Loop

User-reported actual times improve calibration over time. Known-bias correction prevents drift from self-report inflation.

Quick Capture

Drag this bookmarklet to your bookmarks bar. Click it on any ChatGPT or Claude conversation to import.


      

Note: Some sites may block bookmarklets due to security policies. A browser extension is planned.

42 Peer-Reviewed Studies
236K Study Participants
147 Calibrated Estimates
9 Bias Corrections