Skip to main content
ROI & value is the layer that turns telemetry into the kind of numbers a CFO will sign off on. Per-agent, per-project, risk-adjusted, with payback and NPV — and confidence ranges instead of single-point estimates.

ROI Hub

All ROI and value work happens in the ROI Hub (Governance › ROI Hub): risk-adjusted ROI, audit trail savings, alternative comparisons and sensitivity analysis on one page. It opens on Executive Summary. The other tabs are Outcomes, Value Outcomes, Financial ROI, Compliance Audit ROI, Alternative Comparison, Sensitivity Analysis, Project Comparison and Productivity Estimator.

ROI by confidence tier

The Net ROI by confidence panel on the Outcomes tab splits every dollar of value by how it was established. Each tier shows its dollar value, its share of the total and the number of traces behind it. Every trace counts once, at the highest tier available for it, so the three tiers never double-count the same run. When the panel is empty, set up agent baselines, run the productivity estimator, and define an outcome contract plus an experiment to fill in the Verified slice.

Calibration from verified outcomes

The Productivity Estimator tab compares the estimator’s numbers against verified outcomes. It pairs runs that have both a productivity estimate and a verified outcome record, and charts Estimator vs verified so you can see calibration drift over time. On each execution, the estimator reports Calibrated hours: the raw hours estimate adjusted by historical calibration for that agent type. Calibrated hours × the role’s rate is the time-saved value attributed to the run, and calibrated hours are the number the ROI Hub sums.

What’s inside

Value Outcome Units

Group agents by business outcome to translate technical metrics into board-level language.

Financial ROI

Per-agent and per-project ROI: monthly savings, payback, annual ROI %, 5-year NPV.

Compliance ROI

Quantify audit-labour savings and risk reduction from governance automation.

Sensitivity analysis

Monte Carlo simulations across ROI inputs — present a range, not a guess.

Why this exists

LLM ops dashboards report cost and latency. They don’t answer the question executives actually ask: is this AI investment worth it? Observatory’s ROI module exists because:
  • Per-call cost is meaningless without a baseline (“what would humans cost?”)
  • A favourable point estimate hides the downside — what if adoption is half what we expected?
  • Compliance work has a real cost that needs to live in the same calculation
  • “We saved time” is rejected by finance unless it ties to a Value Outcome Unit with revenue or cost attached

The ROI calculation

Each agent has a baseline — what manual effort it replaces (hours, rate, volume). Observatory combines the baseline with actual telemetry-derived cost to compute net contribution. Compliance risk modifies the contribution. The numbers roll up through Value Outcome Units to per-project ROI.

Where to start


AI Registry

The inventory layer ROI rolls up against.

Analytics

Cost telemetry the ROI numbers consume.
Last modified on November 9, 2026