Methodology

Every formula, in the open

A calculator you can't audit is a marketing widget. These are the formulas, assumptions and sources.

Formulas

What does unplanned downtime cost you?. Annual cost = hours/month × 12 × revenue/hour × margin%. Recovery scenarios apply the 20–50% range condition-based programs typically deliver.

What is one OEE point worth on your line?. Extra good units = scheduled hours × ideal rate × (OEE gain ÷ 100). Annual gain = extra units × margin/unit. Assumes the market absorbs the extra output; if not, value shifts to cost-per-unit reduction.

What would cutting scrap actually save?. Annual scrap cost = units × scrap rate × cost/unit. Scenarios apply 30% (typical ML root-cause result) and 58% (best vendor-reported published deployment).

What's on the table in energy?. Annual spend = MWh × €/MWh. Savings = spend × reduction%. The reduction slider spans the published range for ML setpoint and sequence optimization.

Assumptions

All money figures use contribution margin, not revenue, unless a field says otherwise. Downtime recovery scenarios (20/35/50%) reflect the practitioner-reported range for condition-based programs. Scrap scenarios use 30% (typical ML root-cause result) and 58% (best published vendor-reported deployment, JEMBA — labeled as such). Energy scenarios span program-style results (5–15%) up to −20% (vendor-reported). OEE math assumes the market absorbs extra output; where it doesn't, the gain converts to cost-per-unit reduction.

Benchmark sources

Downtime economics: Siemens, The True Cost of Downtime. Market sizing: IoT Analytics industrial AI coverage. Public cases: Renault (predictive maintenance, ≈€270M/yr). Vendor-reported results: JEMBA published deployment figures, always labeled. OEE reference points (85% world-class / ~60% typical) are long-standing industry benchmarks.

Changelog

2026-09-01 — Initial publication: 4 calculators, 6 playbooks, benchmark tables, ROI workbook.

Ownership & independence disclosure

FactoryOptimizationAI.com is published by TEEPTRAK SAS (Paris, France), maker of the TeepTrak performance-tracking system (450+ factories) and operator of the industrial-ML vendor JEMBA. Where either appears on this site it is as a labeled example; formulas and benchmarks carry their sources so you can check the math without trusting us. Rubric-scored comparisons of industrial-AI platforms — including JEMBA, scored under published guardrails — live at IndustrialProcessAI.com, where the same ownership is disclosed. No vendor pays for placement here.