Where the numbers come from
Every headline number, its source, and how strong the evidence is. We say what's measured and what's an estimate.
measuredreal datasurveyvendorestimatemodelvendor = a company that sells a related product · model = our own calculation
The problem: is it big?
| ~$1.4T / yr, 11% of revenue | Unplanned downtime cost at the world's 500 largest companies | Siemens, True Cost of Downtime 2024 (181 interviews, extrapolated) | vendor |
| 44% monthly, 14% weekly | Plants with equipment-related interruptions at least that often | ABB / Sapio Research 2025, n = 3,600 | survey |
| 52% | Smaller US manufacturers (< $500M) saying downtime costs ≥ $50k per hour | SimuTech 2025, n = 200 | survey |
| 83% (699 of 843) | Automation assets discontinued or end-of-life in one real plant audit | Siemens obsolescence audit case, 2025 (one plant, not a national rate) | vendor |
| Only 55% | Plants with a proactive plan to manage obsolete components | ABB 2025 | survey |
| 9.2 years | Average age of equipment in US durable-goods factories, oldest since 2011 | Federal Reserve / BEA data via Silverado Policy Accelerator | measured |
| PLCs 10–24 wks, drives 12–26 wks | Lead times for new automation hardware | Distributor planning ranges, Q1 2026 | vendor |
The market: how many customers?
| 25,799 | US manufacturing plants with 100+ employees (8,992 with 250+) | US Census County Business Patterns 2022, NAICS 31-33 | measured |
| 1,667 | US manufacturing acquisitions in 2024 | Travelers / PitchBook | measured |
| ~670 / yr (450–900) | Of those, deals that get an equipment or controls review | Velocity model on the deal count (no body publishes this) | estimate |
| 67.1% | Share of US mid-market PE deals that are add-ons (each brings a new plant register) | PitchBook, Q3 2026 US PE Middle Market Report | measured |
| 32,000 · $3.8T | Unsold PE-backed companies worldwide (the liquidity squeeze) | Bain, Global Private Equity Report 2026 | measured |
| ~$93B | US maintenance, repair and operations (MRO) spend, 2025: the pool savings come from | Mordor Intelligence (market estimate; definitions vary) | estimate |
| 22% | MRO inventory unused for 5+ years | SPARETECH survey of 300 manufacturing executives | survey |
Today's alternatives: what they cost
| ~64 h, ~$12.8k | On-site machinery appraisal, mid-market plant; photo capture saves only 2–5 h | Velocity model from published appraisal fees and practitioner examples | estimate |
| ~104 h, ~$20k | Controls / obsolescence assessment; photo ID + lifecycle lookup saves 9–34 h | Velocity model; anchors: distributor IBE $5–15k, Avanceon $15k fixed fee, Siemens $290–341/h | estimate |
| ~45 engagements / yr | Consulting-firm breakeven for an expert-reviewed assessment tool ($15k fee uplift) | Velocity model ($600k build, $250k/yr to run) | model |
What plants already know about their parts (data collection)
Velocity-powered research, US plants with 100–1,000 employees. No survey measures these directly, so each is a range (low / base / high).
| 60% (45–75%) | Plants using a maintenance system (CMMS) for at least some records | Velocity research, anchored to Plant Engineering maintenance surveys (53% in 2018, 61% in 2019) | survey |
| 15% (5–30%) | Of those, plants keeping part-level records with install dates (the rest track whole machines) | Velocity research estimate; no representative survey exists. Supported by Emerson field visits: only 'a handful' of plants had complete parts lists | estimate |
| 25% (10–45%) | Installed drives that can report their own fan / capacitor life | Velocity research estimate from Rockwell, ABB and Siemens product documentation | estimate |
| 12% (5–25%) | Plants that actually use that built-in diagnostic data for maintenance | Velocity research estimate; MaintainX 2026 survey: 31% do condition-based maintenance (a broader practice) | estimate |
| ~28% of parts | Parts with a usable install record in a typical plant: 60% × (15% × 70% complete + 85% × 60% original part × 70% complete) = 6% part records + 21% machine records | Our formula on the Velocity numbers above. Completeness (70%, from a measured study where record fields went from 20% to 90% filled) and original-part share (60%) are our assumptions | model |
Plant economics (inputs to the price model)
Values for a ~170-person plant (100–249 employee band), scaled by headcount in the demo.
| $1.05M / yr | Maintenance budget | Velocity #1 (benchmarks: maintenance as % of replacement asset value) | estimate |
| $0.37M | Spare-parts inventory on the shelf | Velocity #1 | estimate |
| 12–30% / yr | Carrying cost of spares (cash part only 3.5–8%) | Velocity #2 | estimate |
| ~$20k / yr | Rush / expedite premiums | Velocity #3 (one vendor case + survey anchors) | estimate |
| $300–1,840 / plant / yr | Our cost to serve (image recognition API + human review) | Velocity #18 at claude-sonnet-5-5 prices | estimate |
Our results on the demo
Plants and companies are fictional. Part numbers, lifecycle statuses and marketplace listings are real.
| 127 / 129 | Part numbers read exactly by image recognition (the 2 misses were cut-off labels, flagged, not guessed) | Measured on rendered test photos of real part labels | measured |
| 100% precision · 98% recall | Parts matched to the right catalog entry; 112 / 112 auto-accepted parts correct | Measured against the demo's planted answer key | measured |
| 30 parts | Lifecycle status (active / end-of-life / discontinued) with date and successor | Manufacturer lifecycle pages: Rockwell, Siemens, ABB, Mitsubishi (each part links its source) | real data |
| 42 listings | Real marketplace prices for 6 parts (e.g. refurbished Kinetix 6000 $4.3–6.5k) | Hand-collected Nov 14, 2026 from eBay (sold), Radwell, DO Supply; screenshots in the repo | real data |
| ~$9.7k / plant / yr | Registry subscription: a fixed yearly fee, tiered by the number of critical parts ($4–10k across the demo plants; Toledo ~$9.7k), plus $1–2k one-time onboarding and a monthly fee per sensor point (price to set). Never a share of savings | Our money model (xray/pricing.py sets the tier amounts); every input labeled in xray/params.py | model |
| 176 h → 40 h · ~$84k at stake | One event: Toledo's Line 2 servo drive fails. Worst-case outage without and with the Fort Wayne shelf spare (at stake: $5–249k) | Our model: outage capped at 10–25 operating days, 40–80% of lost output counted (both assumptions) | model |
| ~$77k / yr | Downtime avoided at Toledo, expected (range $12–205k; ~$232k over 3 years; $622k of output at stake). Shown separately, never in the price, never scaled up to the whole plant | Our model: only the spare rescues we found; each part's chance of failing comes from its health score (1 − health/100 per year), one failure per spare within 3 years, discounted at 10% | model |
| $429k cut + ~$1.7M holdback | Tillman deal view: overdue spending the buyer inherits ($79k control parts + $350k for 6 machines at end of life); holdback at a 10% discount rate | Our model; discount rate per Damodaran Jan 2026, escrow norms per SRS Acquiom 2025 | model |
Full research and citations: docs/research/market-validation.md, deep-dive-numbers.md and velocity.md in the repo.