Product shape
Hardware first, then diagnosis, then the work that follows.
SensorsSmart Trac: vibration, continuous ultrasound, temperature, magnetic field (RPM); Energy Trac: current and voltage; 4G gateway, no plant network needed
→
AutoDiagnosisSignals mapped to named failure modes with severity; 70,000-model bearing library; 14-day learning period
→
Janus engineAnomaly → investigation → insight → decision; JANUS model and agent team; asset health score
→
WorkInsight → inspection → failure event; procedures; work orders in Tractian or pushed into SAP, Maximo, UpKeep
Deployment: cloud over cellular, on-prem or edge, air-gapped, and AWS GovCloud (FedRAMP High since Jul 2026). Optional human review of high-impact alerts by certified vibration analysts.
Monitoring: mechanical, electrical, multimodal
Two value lines in the nav, fused into one asset health score.
Condition monitoring
- Smart Trac sensor: triaxial vibration up to 64 kHz, continuous ultrasound for lubrication and early friction, temperature, RPM from the magnetic field. IP69K, hazardous-area rated, 3-5 year battery.
- Names 35 failure modes on the site (claims "75+"): bearing wear, misalignment, unbalance, looseness, gear and belt wear, cavitation, rotor and stator faults.
- Up to 100 sensors per gateway; "operational within 14 days".
| Line | What it does | Claim (vendor) |
|---|---|---|
| Electrical monitoring | Current and voltage per phase, power quality, VFD stress; electrical signature analysis for rotor bars, stator shorts, insulation, inverter faults; locates the fault to supply, drive, cable or motor | "No swap tests. No shutdown." |
| Utilities and process | Energy by sector, asset and line; gas and water as process variables; kWh per unit | −34% energy per product, −17% demand cost (no baseline) |
| Multimodal asset health | Fuses mechanical, electrical, operational (temperatures, load, valves, vacuum), history and documents into a score, e.g. a chiller at 83/100 (mechanical 79, electrical 86, operational 85) | Install in a day, full view in 14 days |
| Downtime prevention | Live floorplans by status, multi-site rollup, executive reports of downtime and money avoided | −43% unplanned downtime, 383% ROI (customer survey they commissioned) |
Failure workflow
Frames from Tractian's public YouTube walkthrough; their site's image server blocked our browser.
Insights
- One centrifugal pump's alerts in a list: misalignment, mechanical looseness, unbalance, cavitation, each with a date and status (pending, in inspection, check).
- The misalignment insight shows the break over threshold (+60.75% on the axial axis), the spectral component, and a linked inspection with 3 of 7 steps done.
- "Need a recommendation? Supervised Analysis" sends it to a human analyst.
Procedures and the technician
- SOPs per asset type: an annual inspection of an NH3 screw compressor with a preparation checklist and OK / not OK steps.
- Copilot drafts a procedure from instructions and the history of similar assets; "Asset GPT" fills technical data sheets from manuals.
- Voice notes are transcribed in work orders and inspections; "Analyses" reads oil and thermography PDFs onto the asset timeline.
CMMS: Tractian has its own (work orders, PMs, inventory, QR codes, offline app), but every English CMMS page now says "No new system" and pushes pre-filled work orders into SAP, Maximo or UpKeep. The old per-user prices ($60 and $100) are gone.
JANUS and the agents
"Physical AI": their own model, sold as an OpenAI-compatible API to a limited group of partners.
How it works Fact
- A lead agent sends a case to mechanical, electrical and operational expert agents, cross-checks them and issues a verdict.
- Claims are checked against "world models": a physical description of what each asset can and cannot do, from a library of digital twins across industries.
- Inputs: raw vibration, current, process signals, images, text, or an asset ID. Output: verdict, component root cause, severity, days to failure, confidence, recommendation, evidence chain.
Claims Fact
- 89.8% on their benchmark of 5 cases; +47% accuracy vs frontier LLMs and +93% vs certified analysts (their numbers).
- NVIDIA case: 50 specialised agents, 500M inferences a day, a 500-GPU cluster.
- Price: $12.50 per 1M input tokens, $35 per 1M output, on AWS Marketplace; US inference only, no data retention.
Use cases
Eleven industry pages, mostly reusing the same few cases.
| Industry | Use cases named | Proof shown |
|---|---|---|
| Food and beverage | Unplanned downtime, audits, CIP and washdown, refrigeration, food-safety SOPs | Ingredion $1M, 168 h; −68.9% downtime |
| Automotive | Downtime, technician backlog, paper checklists, OEM audits | Stellantis welding robots −17% energy per weld; Pirelli |
| Chemical | Line shutdowns, wear, OSHA/EPA/PSM/ATEX compliance | ICL OEE +41%; CP Kelco $447K |
| Oil and gas | Field inspections, shift reports, harsh sites, turnarounds | CP Kelco numbers reused |
| Mills and agriculture | Peak-season uptime, keeping knowledge in the system, lean crews | Ingredion reused |
| Mining and metals | Close the loop on failures, PM efficiency, MSHA safety | Reused |
| Data centres | Cooling faults (chillers, CRAH), backup power readiness, one ranked queue | Logos only |
| Government | Water pumps, defence depots, federal HVAC and backup power | FedRAMP High |
| Heavy equipment | PMs by engine hours, fleet tracking | CZM: downtime 176 → 31 h per rig per month |
| Facilities, manufacturing | Unplanned failures, paper work orders, multi-building visibility | Georgia Aquarium |
Assets covered: 37+ types on the monitoring page, from fans, pumps and compressors to CNCs, presses, extruders, chillers and cooling towers; bottling-line machines and data-centre power on other pages.
Proof and pricing
25 cases with numbers, all vendor-published. Both aggregate studies were commissioned by Tractian.
| Customer | Result (vendor) |
|---|---|
| Unilever (LatAm) | $796K corrective cost protected, 19 failures prevented, 117 h avoided in 112 days |
| Ingredion | $1.0M production savings, $223K maintenance, 168 h avoided |
| Whirlpool | Over $1M avoided; 85% of insights validated |
| ICL | OEE +41%; one of two annual shutdowns removed |
| Bosch | Recurring failures −29% (Bosch's own NLP tool on top) |
| Danone | $600K impact avoided from one alert |
| Johnson Controls | $2.6M savings |
| Inpasa | 1,659% ROI, ~25-day payback |
Verdantix study (commissioned): a hypothetical $200M manufacturer, 5 interviews, 401% 3-year ROI, $5.14M benefits, 8-month break-even. The "36% downtime reduction" appears only on the homepage. Pricing: quote per asset after an on-site check; ~$35 per sensor per month is a third-party estimate.
Analysis
Fact = from Tractian's site or press. View = our assessment.
Go-to-market Fact
- Plant manager persona; four-step sale: discovery, demo, business case, deployment.
- US and Mexico about half of revenue; Brazil origin; Europe from 2026 (Poland).
- G2 Leader in industrial IoT and asset management; no Gartner or Forrester placement.
Moat View
- Real: a fleet of ~200,000 monitored machines and a lab that runs machines to failure; hardware that installs in minutes; money and speed.
- Thin: the JANUS benchmark is tiny and self-run; industry pages reuse the same cases; value depends on buying their sensors.
Weak spots View
- Rotating equipment and maintenance only; no operator, production or quality capture.
- No historian or PLC connectors; customer counts disagree on their own site.
- Retreat from CMMS leaves them dependent on SAP and Maximo for the work itself.
For Minder View
- Take: a dollar figure on every case; a health score split by cause; insight → inspection → event.
- Win on: the whole plant (people, production, quality, energy), no new hardware, Vietnamese-first capture. Consider their sensors as one more data source.
Tractian vs Minder
Where the two products start, and where they meet.
| Tractian | Minder | |
|---|---|---|
| Starting point | Their own sensors on critical rotating assets | First-party event capture from workers and signals; legacy systems as adapters |
| Primary user | Reliability engineers, maintenance technicians, plant managers | Frontline workers and managers on one event log |
| Fault diagnosis | Named failure modes from vibration, ultrasound, current | State and fault diagnosis on the plant model, from whatever data exists |
| AI | JANUS model with expert agents and asset "world models" | Agent over the operational graph; world model trained on decisions and outcomes |
| Frontline | Technician inspections, procedures, voice notes | Operators, production, quality and maintenance events in their own language |
| Value story | Downtime avoided in dollars, per asset | Output, Quality, Energy, Workforce on the customer's baseline |
| Geography | US, Brazil, Mexico, Europe from 2026 | Vietnam proving ground, Europe next |
Sources
- Product screens: frames from Tractian's YouTube videos "Failure Management through Inspections and Events" and "Condition Monitoring 101" (captured 30 Oct 2026). Tractian's image server blocked our browser, so site screenshots are not used.
- Product pages: condition monitoring, multimodal, JANUS models, electrical, sensor specs, pricing, integrations
- Proof: Unilever, Ingredion, ICL, Verdantix study, NVIDIA case
- Company: Series D release, Bloomberg Línea (ARR), AJC, JANUS on AWS Marketplace
Conflicts left visible: customers 2,000+ vs 1,900+ vs 1,500+; failure modes 35 vs "75+"; Series C valuation $720M vs $800M; third founder named differently by AJC. All savings figures are vendor-published.