Reliability intelligence for process industries

Find the real root causes of failures. Not just the symptoms.

Subatix connects plant data across systems, logs, and engineering knowledge to find why failures happen, what to fix next - and the production capacity hidden behind them.

Built for mills, mines, and refineries running on legacy systems, Excel, and tribal knowledge.

See in actionTalk to the team
Rank 01 · open findingReliabilityEngine

EAF-2 Hydraulic Power Unit

P-204B · EAF-2 · area MS2-EAF

$1.24M
Loss / 12mo
7
Events
78%
Confidence
Primary · root cause62%

Discharge-side check-valve degradation on P-204B drives pressure transients on each pump start - consistent across 7 work orders and historian startup spikes.

CMMS7 work orders · 5 seal replacements
PI historian14 startup pressure transients
Shift logs“P-204B knocking on start”
Recovery $0.31M / yrApprove

Subatix OS · ReliabilityEngine - live product, example workspace

From our team’s prior work · metals, mining & energy

$0M
Additional EBITDAgold plant ramp-up to design capacity
0%
Downtime reductioncopper plant · RCA & maintenance planning
$0M
First-quarter impactops control tower, US energy company
$0M
EBITDA potential foundOpEx diagnostic, gold producer
0%
Downtime reductioncopper smelter · turnaround acceleration
$0B
Modernization programcapital strategy, copper producer
01The problem

The losses are known. The reasons never are.

Recurring failures, lost throughput, yield give-away, a swelling maintenance backlog - every plant can name them. Why they persist is a harder question, and it stays unanswered for the same five reasons.

Most operations are sitting on production capacity hidden in data they already have.

  1. 01

    Fragmented systems. Historian, CMMS, MES, LIMS, shift logs, email - no single source of truth for cross-functional analysis.

  2. 02

    Manual data stitching. Analysts spend most of their time finding and preparing data, not analyzing it.

  3. 03

    Shallow analysis. Reporting stops at “what happened.” It rarely reaches why - or what to do about it.

  4. 04

    Nothing stays. Findings go stale as operations evolve, because no system stays behind to keep them current.

  5. 05

    Tribal knowledge. The context that explains the data lives only in the heads of individual employees.

02How it works

One loop, run continuously.

Not a one-off diagnostic. Subatix runs the same loop your best engineers would - across all of the data, all of the time.

01

Connect

Connects the data you already have - historian, CMMS, MES, LIMS, shift logs, even email chains - and enriches it with operational context.

HistorianCMMSMESLIMSShift logsEmail→ one contextualized layer
02

Find

Searches across every source for losses and uplift opportunities, and translates each one into recoverable dollars.

01EAF-2 hydraulic - recurring trips$1.24M
02Caster speed-limited on grade B$0.86M
03

Recommend

Domain modules turn findings into prioritized actions - every claim linked to the records behind it.

Primary · root cause62%

Replace discharge check valve & seal kit; move PM to condition-based on oil-Fe trend.

04

Act

Your team decides and executes. The system tracks outcomes, updates as new data lands, and keeps learning your plant.

ApproveAssignDismissyour team decides
03The output

Not a dashboard. A finding you can audit.

Rank 01 · open finding$1.24M loss / 12mo · 7 events · 38.5h downtime

EAF-2 Hydraulic Power Unit

EAF-2 › hydraulic power unit › discharge pump P-204B › area MS2-EAF

Primary · root cause62%

Discharge-side check-valve degradation on P-204B drives pressure transients on each pump start, leading to repeated mechanical-seal failure. Consistent across 7 work orders and historian startup spikes.

Evidence chain

4 sources · 17 records
CMMS7 work orders in 12 mo - 5 mechanical-seal replacements on P-204B
PI historianDischarge-pressure transients at startup - 14 flagged events
Shift logs3 operator notes: “P-204B knocking on start, vibration”
LIMS · oilFe wear particles rising 320 → 610 ppm across 3 samples

Replace discharge check valve & seal kit · recovery $0.31M / yr

ApproveAssignDismiss
Ranked hypotheses

Every finding carries competing explanations with explicit confidence - never a single unexplained verdict.

Evidence chain

Each claim is traced to source records and timestamps. Auditable by any engineer, not a black box.

Action & recovery

Findings end in a corrective action with an owner and a recovery value - ready to approve, assign, or dismiss.

04The platform

Starts with reliability. Runs the whole operation.

Eight modules on one data foundation. Land with the reliability wedge, then expand across maintenance and production as the value compounds - every module powered by the same contextualized data layer.

Foundational
Reliability
Maintenance
Production
Reliability Engine

Bad Actors

Ranked by failure frequency × downtime × cost impact - evidence-linked to source records

38 assets · 12 mo
Ranked$ impact
01
EAF-2 Hydraulic Power Unit
P-204B · EAF-2
$1.24M
78%
02
CCM-1 Mould Oscillator
OSC-1 · CCM-1
$0.74M
72%
03
Baghouse ID Fan
FN-3 · DEDUST
$0.51M
69%
04
LF-1 Electrode Regulator
REG-1 · LF-1
$0.39M
61%
05
Water Treatment Pump
WTP-2 · UTIL
$0.22M
58%
Rank 01 · open finding

EAF-2 Hydraulic Power Unit

EAF-2 › hydraulic power unit › discharge pump P-204B › area MS2-EAF
$1.24M
Loss / 12mo
7
Events
38.5h
Downtime
78%
Confidence

Ranked hypotheses

2 competing
Primary · root cause62%

Discharge-side check-valve degradation on P-204B drives pressure transients on each pump start, leading to repeated mechanical-seal failure. Consistent across 7 work orders and historian startup spikes.

Competing23%

Cooling-water fouling raising oil temperature, causing viscosity loss and cavitation. Supported by oil-temp drift but does not explain start-correlated pressure spikes.

Evidence chain

4 sources · 17 records
CMMS7 work orders in 12 mo - 5 mechanical-seal replacements on P-204BWO-48817 · 11 May
PI historianDischarge-pressure transients at startup - 14 flagged eventsPT-2041 · 14 ts
Shift logs3 operator notes: “P-204B knocking on start, vibration”Shift B/C · Apr-May
LIMS · oilFe wear particles rising 320 → 610 ppm across 3 samples3 samples · Mar-May

Recurrence

7 events · 12 mo
Mean time between failures52 d
Trend↓ from 78 d
Next due (est.)~2 Jul

Recommended action

OwnerA. Petrov · Reliability Eng.
ActionReplace discharge check valve & seal kit; move P-204B PM to condition-based on oil-Fe trend.
Recovery$0.31M / yr · MTBF +40%
ApproveAssignDismiss

ReliabilityEngine

Finds the root cause behind every recurring failure and drafts the corrective action - evidence attached.

Top bad actor
$1.24M
Root-cause confidence
78%
Recovery potential
$0.31M/yr
05Principles

Built for plants, not demos.

01

Safety is the first constraint

Every recommendation stays inside hard safety and equipment limits. Constraints are inviolable - the system never proposes across them.

02

AI recommends. Humans decide.

Targets and constraints are set by your team. Engineers approve every action before anything changes - Subatix amplifies their judgment, it doesn't replace it.

03

Evidence-linked & auditable

Every finding traces to records and timestamps in a plain evidence chain. Any engineer can audit any claim - never a black box.

04

Aligned to your operating model

Configured to your assets, standards, and workflows - and built for brownfield reality: messy exports, inconsistent coding, tribal context.

06Deployment

First findings in 1-2 weeks. On raw exports.

No data lake. No warehouse. No 12-month integration program. We land on the data you already export, and the system earns its place from the first findings.

Stage 1weeks 1-2

Diagnostic

Two tracks: we dive into your operations and pain points, and assess data and systems readiness. Works on raw exports - CMMS, historian, downtime records. No IT project.

Stage 2first wave

Configure & train

The priority modules, configured to your processes - first ranked findings on your data, not a generic demo. We train your team and fold in feedback until daily use sticks.

Stage 3ongoing

Support & improve

We monitor quality and stability, recalibrate as operations evolve, and roll out improvements with your team. The system stays - and keeps learning your plant.

Works on raw exports · No IT project to start · On-prem or private cloud · Your team operates it

07Our team

Built from real operations, by people who’ve seen ops problems firsthand.

We’ve spent years inside mills, mines, and refineries - running operational excellence programs and building applied AI systems. We’ve watched the answers sit in data nobody could connect. Subatix is that connection, built to work the way operations, reliability, and maintenance teams actually work.

SteelCopperGoldAluminumRefiningPetrochemicals
Alexander Berks
Alexander Berks
CEO, ex-McKinsey

Operational excellence across process industries.

alex@subatix.comLinkedIn ↗ Alexander Berks
Vlad Rozhkov
Vlad Rozhkov
CTO, ex-SciveFlow

AI products and applied systems, full-stack development.

vlad@subatix.comLinkedIn ↗ Vlad Rozhkov
08FAQ

Questions we hear most often.

01How is this different from existing tools?

Existing tools analyze individual systems (CMMS, historian, BI dashboards).

Subatix connects data across systems, logs, and engineering inputs to identify true root causes - not just symptoms.

02Do we need to integrate all our systems first?

No. Subatix can start with raw data exports (CSV, reports, logs) and deliver initial results within 2 weeks.

Integration comes later as the system becomes continuous.

03What data does Subatix work with?

Subatix works with a combination of:

  • historian data (process signals)
  • CMMS data (failures, maintenance)
  • production data (MES)
  • logs, reports, and engineering notes

We adapt to whatever data is available.

04How do you integrate with our systems?

Subatix initially works with exported data from your existing systems.

As the system becomes continuous, we integrate directly into your environment.

05Is our data safe?

Yes. Subatix does not require uploading raw plant data to external systems.

We can operate using local exports initially and deploy on-premise for continuous use.

06Can it run fully on-premise?

Yes. Subatix deploys on-premise or in your private cloud, inside your security perimeter.

Your data stays in your environment; your team operates the system.

07Can recommendations violate safety or equipment limits?

No. Hard safety and technical limits are inviolable constraints - the system never recommends across them.

Targets and constraints are set by your team, and engineers approve every action before anything changes.

08What does the output look like?

A prioritized list of failure root causes with specific corrective actions.

Subatix shows which issues drive the most downtime and what to fix next.

After deployment, it also supports tracking and verifying implemented actions.

09How long does it take to see value?

Initial insights are delivered within 2 weeks.

Value increases over time as the system becomes continuous and more data is incorporated.

10Is this replacing our reliability team?

No. Subatix augments reliability teams by handling data analysis and identifying root causes.

Engineers remain responsible for decisions, execution, and field operations.

11How accurate are the results?

Subatix combines data from multiple sources and cross-validates patterns to identify consistent root causes.

Accuracy improves over time as more data is processed.

12Will this work with messy CMMS and brownfield data?

Yes. Subatix is designed for brownfield environments with inconsistent asset naming, incomplete failure coding, and fragmented data.

It does not require a clean data warehouse to deliver value.

13How do you handle unstructured data like logs and reports?

Subatix uses AI models to extract and connect information from logs, reports, and other text sources, combining it with structured data to provide full operational context.

14What industries do you focus on?

Process industries such as metals, mining, chemicals, refining, and power generation - where operations are complex and data is fragmented.

15Why hasn't this been solved before?

Because operational data is fragmented across systems and unstructured sources, and traditional tools cannot combine them effectively and surface actionable insights.

Recent advances in AI now make this possible.

Which failure on your reliability list has been there for two years? And what would it be worth to actually close it?

See in actionTalk to the team
What happens next

A 30-minute call. Two questions. No deck.

See in action