Skip to content
Now booking: 12-week Proof of Value engagements for Q1 2027 →
Industrial AI · Mining & heavy industry

Your data already knows.
We make it answer.

Bringing Predictive AI, Digital twins, Optimisation, AI agents and Real-time analytics to mining and heavy industry.

We turn the operational data you already hold into measurable gains in safety, recovery, reliability and cost, with models that are explainable, benchmarked against the methods they replace, and built for production.

Start a Proof of Value See the results
17+ years industrial experiencePhD-led · control & optimisationOne dataset. One metric.
Processing · Copper recovery
Illustrative
Which variables drive copper recovery from shift to shift?
APPROACH
Learn which process conditions drive recovery, shift by shift, from 18 months of plant history.
Plant historianLab assaysShift logs
Cu recovery, % by shift━ before ━ after
Top drivers
Froth depth
Air rate
Feed grind P80
Outcome in production
+1.0% Cu
−65% recovery variability
Delivered in operations across CopperGoldCoalAluminium & aluminaSmelting & refiningPorts & rail

You already collect the data.
Most of its value is still in it.

We don't sell a platform or ask for new capital plant. We work with the systems you run today and turn what they record into decisions your teams can act on.

01
Process historians
PI, SCADA, DCS tags
02
Lab & assay data
Grade, recovery, block models
03
Maintenance logs
CMMS, work orders, manuals
04
Sensors & vibration
Condition monitoring
05
Geotechnical data
Ground sensors, geomechanics
06
Fleet & dispatch
Trucks, rail, port flow

Four layers deep.
One decision on top.

  1. Decisions in the loop
    Alerts, schedules and dashboards inside the systems your teams already use.
  2. Explainable models
    Machine learning, optimisation and digital twins, benchmarked against what they replace.
  3. Engineered features
    Lead and lag indicators and physics-informed signals, built with your engineers.
  4. Your data
    Historian, lab, maintenance and policy records, cleaned and joined.
Proof of value

We start with one metric.
You choose which one.

Every engagement opens with a data readiness review and a single success measure you already report on — recovery, downtime, throughput. Value is proved or disproved within a quarter.

Data readiness reviewweek 2
Historian coverage · 18 monthsready
Lab assays joined to shiftsready
Sensor gaps > 4 h3 fixed
Success metric agreedCu recovery %
Explainable modelling

We build the model.
You see why it works.

Machine learning, optimisation and control theory, benchmarked against the method it replaces — kriging, fixed alarm thresholds, time-based maintenance. Engineers and metallurgists can check every driver.

Benchmark · ore grade estimation
MethodErrorExplainable
Ordinary kriging (incumbent)[baseline]Yes
Ensemble + spatial features[improvement]Yes · SHAP
Deep network[comparison]Limited
Production & MLOps

We put it into production.
It keeps earning after we leave.

Models are deployed by API into your control, maintenance and business systems, scheduled on pipelines, and watched for drift — with retraining built in so results hold after handover.

Daily pipeline · crusher availability● live
05:00 extract_historian ........ ok
05:04 build_features ........... ok
05:06 score_availability ....... ok
05:07 drift_check (PSI 0.04) ... ok
05:07 push_to_dispatch ......... ok
Forecast within 10% of actual availability · fewer trucks queuing
Decision support

We deliver the insight.
Your teams act on it.

Operator dashboards, executive views and LLM assistants that put model outputs and decades of site knowledge in front of the people making the call — on shift, not next month.

Maintenance knowledge assistant
Why does conveyor CV-03 keep tripping on night shift?
Six of the last eight trips follow belt-tension alarms within 40 minutes. Two work orders and the OEM manual point to the take-up unit. Sources: WO-1182, WO-1240, manual §7.3
>70% less time spent finding information
Continuous improvement

Every shift teaches the model.
Every cycle compounds.

Production is where the learning starts. Each cycle checks what the model got right, finds where the data or the process has shifted, and folds that back in, so accuracy and trust grow the longer it runs.

01
Measure in production
Predictions are scored against actual outcomes every shift.
02
Detect drift and gaps
Sensor faults, missing data and process changes are flagged automatically.
03
Retrain and redeploy
Updated models are validated against the last version before they go live.

Two industries. Deep expertise.
Every service built around them.

We focus on mining, minerals processing and heavy industry. Logistics optimisation and data & AI enablement are delivered as part of that work.

01 · Mining & minerals processing

More recovery from every tonne, consistently.

  • Recovery and grade optimisation
  • Geotechnical hazard and safety prediction
  • Mine planning and extraction
“How much does recovery vary from shift to shift, and do you know which variables drive it?”
02 · Heavy industry & asset reliability

Find faults early. Fix them for good.

  • Fault detection and diagnosis
  • Predictive maintenance
  • Process reliability and root cause
  • Digital twins for plants, smelters and refineries
“Which assets fail without warning today, and what does an unplanned stop cost you per hour?”
Services across both
Mine-to-port & logistics

More throughput from the assets you already have.

  • Pit-to-crusher-to-port flow optimisation
  • Fleet and dispatch scheduling
  • Production and inventory planning
Data & AI enablement

From promising model to lasting capability.

  • 12-week proof of value
  • Production deployment and MLOps
  • Operator and executive dashboards

Not sure where to start?

Tell us the metric that keeps you up at night. We'll tell you, honestly, whether your data can move it.

Book a 30-minute call →

Ask a question.
Get a working view.

A model only matters when people can see it. We build operator screens, management views and ask-your-data assistants that turn predictions into a decision on shift.

See how we deliver
Illustrative views · delivered in Power BI, Spotfire or web apps
Longwall hazard map
PANEL 4 · FACE 0–300 M
6 h
early warning
0 m150 m300 m
Show cavity risk along the face as a heatmap
Rougher circuit,
in focus.
+1.0%↓ 65% variability
Cu recovery vs baseline
Write this shift up like a site report
CRUSHER 2 · AVAILABILITY FORECAST
91%
Next 12 hours
Forecast availability for the next 12 hours
REDUCTION LINE · CELL HEALTH
12 cells to watch
CellFailure riskTrend
C-1140.82↗ rising
C-2070.74↗ rising
C-0310.61→ steady
C-1580.58↘ easing
Flag the cells that need attention this week

Measured in production.
Not in a slide deck.

Every result below started as a narrowly scoped engagement on one dataset and one metric.

Minerals processing · Recovery optimisation
+1%
copper recovery

and a 65% reduction in recovery standard deviation for a copper producer, by isolating the process variables that actually drive performance.

Geotechnical safety · Hazard prediction
74.9%
of cavities predicted 6+ hours ahead

for a multinational coal miner, giving crews hours of notice instead of minutes and cutting downtime by 7.9%.

Asset reliability · Predictive maintenance
US$300k
per cell, per year

Early failure prediction across a fleet of reduction cells, valued on extending each cell's life by 15 days.

Asset reliability · AI knowledge assistant
>70%
faster information retrieval

A domain-tuned AI assistant across maintenance logs, manuals and reports, speeding up root-cause analysis for field engineers.

Mine-to-port · Availability forecasting
<10%
error on crusher availability

Two real-time models deployed at a gold mine, forecasting crusher availability so dispatch can reduce truck queuing.

Mine-to-market · Planning optimisation
20%
productivity increase

in metallurgy and marketing teams, replacing manual spreadsheets with integrated sales, production and inventory planning.

Twelve weeks.
One clear answer.

One dataset, one agreed metric and no new capital plant. At week 12 you decide: scale it, or stop with the evidence in hand.

  1. Weeks 1–2Data readinessAudit what you hold, fix the gaps that matter, agree the one success metric.
  2. Weeks 3–8Model & benchmarkBuild explainable models and test them against the method they would replace.
  3. Weeks 9–11Pilot in operationsRun live alongside your team on real shifts, with a dashboard they actually use.
  4. Week 12DecisionValue measured against the metric. A plan to scale to production — or a clean stop.

Built on the stack you already run.
No new platform to buy.

We connect to the systems your site already trusts, and deliver results back into them.

PI SystemSCADADCSLIMS

Built for the plant floor

We work directly with historian tags, lab results and control-system data, at the resolution your process runs.

SAP PMMaximoPronto

Maintenance history, put to work

Work orders, failure codes and manuals become features for prediction, and answers for your engineers.

SQL ServerPostgreSQLCloud warehouses

Your databases, your permissions

Models read from the databases and warehouses you already govern. Your data stays where it is.

GitModel registry

Versioned and auditable

Code, models and results are versioned, so every prediction can be traced back and reproduced.

AirflowScheduled pipelines

Orchestrated pipelines

Scheduled pipelines extract, score and monitor every day, without anyone running them by hand.

REST APIPower BISpotfire
API

Into the tools you use

Predictions arrive by API in control, maintenance and BI tools, where decisions are already made.

Explainable by design.
Accountable in production.

Benchmarked
Every model is tested against the method it replaces, on your data, before anyone relies on it.
Explainable
Drivers your engineers, metallurgists and regulators can inspect and challenge — no black boxes.
Built on what you have
Your historian, CMMS and warehouse. No new capital plant, no platform lock-in.
Maintained
Drift monitoring and retraining are part of the build, so results hold after handover.

Built by engineers.
Grounded in mathematics.

AnalytixAI is a multidisciplinary team of engineers, mathematicians, statisticians and data scientists. We combine control theory, optimisation and machine learning with more than 17 years of hands-on work inside mines, processing plants, smelters and refineries.

Every engagement brings together the mix of skills your problem needs, and we work alongside your engineers, metallurgists and operators, not around them.

The team behind each engagement
Control & process engineersPlant, sensors, process control
Mathematicians & statisticiansOptimisation, inference, uncertainty
Data scientists & ML engineersPredictive models, LLMs, digital twins
Data & MLOps engineersPipelines, deployment, monitoring
Domain specialistsMetallurgy, geotechnics, reliability
17+
years of industrial delivery
12 wks
to a proven result

Insights

All articles →
Uncategorized · October 24, 2026Hello world!

Questions

Do we need a data platform before we start?

No. We work with the historian, CMMS, databases and spreadsheets you already have. The data readiness review in weeks 1–2 tells you exactly what is usable.

What do you need from us?

Access to one dataset, a person who owns the metric we're moving, and a few hours a week of time from the engineers or analysts who know the process.

What happens after the proof of value?

If the metric moved, we deploy the model into your systems with pipelines, monitoring and retraining. If it didn't, you stop with a clear account of why.

Which industries do you work in?

Mining, minerals processing and heavy industry: mines, concentrators, refineries, smelters and large processing plants. Mine-to-port logistics and data & AI enablement are delivered as part of those engagements.

Bring one dataset.
Leave with an answer.

Tell us the metric that matters. We'll come back within two business days with whether — and how — your data can move it.

Start a Proof of Value Book a 30-minute call