Your plant already produces the data. We turn it into findings, fixes, and proof.
In-Sight watches every signal, finds what matters, investigates the cause with AI grounded in your own documents, dispatches the fix, and proves it worked. While the shift is still on.
An industrial analytics platform built for mining, mineral processing, manufacturing, and pharmaceutical operations. Made by ExciteData in South Africa.
Recovery
81.4%
-3.2σ
Dose rate
3.52
-12%
Confidence
94%
high
AI copilot: Reagent underdosing is the likely driver (Pump Manual p.3)
Work order WO-0142 dispatched to maintenance · evidence attached
Delivered for teams at
ModikwaRichards Bay MineralsThe problem
Most plants are data-rich and decision-poor.
Historians capture everything. Dashboards show everything. And yet the same failures recur, because nothing connects a signal to its cause, the cause to a fix, or the fix to proof. Nothing gets smarter.
Alarm fatigue
Thousands of threshold alerts with no prioritisation and no context. The signal that matters drowns in the noise.
Black-box AI
Models that fire predictions no engineer trusts. No explanation, no provenance, no way to verify.
Knowledge that walks out the door
Root-cause expertise lives in a few heads and a filing cabinet of PDFs. Nothing captures it, and nothing learns from it.
The intelligence loop
Detect. Investigate. Act. Learn. Repeat.
One system runs the full loop from detection to a closed-out fix, and every pass through the loop makes the next one sharper.
Seal temperature drifting toward limit
Model confidence 94% · drivers ranked by SHAP
Detect
Real ML, SPC, and engineering limits surface the anomalies, breaches, and predictive risks that matter. Every finding carries its confidence, severity, and the drivers behind it.
- Trained models, live inference
- Overshoots caught at up to 2 kHz
- ISA-18.2 alarm analytics
Worn sealing element
evidence 87%Film supplier batch change
evidence 41%Ambient temperature shift
evidence 12%Replacement interval exceededSealing Unit Manual · p.12
Investigate
An AI copilot assembles the case: related findings, causes ranked by evidence, and a draft root-cause narrative. Grounded in your historian and your own documents, cited to the page.
- Roll findings into one case
- Hypotheses ranked by evidence
- Cites your SOPs and datasheets
Replace sealing element, Station 2
Act
A confirmed cause becomes a dispatched fix. Work orders and prescriptions reach the right people by email with the evidence attached, and write back to your maintenance system when you want them to.
- Work orders with evidence
- Email dispatch built in
- Optional SAP write-back
Live precision
92%Learn
Every operator verdict becomes ground truth. Models recalibrate against reality, the knowledge base sharpens, and the platform publishes its live precision for everyone to see.
- Confirm or dispute becomes labels
- Live precision vs backtest
- Retraining on real feedback
126M
readings pushed through the ingestion path in load testing
2 kHz
per-signal high-rate capture with real-time overshoot detection
24
industrial connector types, from OPC UA and PI to Modbus, Sparkplug B, Kafka and SAP
ISA-18.2
alarm KPIs, flood detection, and bad-actor analytics built in
The platform
Everything the loop needs, in one platform.
From 2 kHz vibration capture to the Word report on the ops manager's desk, every stage runs in one system that remembers everything it learns.
Machine learning in production
Train, deploy, and monitor models on your historian, with SHAP explainability, live predictive findings, and drift detection. Every model is measured against live outcomes.
AI investigation copilot
Launch a case from a finding, roll several into one, and let the copilot rank causes and draft the RCA. Every answer is grounded in your data and cited to its source.
Document intelligence
Upload datasheets, SOPs, and manuals. In-Sight mines them for engineering limits and specs, and grounds every answer in the exact page.
High-rate historian
Capture up to 2 kHz per signal, detect momentary overshoots in real time, and zoom from months to milliseconds on demand.
Alarm management
Ingest OPC UA alarms and events, measure the fleet against ISA-18.2 and EEMUA benchmarks, and hunt floods, chatter, and bad actors.
Actions and work orders
Turn a confirmed finding into a work order with the evidence chain attached, dispatched by email, with optional SAP write-back.
Reports and rounds
Compose reports with live charts and AI-drafted sections, deliver them as Word documents on a schedule, and capture operator log sheets straight into the historian.
Human-confirmed limits
Extracted specs are never auto-armed. Every engineering limit is a deliberate, provenanced human decision.
Role-based access
Granular RBAC across workspaces, areas, and roles, so the right people see and act on the right things.
By industry
The same loop, shaped to your plant.
By capability
Go deeper on the parts that matter to you.
- Anomaly detectionDetection an engineer can interrogate, not a black box that fires.
- Alarm managementAlarm management measured against ISA-18.2, not against opinion.
- Root-cause investigationInvestigate with evidence. Close with a disposition that sticks.
- ConnectorsConnect the plant you have. Backfill the history you already paid for.
Why In-Sight
Built to be trusted with a plant.
Grounded, not guessed
Every AI answer cites your data or your documents to the page. When the platform cannot source a fact, it abstains. It never invents a number.
It closes the loop
Operator feedback becomes model ground truth. You can watch a model's live precision improve as your team teaches it.
Industry-agnostic by design
No hardcoded assumptions. In-Sight derives everything, from modes and limits to language, from your plant's actual configuration.
Security & trust
Trust is the product.
Industrial teams do not adopt AI they cannot verify. In-Sight is engineered so every recommendation is explainable, every number is sourced, and every consequential action is a human decision.
See your plant's data
finally work for you.
Book a walkthrough on your own equipment and watch the full loop run in one session: detect, investigate, act, learn.