The Missing Link: Connecting Critical Control Failures to Maintenance Data - a Mitchell Services Case Study Stage 2
- 1 day ago
- 9 min read
This is the second instalment of the Mitchell Services case study. Stage 1 — documented in partnership with IBM — established how CaNeTA revealed the gaps in Mitchell Services' critical risk management using incident data, critical risk standards, and their corporate risk register. You can read the Stage 1 case study here.
Stage 2 is a different story. It's what happened nine months later — when the network grew, two independent systems spoke to each other for the first time, and a finding emerged that couldn't have come from incident data alone.
Key Takeaways
Connecting safety verification and maintenance data revealed risks that neither system could show alone. Mitchell Services completed 56,000 Critical Control Verifications in 2025. Their maintenance system logged hundreds of unplanned work orders against the same controls. Those two datasets had never been analytically linked — until CaNeTA connected them, and a critical finding emerged: pressure-energy controls were degrading faster than anyone had assumed, confirmed simultaneously by both systems.
Interventions from Stage 1 were independently verified by the network. Nine months after Mitchell Services acted on Stage 1 findings — tightening vehicle inspections and redesigning a fatal-risk task — the causal network confirmed those actions had worked, without being told. Unsecured loads dropped in the risk hierarchy. The redesigned task disappeared from the network entirely.
337 of 449 critical controls showed active failure signals. When CCV failure data and unplanned maintenance work orders were brought into the network, the scale of control degradation became visible for the first time. The highest-priority signals pointed to pressure-energy controls — the ones managing hazards that are hardest for workers to perceive.
The critical control register contained items that weren't critical controls. Pre-start checklists and tagging-and-testing dates were included in the CCV dataset, diluting its signal and creating false confidence in control performance. Removing them sharpens the picture and directs verification effort where it actually matters.
Four actions emerged that couldn't have come from incident data alone — prioritising pressure-energy control robustness, permanently linking CCV failures to work orders, establishing gas monitor placement standards, and cleaning up the critical control register.
The Context: A Contracting Business With Genuine Risk
Mitchell Services is one of Australia's leading drilling services contractors, operating across some of the most remote and demanding mining environments in the country. Their crews work on mobile and fixed drill rigs where heavy rotating equipment, hydraulic systems, and high-pressure components create both physical and operational risk — every shift, in every environment.
Mitchell Services' HSE team is small and resourceful. They can't rely on large programs or expensive initiatives.
Their GM People, Risk and Sustainability, Josh Bryant, describes their philosophy clearly: "Our focus has never been on adding more safety tasks. It's been on improving how work is designed, set up and carried out."
That philosophy is what makes the CaNeTA story worth telling.
Stage 1 Recap: Finding What the Risk Register Couldn't Tell You
In the first phase of CaNeTA analysis (FY2024/25), Libero AI ingested three core datasets from Mitchell Services: incident reports, critical risk standards, and the corporate risk register. The resulting causal network - built from 1,411 individual events - revealed something the documents alone could not: how risk actually flows through the operation.
The network identified unsecured loads as the highest-priority node in the system, sitting at the intersection of nearly every upstream failure pathway and pointing toward the most severe downstream outcomes. It surfaced a person-struck-by-tooling risk that had not been fully visible in the register. It showed where the system was fragile, where controls were load-bearing, and where interventions would have the most leverage.
Mitchell Services acted on those findings. They tightened their vehicle inspection regime, moving from generalised pre-starts to in-depth inspections of critical vehicle systems. They redesigned the task associated with a high-potential lifting-chain event at BMA, effectively eliminating the exposure from the work entirely.
Then they waited nine months to see what the network would say next.
Stage 2: The Extended Analysis
For Stage 2, the dataset grew substantially:
Stage 1 incident data continued — the same network extended forward in time
9 months of new incidents — post June 2025, capturing the period after Stage 1 interventions
CCV failure data — 56,000 individual Critical Control Verifications completed in 2025 alone, at a 0.81% failure rate (~401 instances of a control not performing as expected)
Unplanned Maintenance Work Orders — for the first time, maintenance system data was brought into the causal analysis
The Phase 2 network grew to 2,510 nodes, identified 449 critical control nodes, and matched 456 unplanned work orders directly to control degradation. Critically, 337 of the 449 critical controls showed active failure signals — either through CCV failures or work order pressure.
This was the first time Mitchell Services' CCV data and their maintenance work order data had ever been causally connected. They are independent systems. Before CaNeTA, they had never spoken to each other.
"CaNeTA formed a relationship between two independent systems." — Josh Bryant, GM People Risk & Sustainability, Mitchell Services
What the Network Showed: Confirmation First
The first thing Josh Bryant looked for in the Stage 2 network was whether CaNeTA was reflecting reality — confirming what the interventions from Stage 1 should have done. It was.
Unsecured loads dropped in the rankings. The tightened inspection regime had worked. The causal network picked it up without being told.
Person struck by tooling dropped. The task had been redesigned. The exposure didn't exist anymore — so there were no repeat events in the data — and the node's position in the network fell accordingly.
Near miss and injury rose in prominence. This sounds counterintuitive, but it's a sign of a maturing safety system. Mitchell Services records near misses as events that resulted in an injury, equipment damage, or neither. CaNeTA surfaced the fact that near misses propagate to outcomes — which is exactly what the data shows. A rising near-miss signal means the recording system is working and the network is truthful.
A new signal emerged: hazardous gases. Mitchell Services had seen a genuine increase in gas events — some from gas intersection during drilling, others from the failure of client infrastructure near drill sites. CaNeTA picked it up independently. It matched what Josh was seeing on the ground.
"Nine months on, CaNeTA showed me what I'd hoped to see. Unsecured loads — down. The lifting exposure we redesigned out of the work — gone. The new signals coming through, like hazardous gases, match what I'm seeing on the ground. The network is reflecting reality. That gave me the confidence to go deeper." — Josh Bryant
The Unlock: Two Systems That Had Never Spoken
With confidence in the network established, the real Stage 2 work began — connecting the CCV data and the unplanned maintenance work orders into the causal model for the first time.
The scale of the CCV dataset alone made this significant. 56,000 individual verifications in a single year. A 0.81% failure rate. That's 401 instances of a control not performing as it should.
As Josh put it: "There's got to be something connecting those — they can't just be 401 isolated opportunities for improvement."
CaNeTA answered that question. And the answer was the pressure-energy finding.
The Mic-Drop Moment: Pressure Energy
Mitchell Services uses an energy-based safety model — gravity, motion, electrical, pressure, temperature. It's a framework their people understand and work within every day. Here's what Josh observed:
"Our people are really good at picking up the hazards from motion and gravity. They're not so good at the silent ones — pressure, temperature — hidden behind a hose. So as a business we put controls in place so they don't have to. What CaNeTA has shown me is that those controls are degrading faster than I thought — and the work orders and the CCV failures both confirm it. That's a perfect storm we can now act on — with data, not gut feel."
The controls in question — pressure relief valves (PRVs), Kevlar sleeving, hydraulic burst protection — are the last line of defence for an energy type that people struggle to perceive. They are the controls that matter most precisely because the hazard is invisible.
CaNeTA showed that both the CCV failure data and the unplanned maintenance work orders were pointing at these controls. They were degrading at rates inconsistent with the assumption that they were reliable. And without connecting those two data systems, the signal was invisible — spread across a maintenance system and a verification system that had never been looked at together.
The network also drew a distinction between the surface fleet and the underground operation:
Surface fleet: wear and tear is the dominant failure mode for hydraulic controls
Underground: the issue is placement and discipline of use — controls present but not correctly positioned or calibrated
And on gas monitoring specifically: some monitors not in place, some not calibrated, some not where they should be. A discipline issue now visible in the data, not just in supervisors' observations.
The Register Clean-Up: What CaNeTA Found That Wasn't a Critical Control
The Stage 2 analysis surfaced another finding with significant operational implications: not everything in Mitchell Services' critical control register was actually a critical control.
CaNeTA identified two categories of items that were functioning as supporting activities, not barriers between energy and harm:
Vehicle Pre-Start Checklists — a supporting inspection activity. Pre-starts verify that controls exist; they are not themselves a control that acts on an energy pathway.
Tagging and Testing Dates — an administrative compliance check. Whether a tag date is current says nothing about whether the control will perform when energy is released.
The inclusion of these items in the CCV dataset was creating false confidence. The reported performance figures — including the 0.81% failure rate — were being diluted by activity that wasn't measuring control effectiveness. Removing them sharpens the signal and ensures that verification effort is directed at real barriers.
The true critical controls CaNeTA confirmed in the register: emergency stop buttons, Kevlar/hose burst protection, pressure relief valves, rotation guarding.
Four Actions That Couldn't Have Come From Incident Data Alone
The Stage 2 analysis produced four specific, actionable findings — all of them driven by the integration of CCV and maintenance data into the causal network:
01 — Prioritise pressure-energy control robustness PRVs, Kevlar sleeving and hose burst protection are degrading at rates inconsistent with the assumption they are reliable. A targeted improvement programme is required.
02 — Link CCV failures to work orders as a permanent monitoring cycle This connection must be maintained permanently. It is the leading indicator that shows whether controls are holding between scheduled verification cycles. CCV failures predict work orders — which means the system can act before failure reaches the field.
03 — Establish gas monitor placement and calibration standards Discipline-of-use issues require a defined placement standard with field verification of position, calibration status, and functionality — not just presence.
04 — Remove non-critical controls from the CCV dataset Pre-starts and tagging-and-testing dates are supporting functions. Removing them improves signal quality and ensures verification effort is directed at real barriers.
The Role of AI: From Visibility to Autonomy
One dimension of Stage 2 that Mitchell Services found particularly valuable was the ability to interrogate the causal network directly using Libero AI's Ask solution — an AI assistant trained on the CaNeTA analysis. Josh Bryant's view was direct:
"I don't want to have to reach out to a consultant every time I want to ask a question of my own data. We're not that kind of business. The AI lets me follow the thread myself — show me upstream of this node, tell me what happens if the PRV fails and the e-stop doesn't trip. CaNeTA gave me the visibility. The AI gave me the autonomy."
The ability to ask "what would happen if X failed and Y failed simultaneously?" — and receive a network-grounded answer rather than a consultant's opinion — changes the relationship between a safety leader and their data. It allows for rapid hypothesis testing. It allows for consequential thinking before events happen.
With critical risk legislation now live from June 2025, that kind of direct, evidence-based access to operational risk intelligence is no longer a nice-to-have.
What This Means for the Industry
The Mitchell Services Stage 2 story is significant beyond the operation itself. It demonstrates three things that are hard to achieve in practice:
1. Verification that interventions are working. Most safety programs have no reliable mechanism for confirming that an action taken actually changed the risk profile of an operation. CaNeTA provided that confirmation, without requiring a new incident to prove the point.
2. Integration of siloed data systems into a unified risk picture. The CCV system and the maintenance work order system had never been analytically connected before CaNeTA. The pressure-energy finding only exists because those two systems were joined through causal analysis. No amount of lagging indicator reporting would have surfaced it.
3. The shift from reactive to proactive. The four actions produced by Stage 2 are all leading-indicator-driven. They address control degradation in process, before failures cascade into incidents. That is the practical definition of proactive risk management — not a philosophy, but a data-driven practice.
About CaNeTA
CaNeTA (Causal Network Topology Analysis) is Libero AI's causal intelligence solution for safety-critical industries. It uses AI with human oversight to build causal networks from an organisation's existing data — incidents, risk registers, verifications, maintenance records, and more — revealing how risk actually propagates through an operation and where interventions will have the greatest effect.
CaNeTA is part of Libero AI's integrated solution set, which includes Compose, Enrich, Ask, CaNeTA, Orchestrate, and Pilot.
Ready to Unlock AI for Your Business?
Contact us today to learn how CaNeTA can help you:
Expose hidden risk pathways
Prioritise critical controls
Build resilience into your operations
