AI Case Study
Mining & Bulk Haulage
AI-Assisted Fleet Cost Intelligence
When standard data matching failed on messy industrial data, we deployed AI language models to extract the cost insights that conventional approaches couldn’t reach.
Client
A Fast-Growing WA Bulk Haulage Operator
1 Month (Cost Study)
Scope
200+ Vehicle Fleet, Lifecycle Cost Analysis
The Situation
Growth Without Visibility
A fast-growing Western Australian bulk haulage operator — $259M in revenue, 200+ prime movers, 550+ trailers, and 41 million kilometres driven annually — needed visibility into what its fleet actually cost to maintain and operate. The company had experienced extraordinary growth, achieving a 36% revenue compound annual growth rate, but that growth had outpaced its ability to understand its own cost base.
Purchase orders and maintenance records were captured in free-text descriptions across multiple disconnected systems. No structured cost categories existed. A brake pad replacement might appear as “brake pads KW T909 front axle LH” in one system and “front brake service — Kenworth” in another. There was no way to aggregate, compare, or analyse these records at scale. Equipment was being replaced, maintained, and fuelled based on schedules and gut feel rather than data-driven decision-making.
The client's team had attempted this work and failed to deliver. The operator’s leadership was sceptical but recognised the problem was only getting worse. They needed a structured view of total cost of ownership across the fleet — but the data was too messy, too inconsistent, and too industrial for conventional analysis tools to handle.
The Challenge
Why Conventional Approaches Failed
The data quality problem was not a minor inconvenience — it was the central obstacle. Standard data tools and matching approaches could not process what this operator’s systems had produced over years of rapid, operationally-focused growth.
Data Quality
Unstructured Data
Purchase order descriptions were captured in free text by procurement officers, mechanics, and parts suppliers. “Brake pads KW T909 front axle LH” cannot be matched to a maintenance category using standard keyword or code-based matching. The language was industrial, abbreviated, inconsistent, and deeply contextual.
System Fragmentation
Multiple Systems
Procurement, maintenance, fleet management, and financial data lived across disconnected platforms. Each system captured information differently, with different identifiers, different granularity, and different update frequencies. There was no single source of truth for what a vehicle cost to operate.
Track Record
Previous Failure
An internal team attempted to deliver fleet cost analysis and could not complete the work. This meant expectations were tightly managed — the client had experienced a failed attempt and needed confidence that a different approach would produce a different result.
Operational Scale
Scale & Growth
More than 200 prime movers, 550+ trailers, $259M in annual revenue growing at 36% per year. The fleet was expanding faster than the organisation’s ability to understand its cost structure. Every month of delay meant more capital deployed without lifecycle cost intelligence.
What We Did
AI Where It Actually Mattered
Adaptive deployed a focused, practical approach that combined conventional data engineering with targeted AI application. The AI was not applied everywhere — it was applied precisely where conventional tools failed: classifying unstructured, free-text purchase order descriptions into meaningful maintenance categories.
01
Data Extraction & Unification
Pulled procurement records, maintenance histories, fleet registrations, and financial data from multiple source systems into a unified analysis environment. Established vehicle-level linkages across systems that had never been connected, creating the foundation for lifecycle cost attribution.
02
LLM Classification
Applied Python-based large language model processing to match unstructured purchase order descriptions to structured maintenance categories. The LLM could interpret industrial shorthand, contextual abbreviations, and inconsistent naming conventions that deterministic regex and keyword matching could not solve. This was the critical differentiator — the step that their internal team could not deliver.
03
Cost Intelligence Build
Built lifecycle cost visibility per vehicle, per maintenance category, per time period. For the first time, the operator could see what each prime mover actually cost to maintain across its lifecycle — revealing total cost of ownership patterns that were invisible in the raw, unstructured data.
The Outcome
From Unstructured Data to Operational Intelligence
The engagement delivered what the internal team could not: structured fleet cost data from unstructured industrial records. But the impact extended well beyond a single cost analysis. It established a data foundation, proved AI’s practical value, and launched a broader advisory relationship.
Lifecycle Cost Visibility
Fleet cost data structured and categorised for the first time across 200+ prime movers. Total cost of ownership per vehicle now visible where previously only raw, unstructured purchase orders existed.
LLM Proof of Concept
Demonstrated that AI language models solve real industrial data problems that conventional tools cannot. Not theoretical AI — practical intelligence applied to messy, real-world operational data.
Trust Established
The engagement expanded from a one-month cost study to an ongoing advisory relationship. Delivering where a previous consultant had failed rebuilt confidence in external advisory support and opened the door to broader transformation work.
Data Foundation
A structured data asset that can now feed ongoing fleet analytics, predictive maintenance modelling, and the AI-for-BI platform. The classified cost data becomes more valuable over time as new records flow through the same LLM pipeline.
Why This Matters
AI in Industrial Operations Is Not About Chatbots
This engagement proved something important: AI in industrial operations is not about chatbots or dashboards. It is about solving the data problems that have defeated conventional approaches — messy, unstructured, real-world operational data that does not fit neatly into categories.
The LLM did not replace human analysis. It made analysis possible where it was not possible before. A purchase order reading “brake pads KW T909 front axle LH” is meaningless to a standard database query. To a language model trained on industrial context, it is a brake maintenance event on a specific vehicle type that can be classified, aggregated, and analysed alongside thousands of similar records.
This is the pattern for practical AI in heavy industry: not replacing people, but reaching the data that people could never systematically access. When eighteen years of domain expertise meets modern AI capability, you get intelligence that neither could produce alone.
Your fleet data probably has the same problem. Let’s find out.
If your maintenance records live in free text and your cost reporting relies on manual categorisation, there is intelligence locked in your data that conventional tools cannot reach. We can show you what AI makes possible.