Case studies / Data foundations
Data foundations · Fleet operations
A production data warehouse that unifies a fragmented fleet-operations estate
A large fleet operator's business ran across systems that never talked to each other — agreements, maintenance, fleet management, a legacy back-office system, and third-party trip data — so no report could see the whole business, and the system most people trusted as the source of truth was quietly missing years of records. We built and run the production data warehouse that consolidates those sources into one canonical model, automated the reporting on top of it, and operationalised the analytics that sit on the foundation.
The client: a large vehicle-fleet operator, with vehicles, drivers, agreements, maintenance, and trip data across the business. Our role: engineering delivery partner.
Running in production
A production data warehouse on Google Cloud consolidating the operational sources into one queryable model.
The unified model recovers years of records the legacy “system of record” had been missing.
Operations, maintenance, receivables, driver-revenue, and marketing-attribution reporting now run on automated pipelines.
An accident-cost prediction model operationalised into the production pipeline, with the conversational assistant kept as an early pilot.
The challenge
A business running on data that couldn't see itself
A fleet operator at scale runs on data spread across systems that were never designed to connect — agreements in one place, maintenance in another, fleet management in a third, a legacy back-office system holding historical financials, and third-party trip data on top.
Not only could no single report span the business, but the legacy system most people treated as the source of truth was quietly incomplete — missing whole years of records that lived only in other systems. Decisions were being made on a partial picture, without anyone knowing how partial it was.
The engineering challenge was to reconcile those heterogeneous sources into one trustworthy canonical model, supplement what the legacy system missed, and make it stable enough for automated reporting, machine learning, and analytics to share the same foundation.
Before
- Operational data fragmented across agreements, maintenance, fleet management, legacy back-office, and third-party feeds
- The legacy “system of record” silently missing years of records held elsewhere
- Reporting compiled manually and ad hoc; no whole-business view
- Analytics and ML ambitions blocked by the absence of a trustworthy foundation
After
- One production warehouse with a unified canonical model spanning the sources
- A unified view that recovers the records the legacy system was missing — evidenced, not asserted
- Automated production reporting across operations, maintenance, receivables, driver revenue, and marketing attribution
- An accident-cost prediction model running in the production pipeline; a conversational assistant in early pilot
How we approached it
Unify the sources. Automate the reporting. Layer AI in the right order.
Unify the sources — and prove the gain
We built the warehouse so the unified layer reconciles the operational systems and legacy back-office data into comprehensive coverage — and showed concretely how much the legacy source of truth had been missing. Trust comes from completeness you can evidence, not assert.
Automate what used to be manual
With a reliable foundation in place, we moved operations, maintenance, receivables, driver-revenue, and marketing-attribution reporting onto automated production pipelines, so the numbers are current and consistent rather than hand-assembled.
Layer AI on the foundation
We took an accident-cost prediction model out of notebook form and integrated it into the production pipeline for stable, scheduled execution, then stood up a conversational assistant over the curated data as an early pilot.
The warehouse, source unification, automated reporting, and the accident-cost prediction model are in production. The conversational assistant is an early pilot — each described in the tense it has earned.
Under the hood
Next step
Is your business running on data that can't see itself?
We consolidate fragmented operational estates into one trustworthy production data foundation — then layer reporting and AI on top, in the order that keeps them honest.
Case study · details anonymised to protect client confidentiality