An AI- and IoT-based fleet management platform

A custom, modular fleet management platform for a confidential industrial partner, built around a formula engine that turns contract-specific pricing rules into configuration instead of code — plus live vehicle telemetry, AI-based document understanding, and a finance module deep enough to post straight into the books.

Client

Confidential industrial partner

Industry

Fleet management / automotive services

What we built

An AI- and IoT-based fleet management platform

Service

Product development · System architecture · AI & IoT integration

The challenge

Running a large vehicle fleet business means constantly juggling contracts that are never quite alike: different financiers, different pricing and valuation rules, different fee structures — and a new contract shape shouldn't mean waiting for the next software release.

The business needed one system instead of many scattered tools and spreadsheets: CRM, sales, vehicle and service records, finance, reporting, notifications and access control were split across separate tools, and vehicle telemetry — location, engine diagnostics, tyre pressure — sat completely outside any of them.

On top of that, a constant stream of paperwork — dealer offer PDFs, leasing documents, incoming and outgoing invoices — needed structured data pulled out of it by hand, one document at a time.

The solution: a modular platform

We build the platform as a technology partner, and it's in active development today. It's a domain-driven modular monolith over a single relational database, explicitly designed to be split into services later if the business needs it — with modules for CRM, sales, vehicle, service, finance, reporting, notifications and access control.

An expression engine, not a formula field

At the platform's core is a full formula language — a tokeniser, parser, abstract syntax tree and interpreter — with Excel-compatible built-in functions (including PMT, the financial annuity calculation), automatic dependency resolution with cycle detection and topological sorting, and an editor with syntax highlighting and cursor-aware autocomplete. Pricing, fee and valuation rules that differ per contract are expressed in the engine itself, so a new contract shape needs no code release.

Finance deep enough to post into the books

Leasing contracts carry a principal/interest split with residual-value tracking, grouped and reported per financier. A financier's invoice is broken down line by line, with separate interest-discount documents merged back into the original by contract number. Principal and interest post separately per vehicle into the bookkeeping system, with paid status read back automatically from a daily bank feed. Incoming invoices go through payment verification, approval routing, and automatic matching against tax-authority records and email attachments; outgoing invoicing handles recurring fees, edit-before-issue, distribution and dunning. When a contract is extended, residual value recalculates automatically — and if a financier declines the extension, the contract is quietly novated to a different financier behind the scenes, while the customer simply experiences a normal extension.

Live telemetry from every vehicle

Vehicle location, engine diagnostics and tyre-pressure data stream in continuously and in real time, and the system alerts us the moment something falls outside the normal range.

AI that reads documents, not just chats

A provider-abstracted LLM layer extracts structured vehicle parameters straight out of dealer offer PDFs — document understanding rather than a chatbot bolted on the side.

Built to connect, not to be rebuilt

External ERP, CRM and financial systems connect through an adapter-and-plugin pattern instead of point-to-point code, a central history module tracks every field-level change with diffs, and the whole platform is multi-tenant with tenant isolation, enriching counterparty records automatically from a company-data API.

How we planned and delivered it

Feature-level planning: a per-feature effort and dependency plan with explicit definitions of done, tracked quarterly against actuals — not a single up-front estimate.
Architecture built to split later: a domain-driven modular monolith over a single relational database, explicitly designed so any module could be pulled out into its own service if the business ever needs it.

Do you also plan to develop your own ERP?

Let's start with a conversation about your workflows.

Book a time →