Bulgarian road-construction group · ~300 employees · €100M+ revenue · in-house asphalt, concrete and fuel sites

Road Builder ERM: Fleet, Plants, Sites

Some 300 people, 180 trucks and machines, in-house asphalt, concrete and fuel sites, and more than €100M a year in public contracts. All of it ran on paper, Excel and what the accountants remembered. Each number was keyed in three times. Management only saw each month's results three to four weeks after the fact, while insurance and inspection deadlines sat in a single person's spreadsheet. We designed one platform to run the entire operation.

Industry: Construction & Infrastructure
Function: Operations
Region: Bulgaria

Results

< 1 day
to management reports, instead of 30: Projected figure, live data.; 180
vehicles, machines and plants managed in one place: Each expiry date under watch.; 20+
document types handled without manual entry: Projected: manual entry falls from 100% to below 10-20%.

The challenge

The client is a road-construction group in Bulgaria, with around 300 people and over €100M in yearly revenue. It builds and rehabilitates roads, motorways, bridges and tunnels under public contracts. Across several sites it runs asphalt and concrete plants, quarries, stone crushing and fuel stations of its own. Its fleet counts about 180 assets: heavy trucks, vans, cars, road machines and production plants.

Mechanisation, technical operations and procurement all worked on paper and in Excel. The only point where company data met was finance, so every other department waited on accounting.

The price of working this way

Everything typed three times. A figure went onto paper, then into Excel, then to the accountants. That cost 40-100+ hours a month, plus one to two weeks each month spent on consolidation alone.
Decisions without current numbers. Reports trailed reality by 21-30 days. Nobody could see live costs per site or per motorway lot.
180 assets, compliance tracked by hand. Dates for insurance, technical inspection, vignettes and tachographs lived in a spreadsheet. Miss one and a truck drives uninsured or a machine stops on a public contract.
A mountain of driver paperwork. By law each vehicle needs a waybill every day. That is about 2,000 paper waybills a month, stored for years.
Approvals by chat. Repairs and purchases were signed off in messaging apps, leaving no audit trail.
One person held the fleet. Knowledge about the vehicles sat with a single employee.

What we did

We started at "the paper is killing us" and finished with a platform that is specified, prototyped and costed. That meant process discovery in three departments, a complete specification, a working prototype of the interface, the architecture, and a delivery plan solid enough to sign a contract on.

Enterprise Resource Management built for this business, not a generic ERP

The platform is driven by documents and rests on three pillars: document AI, a single central register and automated workflows. It has its own web and mobile interface and connects to the accounting software finance already runs.

Area	Documents handled
Finance	Fuel invoices, incoming invoices, acceptance protocols, bills of quantities
Fleet	Waybills, log books, registration papers, fuel limit cards, insurance policies, inspection certificates, service documents
Site operations	Weighbridge tickets, daily work records, construction acts, bills of lading, accident reports, permits
People and contracts	Health cards, employment contracts, rental contracts, powers of attorney, customs declarations, calibration certificates

17 modules covering the business

Transport and mechanisation. Each vehicle and machine has a complete file. Fuel cards and stock at the company's own fuel stations are tracked, and fuel is reconciled against supplier statements. GPS routes are recorded. Waybills are exported in the format the traffic police require, so they can be retained as the law demands. Service requests follow approval thresholds set by amount.
Compliance calendar. Insurance, inspection, vignettes, tachographs, licences: every expiring document, with reminders 30, 15 and 5 days ahead.
Projects and costs. Work is organised into projects, sub-sites and cost categories. Invoices are split across projects, and cost per site and lot is live.
Procurement, staff and assets. Requests and purchasing, staff files including licence categories, fixed assets with depreciation.
Dashboards for management fed by live data, not a report that is a month old.
Link to accounting. Our own Microinvest connectors, running on self-hosted n8n, keep Microinvest and Plus-Minus in two-way sync. Invoices go out; payroll, clients and depreciation come back. Finance stays on its software, and nobody types the figures a third time.
Bulk import from 14 source formats, covering everything from fuel statements and GPS routes to insurance and payroll.
Security. Permissions set per role, per module and per action, two-factor authentication and a complete audit log. The interface scales to a phone, so drivers send waybills straight from the cab.

A prototype people could click through

We designed every key screen and built it as an interactive prototype. Mechanisation, operations and procurement tried their own workflows in it and gave feedback before any production code was written. In practice, this is where most requirements come to light.

A plan built to be delivered

Phase	What it covers
1 Foundation	Infrastructure, authentication, document AI and automation server
2 Core	Document core, transport and mechanisation, staff, administration
3 Business	Accounting integration, procurement, technical operations, reports, dashboard
4 Data	Import and export, security, performance
5 Go-live	QA, DevOps, acceptance

The plan runs five phases over 30-36 weeks, with about 3,800 build hours. Each phase has written sign-off and each deliverable has acceptance criteria. Defects follow a severity model with set resolution times, and support comes with uptime commitments.

Stack

Layer	Design choices
Front end	TypeScript, React, Vite, shadcn/ui, Tailwind CSS
Back end	PostgreSQL, Node.js, BullMQ and Redis, S3-compatible storage
Automation and AI	Google Document AI splitter, custom extractors, n8n (self-hosted)
Integrations	Microinvest via our own connectors, Plus-Minus, fuel-card data, GPS tracking
Operations	Docker, nginx, Sentry, GitHub Actions, hosting in the EU

The outcome

Every figure in this section is a projection based on the discovery work and the specification.

	Now	Projected, with the platform
Data typed by hand	100%, done three times	Below 10-20%
Management reporting	21-30 days behind	Live, in under a day
Lapsed insurance or inspection	Discovered by accident	Reminders 30, 15 and 5 days ahead
Document processing	Manual	60-90% faster, about 80% fewer errors
Approvals	Chat apps	Amount thresholds and an audit log

Fleet, plants, sites, procurement, people and reporting in one platform for a company of 300 people and 180 assets.
Over 20 document types handled automatically.
More than 100 hours a month given back, with accounting no longer the bottleneck.
Pays back in under 18 months, for about one fifth of what a comparable ERP rollout would cost.
The specification and prototype have value on their own. Each problem the client brought, from triple entry to slow reports to hidden compliance dates, maps to the module that fixes it.

The risk teams tend to overlook

A packaged ERP knows nothing about Bulgarian waybills or your plants. If documents do not flow in by themselves, within one quarter people are back in Excel and the system goes unused.

More case studies

Similar problems, measured the same way.

All case studies

EU funding agency, public sector · internal audit · nearly €1 billion paid out yearly

An Audit Assistant That Runs On Premise and Cites Every Line

auditor time saved on each control plan: ~80%; auditor time saved on each control plan
of generated lines backed by a checked citation: 100%; of generated lines backed by a checked citation

Independent school with nursery · London · roughly 200 children

An Automation Plan for an Independent School's Finances

of finance work each week: 20 h → 1 h; of finance work each week
freed every year: 780 h; freed every year

Bulgarian accounting practice · 10-15 people · bookkeeping for 100+ companies

AI That Reads and Posts Accounting Invoices

of documents go through uncorrected: 97%; of documents go through uncorrected
less data entry every year: 576 h; less data entry every year