We're Building a Construction ERP — The Safe Way

Editor's note (July 2026): This article is the manifesto for The ERP Experiment — our attempt to build a full construction ERP for UK SMEs, in the open, one module at a time. The live module board and roadmap show exactly where the build stands today, and every release gets a Build Log entry in Resources. Start here, then follow along.

AI can generate a construction ERP. It still shouldn't be trusted with job costing, retainage, or month-end close. That is the real test.

In construction, the hard part is not screens. It is whether the system holds up when approvals are partial, labour moves across jobs, back charges hit late, and billing exceptions pile up.

That's why we're building The ERP Experiment in the open — as a series of focused modules, released one at a time, with the financial core treated with the caution it deserves.

👉 See the ERP hub for the live module list, or the full roadmap for the phased plan.

What AI does well

AI is strong at first-pass output. It can scaffold CRUD apps, generate schemas, draft service layers, build approval flows, create role-based screens, and spin up reporting views fast.

That is useful when teams are building around the ERP — custom reporting, document capture, approval routing, subcontractor compliance, portals, and project-specific workflows. These are bounded problems. Lower blast radius. Faster iteration. Clear payback.

Where construction ERP breaks

Construction ERP is a system of financial truth tied to project execution. This is where teams get fooled: the demo looks clean because the hard part is hidden.

The real complexity sits in job costing, work in progress, retainage, subcontractor billing, payroll allocation, change orders, cost transfers, compliance, and month-end reconciliation.

Failures usually start in the gaps between systems and timing. A retainage release posts wrong across partial approvals and back charges. Now billing is off. Cash expectations are off. Month-end close is off. Executives are making decisions on distorted numbers. That is not a workflow bug — it is a financial systems bug.

ERP bottlenecks are rarely about coding speed. They are about requirements accuracy: does the data model match how the business actually runs, do permissions preserve control, can posting logic survive ugly exceptions without corrupting financials, and can someone audit the result six months later and explain exactly why a cost hit the wrong job?

AI does not remove that work.

The safest adoption path

Build around the ERP before trying to replace the ERP. We use AI as a co-builder for extensions and experience layers, not as a rip-and-replace engine for the accounting core.

Use AI where speed matters and mistakes are recoverable:

• custom reporting apps
• approval routing
• document capture
• vendor or subcontractor portals
• project workflows
• tools that read from ERP data without rewriting the posting logic underneath it

Keep humans firmly in charge of:

• system architecture
• source-of-truth design
• posting rules
• identity and permissions
• exception handling
• auditability
• financial validation over time

AI can write screens. It cannot absorb the cost of broken posting logic.

How we're sequencing it

That principle drives our roadmap. We ship the low-risk modules first — document capture, approvals and purchase orders, subcontractor portals, reporting — then move into the financial core (job costing, applications for payment and retentions, payroll allocation, and finally WIP and month-end close) only once the foundations are proven.

Before any module touching financials rolls out, we test it against live historical project data across retainage, change orders, payroll allocation, and month-end close exceptions. That validation is not optional: a system that works in a sandbox can still fail fast on real projects.

One subscription, every module

The whole suite is unlocked by a single ERP Experiment subscription — not per-module pricing. Subscribe once and each new module becomes available automatically as it's released. You can start with the low-risk modules today and grow into the financial core as it's validated.

So yes — you can build a construction ERP with AI. But a production-grade one is won in accounting logic, controls, exception handling and trust. AI is an accelerator. It is not a shortcut around construction complexity.

👉 Ready to follow along? Explore the ERP hub and the roadmap.