Where the margin leaks on a construction project

Ask a construction business where a project lost its margin and you will usually get a story about one event — a client who would not pay for a variation, a subcontractor who walked, a material price that moved. Occasionally that is true. More often the single event is the one people remember, and the margin actually went in small amounts across the whole job, in places nobody was watching in time.

The common factor is lag. Committed cost is known weeks after the commitment was made. By the time the report shows an overrun, the concrete is poured and the argument is unwinnable.

Leak one: cost is reported, not controlled

Most cost reporting is built from invoices, which means it tells you what has been billed rather than what has been committed. A purchase order raised on Tuesday for material arriving in three weeks and invoiced a month later does not appear in any report until the money is effectively gone.

Controlling cost rather than reporting it means the commitment enters the system when it is made. Budget, committed and actual sit side by side per cost code, and a requisition that would breach the budget is stopped at the point of raising it rather than explained afterwards. That single change — checking the budget before the commitment instead of after the invoice — is the highest-value thing most contractors could build.

A cost report built from invoices tells you where the money went. A commitment ledger tells you where it is about to go, which is the only point at which you can still act.

Leak two: variations agreed on site, disputed in the office

A client representative asks for something on site. The site engineer agrees, because refusing is awkward and the change is small. Nobody raises a variation. Three months later the subcontractor claims for it, the client disputes it, and the only evidence is two people's recollection of a conversation next to a scaffold.

This is where an offline-capable mobile app earns its cost, and where most implementations get the design wrong. It has to be usable in under a minute, standing on site, possibly wearing gloves, with no signal. Photo, voice note, location, who requested it, timestamp — captured at the moment, synced later. If it requires a laptop back at the site office, it will be done at the end of the week from memory, or not at all.

Leak three: procurement leverage that exists on paper

Head office negotiates a rate contract. Sites buy locally anyway, because it is faster and the site engineer under time pressure does not know the contract exists. The saving is real in the document and absent from the ledger.

Enforcement has to happen at the point of purchase, not in a monthly review. When a requisition is raised for an item covered by a contract, the contracted supplier and rate should be the default, and buying outside it should require a reason that is recorded. Not blocked — construction sites have genuine emergencies — but recorded, so the pattern becomes visible.

The same fragmentation shows up in stock. One site orders material while another has it sitting unused, because there is no shared view. A consolidated position across sites, with transfer as an ordinary workflow, recovers a surprising amount.

Leak four: subcontractor bills that nobody measures against

Subcontractor claims arrive as a percentage of completion or a quantity of work. Verifying them properly means measuring against the BOQ and against what was actually certified — and under time pressure they get approved on trust, because the alternative is delaying a payment and a difficult conversation.

Structured measurement against the bill of quantities, with retention, advance recovery and previously certified amounts calculated rather than tracked in a spreadsheet, removes most of this. It also produces the record that makes a dispute winnable, which matters more than the individual overpayment.

Leak five: plant that nobody is accountable for

Equipment moves between sites without being recorded. Nobody is quite sure what is hired and what is owned, or whether the hire on that excavator was terminated when it came off site. Utilisation is unknown, so hire-versus-buy decisions are made on instinct, and the same machine is hired in while an identical one sits idle forty kilometres away.

A register with scan-based custody transfer fixes the record, and the record is what makes the commercial decisions possible: utilisation per machine, cost per hour including maintenance and idle time, and hire contracts with end dates that surface before they auto-renew.

The compliance leak, which is different in kind

Certificates, insurance, licences and subcontractor compliance documents tracked in a spreadsheet and discovered expired on the day they are needed. This one does not leak margin gradually — it stops work, and occasionally it stops work during an inspection, which is considerably more expensive.

Expiry dates as data with escalating reminders is a trivial thing to build and one of the highest-return items on the list, precisely because the failure mode is discrete and severe rather than gradual.

What to build first

In order of return, and it is fairly consistent across contractors we have looked at: the commitment ledger with budget checks first, because it moves cost from reporting to control. Then the site app for variations and daily progress, because it captures evidence at the only moment it exists. Then subcontractor measurement, then plant custody, then compliance expiry.

Document search across drawings, specifications and RFIs is worth mentioning separately. It is the one place a language model clearly earns its cost on a construction project — answering "what did we agree about the basement waterproofing detail" with the source document cited, instead of three people searching a shared drive for an afternoon. It is also a later phase; the cost and commitment layer comes first.