Articles · Construction

Full order book but no money. The builder's estimates were accurate

12 October 2026

One job follows another, the teams are busy, and at the end of the quarter there is less money than there ought to be. The owner of a building firm has a number for this and knows it by heart: on finished fixed-price jobs, hours came in 20.5% over the estimate. The conclusion suggests itself — the estimates are too tight. A vendor has already turned up with AI estimating.

The number comes from the report on Skarvholmen Bygg og Rehab. It is an invented Norwegian company from our sample reports, and the report on its data was written by our AI agent. The owner's number turned out to be right. What was wrong was the conclusion drawn from it.

Of the 20.5%, two points belong to the estimate

The firm's three carpentry teams log their hours every day, split by type: contract work, extras, time-and-materials work, warranty work. The answer had been sitting in the time records all along. The hours were simply being added into one total per job.

Split them out, and this is the picture. Across 19 finished jobs, contract work took 4,156 hours against 4,070 estimated. That is 2% over. All the rest of the overrun is extra work.

Hours are 20.5% over the estimate. Contract work is 2% over. The rest is extras.

So the estimates are accurate. Bathrooms and renovations look worse than other jobs because they have the most extra work, not because the estimates are off.

Extra work is done today and priced later

Extra work ought to bring money in, not eat it. Here it eats it, even though every step, taken on its own, is sensible.

When something unexpected turns up on site, the team foreman logs a change order with a photo the same day, and the work is done straight away. The customer agrees verbally, on the spot. The project manager is then supposed to price the change and send the price to the customer. Each one takes around 20 minutes, and 58.67 of them arrive a month.

From here, changes come in two kinds. What the customer has asked for is always priced: 56 out of 56, in 5 days. Unforeseen work was priced in 46 cases out of 116, and where it was priced, it took 17 days. It waits until the project manager finds the time, and on the renovation projects that manager is running ten to twelve small jobs at once.

The last step springs the trap. The accounts team invoices only changes that have an agreed amount. The final invoice goes out on average 7 days after the job is finished, and whatever is still unpriced on that day is closed without an invoice. No claim has ever been sent to a customer after a final invoice. None of the people asked could say who decides when the final invoice is sent.

68 pieces of finished work that were never invoiced

In three months, 103 of 172 completed changes did not make it onto an invoice. Of those, 68 are closed for good — 499.5 hours of work. Of the 68, 53 were never priced, and 15 were priced and sent to the customer, for 134,500 NOK excluding VAT, but the final invoice went out before the customer replied.

At the firm's rate of 720 NOK an hour, those 68 changes are worth about 409,180 NOK excluding VAT, not counting materials. That is an estimate of the work done, not a sum anyone is promising to get back.

From the inside the loss cannot be seen, because it has no line of its own. The team worked, and the hours were paid. The customer agreed. The accounts team invoiced everything that had an amount. An invoice that was never raised appears neither in costs nor in what customers owe. It surfaces in one place only — hours over the estimate — and there it gets taken for an estimating error.

A second hole of the same design turned up nearby. In 152 of 345 cases where goods were collected in store on the company account, no project number was given. That is 281,331 NOK in three months: the goods were booked as sundry consumables and never reached a customer invoice. How much of it should have been charged on to customers, nobody knows.

Of the vendors' four offers, none is needed now

• AI estimating — 60,000 NOK to set up and 8,500 a month. It improves something that is already accurate.
• An AI agent for payment reminders — 15,000 and 3,900 a month. Customers pay 11 days after the invoice date. Of 1,519,898 NOK in outstanding invoices, 76,712 is overdue, and 31,875 of that is agreed retention. The agent would chase that as well.
• Digital signing by the customer — 25,000 and 2,400 a month. On its own it will not help: the amount still has to be set by the project manager.
• An assistant that drafts price proposals — 35,000 and 5,900 a month. The only one of the four aimed at the bottleneck. Even so, it is worth trying no sooner than eight weeks from now, and only if unforeseen work is still not being priced the same week.

Two rules come first, and nobody has to be paid for them. All changes are priced on the same day each week, and the final invoice does not go out while the job has open changes; for that, the owner decides who releases the final invoice. And nobody collects goods on the company account without a project number — something to agree with the store.

The rules are free in money, not in effort. The project managers' pricing time goes up from 11.78 to 19.56 hours a month between them. No payment the firm makes goes away, so the guaranteed saving is 0 NOK. Any additional invoicing has to be measured; it cannot be promised.

Three numbers a building firm can work out today

1. The overrun in hours, separately for contract work and for extras. Time records, finished fixed-price jobs over the last quarter. If your hours are logged as a single line per job, that is your first finding: start logging them by type of work. The warning sign is a gap between the overall overrun and the overrun on contract work — 20.5% and 2% in the example.
2. The share of completed extras closed without an invoice. Take your last five finished jobs. In one column, everything the foremen logged as changes, wherever they logged it. In the other, the lines on the customer's invoices. In the example, 52% of changes on finished jobs were left without an invoice. Any piece of work that was done and is missing from the invoices deserves a look, all the more so if every such piece is of one kind: unforeseen.
3. How many days a price for unforeseen work takes to reach the customer — and how many days after the job is finished the final invoice goes out. The first comes from the dates in the change log, the second from the completion dates and the final invoice dates. In the example, 17 and 7. As long as the first number is bigger than the second, changes that turn up late in a job will struggle to make it in time, whoever does the pricing.

If the first of these three shows contract work itself running over the estimate by the same percentage as the overall total, then the estimates really are the problem, and a conversation about estimating software makes sense. In our example the estimates needed no fixing. What needed fixing was the week between the end of the job and the final invoice.

All articles