AI Job Costing for Trades: Which Jobs Actually Made Money

You quoted 40 jobs last month, and AI job costing for trades helps reveal what happened next. You won 14. You know exactly what those 14 invoiced out at, because that number sits in your accounting file where you can see it. Ask which of them actually made money and the honest answer is usually a shrug and a gut feel about the two that felt painful.

That gap is what this approach closes. Not another app to log into. A layer that reads the data you already produce, quoted hours, timesheets, supplier invoices, purchase orders, variations, and hands you the margin on each job while you can still do something about it.

This post covers where margin actually leaks on a trades job, the four numbers every completed job should give you, and how to start without replacing a single system you already run.

Key Takeaways

• AI job costing for trades pulls quoted hours, timesheets and supplier invoices into one margin figure per job, not one figure per month.
• Most trades businesses find out a job lost money six to eight weeks after the crew left site, which is far too late to fix anything.
• Variations are the single biggest leak. Extra work agreed verbally on site and never invoiced is pure margin gone.
• Four numbers per job tell you almost everything: quoted versus actual hours, materials variance, variations captured, gross margin.
• Labour is where jobs die. A 10 per cent hours blowout on a labour-heavy job costs you more than a materials price rise.
• Start with your last 20 completed jobs. The pattern in those is worth more than any new pricing model.
• You do not need new software. Job costing runs on data already sitting in your accounting file, your job app and your supplier statements.

Why Trades Businesses Find Out Too Late

The information arrives in the wrong order. Timesheets come in on Monday, sometimes. The merchant statement lands on the 20th of the following month. The variation you agreed to on site on a Tuesday afternoon exists as a text message and nowhere else. The job is marked complete in the app, invoiced, and mentally filed as done long before the last cost has even shown up.

By the time the monthly numbers are reconciled, you are looking at work your crew finished six to eight weeks ago. You cannot go back and re-price it. You cannot go back and invoice the extra day. The only thing that number is good for now is regret.

Then there is the aggregation problem. A monthly profit and loss tells you the business made money. It does not tell you that three jobs carried the month and four quietly bled. Averages hide exactly the thing you need to see. If your gross margin sits at a comfortable-looking number every month, it is entirely possible that half your work is subsidising the other half, and you are still quoting the losing half the same way you always have.

The last piece is who does the work of finding out. It is you, at 9 pm, with a laptop and a pile of dockets, because nobody else has the full picture in their head. That is the same pattern behind admin overload in a small business: the owner is the only person who can assemble the truth, so the truth only gets assembled when the owner has a spare evening.

What AI Job Costing for Trades Actually Does

It sits between the tools you already use and does the matching you never get around to.

The estimate lives in one place. Labour hours live in a job management app or a paper timesheet someone photographs. Materials arrive as supplier invoices in your accounting file, often bundling three jobs into a single bill. Variations live in texts, emails and site notes. Nobody built these systems to talk to each other, so the reconciliation has always been manual, which means it happens late or not at all.

AI job costing for trades reads all of it and allocates costs back to the job. A merchant invoice covering three sites gets split by the purchase order references and the delivery addresses. A photo of a docket gets read and coded. A text saying “yep, go ahead and do the extra two power points” gets flagged as an uninvoiced variation before it evaporates.

The part that changes behaviour is the timing. Instead of a post-mortem, you get a warning. The job is at 78 per cent of quoted labour with the fit-out still to go. That message on a Wednesday morning is worth more than a perfect report in eight weeks, because on Wednesday you can still ring the client, still pull a second body off another site, still price the variation properly.

It does not need to be perfect on day one to be useful. Human in the loop is the right setting: the system proposes the allocation, you confirm or correct it, and it learns your job codes, your supplier naming, your quirks. Ten minutes a week of correction beats an evening a month of archaeology. It also builds something you have never had before, which is a clean audit trail showing exactly where each dollar of cost landed.

The Four Numbers Every Job Should Hand You

Quoted hours versus actual hours. The gap here is the single most useful number in a trades business. Not because you will hit the quote every time, but because a consistent gap tells you your estimating is wrong in a predictable direction, and predictable is fixable. If bathroom renovations always run 18 per cent over, that is not bad luck four times in a row. That is a pricing input.

Materials variance. Quoted materials against invoiced materials, including the bits bought twice because nobody checked the van. Price rises since you quoted show up here too, which matters if your quotes stay open for 30 days in a market where merchant pricing moves.

Variation capture rate. Of the extra work agreed on site, how much made it onto an invoice? Most trade businesses have never measured this because measuring it requires knowing what was agreed, and what was agreed is scattered across texts and memory. This is often the largest recoverable number in the whole exercise.

Gross margin per job, and per job type. Job by job matters, but the pattern across job types is where the decisions live. Maintenance callouts, new builds, insurance work and commercial fit-outs behave completely differently. Once you can see them separately, you can stop quoting the category that never works.

A fifth number is worth tracking if you can get it: rework and callback hours. Those never appear on a quote, so they never appear in your margin, and they land on the jobs that already went badly.

Where the Margin Actually Leaks on a Trades Job

Labour first, always. Materials are visible because they come with an invoice attached. Hours disappear quietly. Travel between sites, yard time loading, the half hour waiting for a delivery, the apprentice sent back for a fitting. None of it is on the quote and all of it is paid.

The second trip is the classic. A job priced as one visit turns into two because a part was wrong or the site was not ready. On a small job, the second trip can be the entire margin. Count those. Most owners are shocked by the frequency.

Then variations. Somebody rings you and asks for one more thing while the crew is on site. You say yes, because saying yes is good service and arguing about a hundred dollars in front of a client is not how you want to run a business. Nobody writes it down. Multiply that across a year, and it is a real number, often larger than the price rise you have been nervous about putting through.

Client-level patterns are the last one, and they are uncomfortable. One builder’s jobs always run over because the site is never ready when they say it will be. One property manager’s callouts always take longer than the description suggests. You cannot see this from a monthly P&L. You can see it immediately when the margin sits next to the client name.

Quoting time is worth naming as well. If you spend six hours a week pricing work and win one in three, the cost of the two you lose is loaded onto the one you win, and it never shows up anywhere.

How to Start AI Job Costing for Trades This Month

Start with history, not systems. Take your last 20 completed jobs and put the quote, the final invoice, the labour hours and the supplier costs side by side. Do it manually the first time if you have to. It will take an afternoon, and it will tell you more about your business than the last year of monthly reports. The pattern that shows up in 20 jobs is usually the pattern in all of them.

Second, fix the coding. Costs can only be matched to jobs if jobs have consistent references and your crew and suppliers use them. This is the boring part, and it is also the whole game. Purchase order numbers on merchant accounts, job numbers on timesheets, one naming convention everyone follows. Automation cannot allocate a cost that has nothing on it to allocate against.

Third, automate the pull. Once the coding is consistent, the weekly cost summary on open jobs can be generated without anyone touching it, and it can land in your phone on a Monday morning instead of living in a report you have to remember to run.

Fourth, add the alerts. Labour burns past a threshold on an open job. A variation mentioned in an email that has no matching invoice. A job closed with costs still arriving.

One warning. Do not start by trying to automate the estimating judgement itself. An MIT study found that around 95 per cent of AI initiatives deliver no return, and the ones that fail almost always start with the clever bit instead of the boring bit.

Judgement calls about how long a tricky job will take are yours. The collecting, matching and flagging is what the system should be doing. That order is the difference between this working and it joining the list of AI projects that quietly get abandoned. It is worth checking your payment timing at the same time, since costs and cash rarely move together in trades work, and Xero’s Small Business Insights tracks how long New Zealand businesses actually wait to get paid.

Conclusion

Knowing which jobs made money is not an accounting exercise. It is the difference between pricing on evidence and pricing on hope. Every trades business already generates the data. It just sits in four places, arrives late, and requires the owner to assemble it, so it never gets assembled.

That is the real problem, and job costing is only one symptom of it. The business has no brain of its own. Every number that matters lives in a different tool, and the only thing that connects them is you, at night, when you should be finished. Implementing AI job costing for trades alongside an AI workforce that does the matching and flagging every day without being asked changes what you can see and how fast you can act on it.

Fix the visibility first. The pricing decisions get obvious after that.

Ready to see what your jobs are really returning?

If you want to work out where your margin is going and what it would take to see it in real time, book a 30-minute Discovery Call. No pitch deck. I will ask what your last five jobs looked like and tell you honestly whether this is worth doing in your business.

Frequently Asked Questions

About Octavius