Trades and field service businesses get pitched AI constantly and almost none of it survives contact with a job site. The useful question is not whether to use open source models, it is which parts of a trades business have enough repetitive text and data work to be worth automating at all.
Where the actual time goes
In most Australian trades businesses the office work, not the field work, is what limits growth. Quoting, scheduling, chasing payment, writing up job reports and answering the same twenty customer questions consume the hours that would otherwise go into more jobs.
Quoting from a site photo, a rough note and a price list
Turning a technician's voice note into a job report a client can read
Scheduling and rescheduling around cancellations and weather
Chasing invoices and following up quotes that went quiet
Every one of those is text and data handling. None of them requires a model to be brilliant, which is precisely why the cheaper end of the market is worth a look here.
Where open source fits
The case for an open-weight model in a trades business is almost never capability. It is cost at volume and, occasionally, working offline. If you are processing hundreds of job reports a month, per-token pricing starts to matter in a way it does not at ten a week.
The case against is that you now own infrastructure. A plumbing business with eight vans does not want a GPU server and the person who maintains it, and the honest advice is that a hosted model is the right answer until the volume genuinely justifies otherwise.
The realistic first project
Job reports. A technician records a voice note at the end of a job, and the system produces a client-ready written summary plus the internal notes. It removes the worst-loved admin task in the business and nothing about it is risky.
Quoting is the more valuable second step but a harder first one, because it needs your price list, your margins and your judgement about site conditions. Get the reporting right, build trust in the output, then move upstream.
What it is worth
A business with eight field staff, each losing forty minutes a day to write-ups and admin, is losing roughly twenty-six hours a week. At a charge-out rate of $110 an hour, even recovering half of that is over $70,000 a year in billable capacity that already exists.
The saving is not in wages, it is in throughput. The same crew completes more jobs without working longer, which is the only lever most trades businesses have short of hiring in a tight labour market.
Connectivity is the constraint nobody mentions
Field work happens in basements, on rural properties and inside metal sheds. Any workflow that assumes a live connection at the moment of capture will fail regularly and be abandoned within a month.
Design for capture offline and process later. The technician records or photographs whatever the situation allows, and the processing happens when the device reconnects. This single design decision separates the tools that get used from the ones that do not.
Keeping it honest with customers
A job report that reads as though a machine wrote it will be noticed. Feed the system your own past reports as examples so the output sounds like your business, and keep a person glancing over anything that goes to a client.
There is also a straightforward accuracy point: a summary that overstates what was done or promises a warranty position you did not intend is a commercial problem, not a writing one. Review is cheap; a disputed report is not.
Privacy and site photos
Site photos routinely capture more than the job: household contents, other people, occasionally documents. Under the Privacy Act and the Australian Privacy Principles that is personal information you are now collecting and storing.
Nothing about this is prohibitive, but decide early where images are stored, how long they are kept and who can see them. It is far easier to set that up at the start than to unpick after four years of accumulated photos.
What not to conclude
None of this replaces a tradesperson, and any pitch that suggests otherwise misunderstands the work. What it replaces is the second unpaid shift of paperwork that happens after the last job of the day.
Nor should a small operation start with self-hosting. Open source matters here as a cost option at scale, not as a starting point. Prove the workflow on something hosted and cheap, and revisit the infrastructure question only if the volume ever justifies it.
If your office admin is the thing capping how many jobs you can take, book a short call and we will look at which part is genuinely mechanical.
