The AI your trade business could be running in 2026, and why it isn’t

1. It can be done. It just isn’t packaged

Go through the field service AI that gets talked about in 2026 and sort it into two piles: things nobody can do yet, and things large companies already pay for. The first pile is nearly empty. Agents that book and reschedule, help on site from a photo, reports drafted from what the tech said, prediction from sensor data: all of it is sold today, mostly inside enterprise platforms.

So the gap for a plumbing, electrical, air conditioning or fire business with a handful of vans is not “can’t be done”. It is “not packaged as a simple standard that sits beside the system you already run”. The enterprise version assumes you move onto the platform, hire someone to configure it, and have years of clean data. You have ServiceM8, Simpro or AroFlo, a busy week, and no wish to switch.

The seven sections that follow take one area each, in the same order: what is possible today, the reality check, and what it means for a 2 to 20 person trade business. One of them, predictive maintenance, we tell you to leave alone.

2. Knowing a unit will fail before the customer rings

Possible today

Predictive maintenance is real and sold today. Cumulocity sells condition monitoring and predictive maintenance built on sensor data streamed from equipment. eMaint, from Fluke Reliability, takes readings from wireless vibration sensors, raises alarms on patterns such as bearing wear or misalignment, and can open a work order in its maintenance system automatically. Salesforce describes predicting when, and why, an asset will fail. Equipment makers increasingly monitor their own units too.

The reality check

Prediction needs history: sensors on the equipment, readings over months, and a record of what actually failed. It also needs someone to pay for the sensors and the platform. That suits a plant room, a hospital or a fleet of chillers. It rarely suits a run of domestic split systems and hot water units.

What it means for a small trade business

Out of scope for us, and probably for you. If you service commercial plant, the equipment maker or the building’s own monitoring is the place to start. For a small trade business the useful version is plainer: a service due date on every unit you install, and somebody ringing before it lapses. That is a calendar, not a model.

3. An AI that runs the whole day

Possible today

Agents that do real field service work are live, at the enterprise end. Salesforce announced Agentforce for Field Service in April 2025: it books, reschedules and cancels appointments with customers over web and messaging, matches technicians by location and skills, helps a dispatcher fill a gap left by a cancellation or a no-show, walks a technician through a repair step by step, and drafts the job report at the end. Salesforce now describes the product as an agentic platform for frontline operations.

The reality check

Each of those agents does one job inside one large platform, and the dispatcher still picks from the options it offers. A set of agents that owns the whole day, from the first call to the last invoice, deciding and acting on its own, is early everywhere. Nobody should sell you that as finished.

What it means for a small trade business

You do not need the platform to get the pattern. The pieces that work are narrow: one agent, one job, one record it writes to, and a person who approves anything that matters. That shape fits beside ServiceM8 or Simpro without moving your business onto something new. We wrote up the wider idea in agents that run the work.

4. Help on site, with your hands full

Possible today

Talking the job into the record, and photos that become part of it, are shipping now. Salesforce describes a technician listening to an audio brief of the work order on the way, getting step by step help that can read a photo, and having the job report drafted from what was collected during the job. Remote expert help on a phone or tablet is also sold: Salesforce’s Visual Remote Assistant connects a technician to an expert over live video.

The reality check

Voice and photos work because the technician already carries a phone. Augmented reality on a tablet exists, and remote video is genuinely useful for a junior on a tricky fault. Headset augmented reality, the glasses in every trade show video, has low adoption: it is costly, it is awkward in a roof space, and most jobs do not need it.

What it means for a small trade business

The win is the record, not the glasses. A tech who says “replaced the element, isolator was corroded, quoted a new one” into the phone, and finds it written against the right job with the photos attached, has saved the evening paperwork. That is a small, buildable step on the system you already run.

5. The right part already in the van

Possible today

Every piece of this exists. Van stock can be tracked in most job systems. Some trade suppliers offer price and stock information to software. An agent can read the job, check the van, check the supplier and place the order.

The reality check

It falls over on data, not on the AI. Van stock is only as good as the last person who updated it, and a job that says “not cooling” does not say which part. Pre-staging, where the part is waiting at the counter or in the van before the tech arrives, is an extension of the same data: it only works once the first two are true.

What it means for a small trade business

Start with the order, not the prediction: the parts list from an accepted quote turned into a supplier order a person sends. We cover that step in material ordering for trades. Guessing the part from the fault description comes later, and should always be a suggestion.

6. When the day falls apart at 10:40

Possible today

Re-planning the day as it changes is standard in enterprise field service software. Salesforce describes in-day optimisation and agents that handle in-day conflicts such as absences and vehicle breakdowns, and other enterprise field service platforms sell the same kind of scheduling engine.

The reality check

The Australian trade platforms help you plan the day. ServiceM8 sells an Auto Routing add-on that puts a staff member’s jobs in a more efficient order. Simpro’s AI Scheduler proposes who should take new work, but it leaves blocks already on the schedule where they are, and it does not re-plan the day by itself when an emergency lands or a job cancels. That middle-of-the-day step is what is still missing at this size: noticing that a job has overrun and the plan has broken, and proposing the fix while everyone is already on the road. For most 2 to 20 person businesses today, that step is the owner on the phone.

What it means for a small trade business

This is a real gap, and the safe first step is small. An agent that watches the board, notices the overrun, and sends the owner one proposed move by text: “Move the Smith service to Dave at 2pm? Reply yes.” It suggests; you approve. It never moves a job by itself. One part of this already runs: when today’s job slips on the board, Office Voice rings the customer with the new time (see running-late calls). The re-plan itself is not built yet.

7. What the customer can see

Possible today

Live arrival tracking and online booking are common. ServiceM8 sells Track My Arrival, a map link in the on-the-way text, and online booking forms that let customers book quotes or jobs. Salesforce lets customers book, reschedule or cancel over web and messaging.

The reality check

The next idea is a shared view of the customer’s equipment: what was installed, when it was serviced, what was found last time, what is due. In the enterprise world that is called a digital twin, which just means a live copy of the asset’s record. It is not new technology. It is integration work: the install, the service history and the invoices usually sit in different places.

What it means for a small trade business

Switch on what your platform already sells before buying anything new. The shared equipment view is worth doing once every job is filed against the same customer and the same unit, which is the constraint in section 9.

8. Nobody remembers the last visit

Possible today

Searching your own job history, manuals and photos and answering in plain words is straightforward now: the AI looks up the relevant records first and answers from them, which is called retrieval. The enterprise field service assistants described earlier are built on the same idea.

The reality check

A better pre-job brief (“last time the access code changed, the dog is loose, the isolator is in the garage”) is not an AI problem. It is a data and process problem. If last visit’s notes say “done”, no model will find the dog.

What it means for a small trade business

Two things make it work: notes written against the job and the customer, and a short list of facts about each customer that a person has approved. The second is the Client Memory Build Standard, and the first is the Knowledge Base Build Standard.

9. What actually stands in the way

Read the seven reality checks again and the same four things keep coming up. None of them is the AI.

• One job, one number, all the way through. Every idea above reads the job: the brief, the parts, the re-plan, the customer’s history. If the call, the quote, the job and the invoice each carry a different reference, the AI is guessing which belong together. The foundation is one job id carried from the call to the chase, written up as the Cash Seam Build Standard. Do that first and everything else gets easier.
• It has to fit beside what you run. Nobody with a working ServiceM8, Simpro or AroFlo setup should swap platforms to get a feature. Switching is weeks of retraining, and the gaps follow you. The useful version reads from and writes back to the system your team already looks at. We compared the platforms in job management software for tradies.
• The cost has to suit the size of the business. Enterprise field service is priced and staffed for companies with a dispatch team and an IT department. A small business needs something it can switch on, watch for a month, and switch off if it does not pay its way.
• A ready-made standard beats a custom build. A custom AI project for one trade business is expensive to build and hard to check. A written standard says what the build must do, in what order, and where it must stop, with a test you can run to see whether it does. You, your bookkeeper or whoever builds it can hold the result to the page. That is why we publish them free in the standards library.

10. What we are writing next

These are planned specs, not built ones. Each takes one narrow job from the areas above and writes down the checks a build must pass. None of them is published yet, so none is linked.

• Fault Triage (planned). What the customer says is wrong, turned into a job with the likely cause, the urgency and the questions still to ask.
• Parts Ready (planned). The parts a job needs, checked against the van and the supplier, and an order a person releases.
• Job Brief (planned). The few things the tech needs to know before arriving, drawn from the last visits and the approved customer facts.
• Job Close (planned). What the tech said and photographed, written into the job record and ready to invoice.
• Exception Desk (planned). The overrun, the clash and the missing part noticed during the day and brought to the owner as one proposed fix to approve.

Want to see which leak each of these closes, and in what order to tackle them? That is the job of the field service automation roadmap.

11. What stays with a person

Everything in this paper is safer as a proposal than as an action. The AI notices the overrun, drafts the brief, lists the parts and writes the job record. A person decides to move a customer, spend money with a supplier, change a quote or tell a customer bad news.

The diagnosis stays with the tech, too. A brief that says the isolator was corroded last time is a prompt to look, not a finding. And the customer who has already been moved twice gets a call from you, not from a machine.

12. Sources and method

Specific vendor capabilities on this page were read from each vendor’s own site on 1 November 2026: Salesforce’s Agentforce for Field Service announcement and its field service product page, Cumulocity’s manufacturing use cases, eMaint’s condition monitoring page, ServiceM8’s add-ons page and help centre, and Simpro’s AI Scheduler page. Where we could not read a vendor’s own page, the claim is kept general. No vendor was contacted and none has reviewed this page.

There are no figures in this paper on purpose: no time saved, no adoption rates, no results. Products in this market change monthly. Check the vendor’s own page before you decide, and tell us if something here has gone out of date so we can correct it.