The six ways to put AI into a trade business

A lot of noise has been made about AI and you're probably using it in a number of ways. But you're probably missing out on some of the real value and time savings it can deliver. I've compared six ways of using AI in trade businesses and ranked them.

A lot of content that exists on this subject is little more than a shopping list. Ten tools, a price next to each one, and a claim that you can set it all up in an afternoon. Assessing tools is interesting, but it's important to understand the plumbing too.

I run an electrical contracting business in Auckland. I also build software. I've spent the last two years putting AI into my own business, and I've sat across the table from owners running setups far more automated than mine. What follows is the map I wish someone had given me at the start.

There are six different setups for working with AI. They range in value delivered, time saved and setup costs. Some of them involve a lot of manual handling of data, and some take care of that for you. These can be thought of as rungs on a ladder, as they get progressively more powerful and require more setup and know-how.

The days of comparing a business system to physical paper are long gone. What we're comparing against is a typical job management app on its own, and how these tools can work with or inside job management apps.

Rung 1. Chat tools with manual transfer

ChatGPT or Claude in a browser tab. You paste in the messy site notes, it drafts the scope of works. You paste in the angry email, it drafts the calm reply. USD $20 a month and it works on day one.

This is genuinely useful and is a great place to start. It also gets better over time as your projects build up memory and context about your business. To get started you make a project inside Claude or ChatGPT, tell it to ask you questions about your business and create an MD file for its own reference, and upload things like your pricing to it.

But no matter how much these chats learn about you, it still remains a very manual process. You are the integration between it and other apps. You copy in, you copy out, it advises and you execute, and nothing that happened exists in your job system. The quote it helped you write isn't linked to the job. The email isn't on the customer record. If you stop pasting, everything stops.

Almost every "AI for tradies" article is describing this rung and calling it a stack.

Rung 2. AI features inside your job management app

Whatever your vendor has shipped. Simpro's agents brief and summarise individual jobs. ServiceM8 does AI job allocation and adjusts routes in real time. Tradify suggests quote lines from your job history.

The copying disappears and there's nothing to maintain. The catch is that you get exactly what the vendor imagined and nothing more, and most of them have shipped narrow features scoped to a single job. Useful, but nobody's app is doing powerful admin work yet.

Cost: usually included in your subscription. Effort: zero. Ceiling: low.

Rung 3. Bolt-on AI services

The most marketed AI solution for trade businesses has to be the AI receptionist. It's also the one I'm the least keen on. I figure people calling to inquire about a service want to talk to a person. I've met customers who dealt with an AI answer service who said they don't want to deal with that company because of it.

That said, if you genuinely can't answer calls then it's definitely better to have an AI talk to them. The Australian market runs roughly $99 to $699 a month with setup fees from zero to about a thousand dollars, and most trades-focused options cluster around $200 to $300. The better ones push the enquiry into ServiceM8 or Simpro as a job. There are also transcription tools that turn site visits into notes, and voice tools you ring to say "invoice the Smith job" and it happens in your app.

For a business bleeding missed calls, this pays for itself with one recovered job a month.

Two things to watch. First, vendor sprawl. Each service holds a partial view of your customer, and the outputs mostly land in your inbox rather than on the job.

Second, know what you're buying. A decent share of the AI receptionist companies selling to tradies are running GoHighLevel underneath, white-labelled with their own branding. That is not a criticism. GoHighLevel is very capable software and you can do genuinely clever things with it. And the companies reselling it usually earn their margin: they configure it properly, write the call scripts, map it to your job system, and stay on the phone when something breaks. Most owners would never get the same result buying a licence themselves, because the work is in the setup, not the software. I use GoHighLevel at Legacy Electrical and I'm glad I have someone to ring when things break.

The thing to be clear-eyed about is the architecture, not the vendor. You've added a second CRM that handles the customer relationship, sitting in front of the job system that owns the work, connected by a one-way push covering whatever fields got mapped. You now have two systems, each holding half the record. Fine trade if you're missing calls today. Just know it's the trade you made.

Rung 4. Orchestration

This is the type that AI consultants are running around setting up for everyone. It's where a tool like n8n integrates different apps together in workflows. A call comes in, the audio is transcribed, the transcript processed and a quote drafted. One action in the business has four apps firing off on autopilot.

A lot can be achieved with these systems. The downside is they are complicated to build and complicated to adjust. They also suffer from not having a single source of truth. The call transcript sits in one app, the quote in another, emails and text messages in yet more apps.

Setting these systems up will likely take all of your free time and you'd need to be technically minded. Plenty of consultants can build these for you, but be ready to call them back any time you want to make a change.

The running costs aren't bad. You're likely already paying for all the apps you're linking up. n8n has a self-hosted community edition which is free with unlimited runs. You pay for the box it runs on, which can be a $10 a month server or a Mac mini in the office. Zapier and Make are the hosted alternatives, but they bill per step, so the always-on polling that makes automation useful gets expensive. n8n bills a whole workflow run as one execution, and self-hosted it bills nothing.

The real constraint isn't the orchestration tool. It's your job management app's API. Simpro has one of the deepest APIs in the trades and you can self-provision credentials without talking to anyone. ServiceM8 has a documented API, real webhooks and an SDK. Tradify has no public API and no webhooks, so there is no supported way to automate it at all. If you're on a closed platform, this rung doesn't exist for you, so you may want to explore other options.

The cost of this rung is a person. Someone has to build the workflows, and someone has to fix them when the platform ships a change and half of them break. If that person is you, you've just added a second unpaid job. If you pay an agency, you're renting the capability and you can't modify what they built. I know an electrical business running around a hundred workflows this way. It's genuinely impressive, and it's also a system exactly one person on earth understands, and it eats up a serious chunk of his time.

Rung 5. Your own data layer

This one is an absolute game changer. What's different here is the addition of a central hub for all your business data to live in. Instead of a complex web of apps linking to one another, data is synced into a central database that can be queried by AI. Because it's all in one place, searches are fast and reliable, and summaries of the data can also be maintained, giving agents a quick source of accurate information.

This is done with a hosted Postgres database such as Supabase. Supabase is particularly good for this because it provides a lot of functions too, such as login tools, file storage, an auto-generated API and serverless code functions in one place. The free tier covers a lot of use cases. The paid tier starts at only $25 a month. Neon is the database-only alternative, typically paired with Vercel, and it's what AI coding tools like Replit and v0 provision by default. Both paths work. Supabase simply bundles more of it upfront.

This is how actual software companies build apps, which is exactly why it removes the ceiling. At rung 4 you can only touch what your platform's API exposes. Here you own the tables. Any question you can write is askable. Any workflow you can imagine is buildable. My own business runs on this: every email swept into one database on a schedule, quotes and compliance documents generated from job data, research and knowledge stored where the AI doing the work can query it.

Now the honest costs, because they're substantial.

You are designing, building and maintaining a system. Supabase gives you the auth mechanism and the backups, but the security model is yours to author, and a wrong access policy is your data breach, with no vendor to blame.

The build itself now runs through AI coding tools, and those cost real money. Anything substantial means hundreds to a few thousand dollars in AI usage, and weeks of your attention, which is worth more than the tokens. The "build your app in a weekend" videos are demos. Production software that your business depends on takes months of iteration.

And it's still one person. The system is exactly as durable as your ongoing interest in it. Your business inherits whatever you built, documented however you documented it.

If you code, or you've become the kind of owner who can direct AI coding tools well, this rung is genuinely open to you in a way it wasn't three years ago. It's super cool, but go into it knowing it's a permanent commitment dressed as a project.

Rung 6. An app with AI native to the workflow

The top rung is when the intelligence lives inside the system of record itself. Nothing to wire, nothing to sync, no seam to pay tax at. The AI acts at the point of use: while the quote is being written, while the sparky is on site, while the enquiry is arriving, rather than reacting after the fact through an API.

The second half of this rung is agent access: giving your Claude or ChatGPT the ability to answer from your business data and carry out actions in the app. Pull up the job, draft the invoice, book the visit. The open standard for this is MCP. When the app releases a new feature, the agent gains it too, with no integration work. Compare that with rung 4, where every new capability is a workflow someone builds and then maintains forever.

The honest problem with rung 6 is that the options barely exist. ServiceM8 is the only established trade platform with a first-party MCP server, shipped in late 2025, and it's a real credit to them. Simpro has no MCP server, and the middleware companies selling "Simpro MCP" or "ServiceM8 MCP" endpoints are wrappers over the ordinary API. That's rung 4 in a rung 6 costume. A first-party agent surface over the platform's own data is a different thing, and most vendors haven't built one.

This is the gap I'm building Service Ninja into. It runs my electrical business today: quoting, invoicing, scheduling, timesheets, compliance certificates and the field app, with the AI inside the workflow rather than bolted around it, and a first-party MCP server so an agent can operate the whole thing. I'm only describing what's live, because that's the standard I hold the rest of this essay to.

How to choose

The ladder isn't a maturity model. You shouldn't climb it for its own sake.

If you're doing nothing, start at rung 1 today and add a receptionist at rung 3 if you're missing calls. That combination costs a few hundred a month, takes a week, and recovers real money.

If you're technical, or you have someone who is, rung 4 is legitimate, provided your platform has a real API. Go in knowing you're taking on software ownership without a support contract.

Rung 5 is for a very specific person: the owner who codes or is technically minded, has months to spend, and accepts that they're building a permanent second job. It's more possible than it's ever been and it's still rarely the right call.

For everyone else, the move is to watch rung 6 hard, because it's where this ends up. The question to put to your software vendor is simple: where does your AI actually run, and can an agent operate your platform through a first-party interface. The answers will sort the field quickly.

The pattern under all of it: rungs 1 through 5 are attempts to add intelligence around a system of record that wasn't built for it, and every seam charges rent forever. The platforms that build the intelligence in will make the rest of the ladder unnecessary.

At Service Ninja we believe that rather than every business building their own internal systems, trade service businesses are best served by those systems being built into a robust software solution. And building that solution is our goal.

Alex is a registered electrical inspector and director of Legacy Electrical in Auckland. He's a co-founder of Service Ninja, which Legacy runs on every day.