The Problem
Jake runs a small electrical contracting business in Brisbane's north side. Two licensed electricians — him and one full-time sparky — plus a part-time admin assistant who works school hours three days a week.
The work itself was good. Consistent residential jobs, a handful of small commercial clients, a solid reputation in the area. The problem was everything around the work.
Every quote took 35–45 minutes to write up. He was doing 15–20 per week. At the low end, that's 9 hours a week on quotes alone — most of it after dinner, once the kids were in bed. Job notes for ServiceM8 were getting done in the van, rushed, and often incomplete. Invoice follow-up for overdue payments was falling through the cracks because nobody had the time to chase consistently.
His admin assistant was doing what she could in her hours, but she wasn't available for the evening admin blitz. And Jake's wife had made it clear that 10pm laptop sessions were becoming a problem.
He came to us not looking for a full technology overhaul. He wanted to stop working at night.
What We Found
Before building anything, we spent an hour going through a typical week with Jake. Not the ideal week — the actual one.
The quoting process was the clearest target. Every quote followed the same structure: site details, scope of work, materials estimate, labour hours, total, terms. Jake was writing each one from scratch in Word, copying in his standard terms at the bottom, then attaching it to an email he also wrote from scratch. He had no template. Every single quote was reinvented.
Job notes were the second problem. ServiceM8 has a notes field Jake was supposed to fill in after each job — what was done, what was found, any follow-up needed. In practice he'd do it from memory at the end of the day, often forgetting details from the morning's jobs. His admin assistant then had to call him to clarify before she could close jobs off and send final invoices.
Invoice follow-up was being done ad hoc. His admin assistant would check Xero on Tuesdays and manually email anyone overdue. Some weeks it didn't happen. He had $14,000 sitting in outstanding invoices over 30 days when we first looked.
Three distinct problems. Different solutions for each.
What We Built
1. Voice-to-Quote in Under 10 Minutes
The quote workflow now works like this:
After a site visit or job estimate, Jake records a 2–3 minute voice memo while he's still in the van. He describes what he found, what needs doing, the materials he'll need, and roughly how long it'll take. Specific, in his own language.
That memo gets transcribed automatically by Whisper (running via an n8n workflow triggered by a shared iPhone shortcut). The transcript goes to Claude with a prompt that structures it into Jake's quote format — site details, scope, itemised materials, labour, total, and his standard terms pulled from a saved document.
The result lands in a Google Doc draft, shared with his admin assistant. She checks it, adjusts the pricing if needed (she knows his rates better than anyone), converts it to PDF, and sends it from his email address.
Total time on Jake's end: 3 minutes in the van. His admin assistant's time: 5–7 minutes to review and send.
Before: 40 minutes per quote, mostly Jake's evening time. After: 3 minutes from Jake, 6 minutes from his assistant. Evening closed.
2. Job Notes by Voice
Same approach, different destination. After each job, Jake records a 60-second voice note: what was done, any issues found, whether follow-up is needed. The transcript gets formatted by Claude into a ServiceM8-compatible job note and dropped into a shared doc queue for his admin assistant to paste in during her morning.
She's stopped calling him for clarifications. Jobs are closing faster. Final invoices are going out same-day instead of 3–4 days later.
3. Automated Invoice Follow-Up
This one was the most impactful for cash flow.
We built an n8n workflow that connects to Xero and runs every Monday morning. It pulls all invoices that are 7+ days overdue. For invoices 7–21 days overdue, it drafts a polite follow-up email via Claude — personalised with the client name, invoice number, amount, and a payment link. For invoices 22–35 days overdue, the tone shifts slightly and it flags them to his admin assistant for a phone call.
His admin assistant reviews the drafted emails each Monday, hits send on the ones that look right, and calls the flagged ones. She's not writing anything from scratch — she's approving and sending.
The $14,000 in outstanding invoices dropped to $3,200 within six weeks of this going live. Not all because of the automation — some of those were just slow payers who needed one nudge. But the consistency of the follow-up is what made it happen.
The Numbers
Across 15–20 quotes per week, the quote workflow alone saves Jake roughly 8–9 hours. Job notes save another hour. The invoice follow-up workflow saves his admin assistant 90 minutes every week and has directly improved his cash position.
Total estimated time saving: 10–12 hours per week. Almost all of it was evening and weekend time.
He now finishes most workdays before 6pm. His admin assistant is more effective in her hours. And the $14k in aged receivables is largely cleared.
What We Didn't Change
Jake handles all client calls himself. He estimates every job on site. He makes every decision about scope, pricing exceptions, and whether to take on a job.
None of that changed. The AI is handling the document production that surrounds the work — not the work itself or the judgment behind it.
His clients haven't noticed a difference in how he communicates. If anything, his quotes are more consistent and better formatted than they were before, which his admin assistant thinks has helped his close rate on larger jobs.
What This Cost to Build
The tools involved:
- Whisper (via API) for transcription — minimal cost at Jake's volume
- Claude API for drafting — roughly $15–20/month at current usage
- n8n (self-hosted on a cheap VPS) — $8/month server cost
- Xero and ServiceM8 — already paying for both
Setup time: approximately 12 hours of our time across two sessions. We documented everything and trained his admin assistant on the review process in a 45-minute call.
Total ongoing cost: under $30/month in AI and hosting. Setup fee paid back in the first week of time saved.
The One Thing That Almost Stopped It Working
Jake's voice memos. For the first two weeks, he was recording them in the office at the end of the day from memory — not in the van right after the site visit. The transcripts were vague, Claude's drafts were generic, and his admin assistant was having to fill in gaps.
We pushed him to record in the van, immediately after every visit or job. That change — which sounds small — is what makes the whole thing work. Specific inputs produce specific outputs. Memory-based inputs produce generic outputs.
Once that habit clicked, the system worked as designed.
If you're in trades and this sounds familiar, the tools and workflows aren't the hard part. The habit change is. But it's a 3-minute habit, not a 45-minute one. Most people find that easy to sustain.