Case: 11 hours a week back, for a 7-person accounting firm.

The client is a seven-person Sydney accounting firm we've worked with for two and a half years. They run a textbook small-practice mix: SME tax compliance, BAS, advisory, with a thin line of bookkeeping at the bottom of the funnel. Profitable, busy, and (like every firm we've ever scoped) losing roughly a full day a week per fee earner to admin that doesn't bill.

Here's exactly what we automated, what it cost, and what we got wrong the first time.

The workflow we mapped.

We started with a one-day process-mapping workshop. The brief: write down every step between "new lead enquires" and "engagement letter signed in Xero". The map came out at 17 steps, owned across three people. That gap (the lead landing through to a billable engagement letter) was where the hours were leaking. Specifically:

• Lead intake. Manual triage from contact form, calendar invite, follow-up reminders.
• Scoping call. Partner runs the call, no structured output, notes sit in someone's notebook.
• Quote generation. Manual fee estimate based on a spreadsheet from 2021 nobody fully trusts.
• Engagement letter. Copy-paste from a Word template, manual edits per client.
• Xero setup. Duplicate data entry from the engagement letter into the firm's practice management system.

What we built.

An n8n workflow chained into HubSpot (their existing CRM) and Xero Practice Manager (their billing system), with Claude doing the language work in the middle. The flow:

• Lead lands in HubSpot via the contact form. Workflow tags by service line (compliance / advisory / bookkeeping) using a small classifier prompt.
• Calendar invite + reminder sequence auto-drafted and sent (still partner-approved before going out; more on that below).
• Scoping call notes: partner records the call, Fathom transcript drops in, Claude summarises into a structured one-pager with services, complexity flags and estimated hours.
• Quote: Claude generates a fixed-fee quote using the firm's pricing matrix as a prompt context. Partner reviews on a single screen and ticks accept.
• Engagement letter: tick triggers a docx render from a templated source, populates client data, generates a PDF, sends via DocuSign.
• Xero handoff: when DocuSign signs the engagement, a final node creates the client record in Xero Practice Manager with the agreed scope and fee.

What we got wrong the first time.

Two things, both worth flagging because they're recurring patterns.

Auto-sent emails were the wrong default. We launched with the workflow auto-sending the calendar invite and scoping call confirmation without partner review. Two weeks in, a confused partner had to apologise to a lead who'd received an obviously templated note signed by the wrong partner. We rolled back to a one-tap-approve step. The lesson: in a high-trust profession, the time saved by skipping human review is not worth the brand cost of one wrong-tone email.

Quote precision was off. First version of the workflow used Claude to estimate fees from the call transcript alone. It anchored too heavily on what the client said they wanted, not on what the engagement actually required. We added the firm's pricing matrix as a prompt-context document and required a partner-review checkpoint. Fee accuracy on the first 20 quotes went from "underquoted on 8 of 20" to "underquoted on 1 of 20".

The numbers.

• Time saved: 11.2 hours/week measured across the team, 90 days post-launch.
• Setup cost: $7,400 (Discovery & Blueprint $1,500 + Build $5,900) plus $190/mo in software (n8n self-hosted, Claude API, DocuSign Business plan they already had).
• Payback: 6.4 weeks at the firm's loaded $95/hr rate.
• Conversion lift on lead-to-engagement: +18% in the first quarter, attributed to faster response times and a more professional onboarding flow.

Nothing about this workflow is exotic. The components are all tools the firm already paid for. The work was in mapping the right 17 steps, picking the right four to automate first, and being disciplined about where human review stays in the loop.