All work
2026
Automate the admin, keep the personal part
For a luxury retailer: automating the follow-ups and record-keeping that slipped when new sellers came on board. The personal service stays with people.
Small business
Automation
In testing
Client
A luxury retailer
My role
Discovery, process mapping and build
Timeline
Since September 2026
Year
2026
A luxury retailer that sells on consignment gets well over a hundred enquiries a week from people who want to sell their pieces. Each one should become a conversation, then an agreement, then pieces in the store. Too many were getting lost along the way.
The challenge
Enquiries arrive through Instagram, the website and the phone. From there, everything runs on email and a shared spreadsheet kept up to date by hand. At that volume, the spreadsheet couldn’t keep up, so it was filled in once a week, after the fact.
There were six points where a seller could simply stop replying, and four of them had no follow-up at all. Nothing marked a lead as finished, so nobody could see which ones were still worth chasing.
The approach
Map it before building anything. I started with a discovery call with the team, then mapped the whole journey in five stages: enquiry, pricing, agreement, collection and stocktake. Every step on the map is marked as one of four things:
done by a person, because it needs judgement
done by hand today, but could be automated
already automated
not sure it happens at all
That map is the brief for every build.
Some things stay with people. The map also records what must not be automated: the first personal reply to a seller, setting the price, preparing the agreement, and checking each piece’s authenticity and condition. The personal service is the brand, and those are the moments that carry it.
Phase one: the enquiry stage.
A record that keeps itself up to date. Every lead has a record in Airtable, a database that works like a modern spreadsheet, and its status changes as things happen.
Follow-ups that happen on time. When the team marks a lead ready, the welcome email goes out, signed by a named person. If there’s no reply, a follow-up goes out a few days later, then a closing email. They’re all in one email thread, so the seller sees a single conversation.
Photos filed where they belong. When a seller replies with photos, the reply is matched to their record, the photos are filed and linked, and the seller gets a note to say they’ve arrived.
A person handles the exceptions. Mail that arrives outside a known thread is labelled by the team, and the automation picks it up from there.
The automations run in n8n, a workflow tool. They were designed around its monthly run limits from the start, so they stay on an affordable plan. Tests check the logic against real replies from the test runs, and the team has a written guide and a test checklist.
No AI agent, and that’s deliberate. Every step in phase one follows a fixed rule: if this happens, do that. A rule is cheaper than AI, faster, and gives the same answer every time. AI earns its place where the rules run out, when the input is messy and needs reading, not matching. The first of those here is a seller’s description of their pieces, written in their own words. That’s where an agent comes in next, drafting the details for a person to confirm.
Built with Claude Code.
The outcome
Phase one is built and in testing, in a test setup kept separate from the live mailbox and records. Once the team’s email copy is in and testing is signed off, it moves to the live inbox. The agreement stage is fully specified and comes next, then collection and stocktake.
Based in the Blue Mountains
Open to contract work
If you’re working out where AI fits in your team, I’d like to hear about it.
Email or LinkedIn is the easiest way to reach me.