Case study · Retail · 6-store group · 45 days
Four person-days a week back to the floor.
A six-store retail group was spending half a working week stitching inventory data together by hand. A nightly reconciliation now lands in the team’s inbox before opening — flagging only the items that genuinely need a human.
Chapter 01 · The problem
Support tickets were doubling every quarter, and the floor staff were the support team.
A six-store retail group in Queensland was running three platforms behind the counter — Shopify for online, a POS for in-store, and two shared inboxes for customer support. The same question arrived through every channel: where is my order, can I swap a size, did the warehouse pick the right colour. Floor staff were splitting the day between serving customers and answering tickets. The support volume had roughly doubled across the prior twelve months. The Founder had hired a part-time helper to keep up. The helper was already at capacity.
~700 tickets a week
Chapter 02 · The approach
We did not replace the inbox. We answered the easy 60% of it.
The fix was not a chatbot. It was a pipeline that read every incoming ticket, matched it against the order in Shopify and the live stock in the POS, and drafted a reply the support agent could send in one click — or rewrite if the situation needed a human eye. The agent stayed in the loop on every reply. The model never auto-sent. The 60% of the inbox that was status checks, swap requests, and stock questions cleared in minutes, not hours.
What we built
A ticket-triage pipeline reading from a shared inbox, joined to Shopify orders and POS stock through n8n, with a custom reconciliation step that catches stock drift between the two. Anthropic Claude drafts the reply against the matched context. The Hermes orchestrator handles retries, queues the agent review, and writes the chosen reply back to the inbox.
Build time · 14 days · pilot phase · followed by 5 weeks of iteration with the Founder and the lead support agent.
Chapter 03 · The outcome
Three numbers that gave the floor staff their week back.
- -0%
support volume handled by floor staff — measured against the baseline ticket queue from the prior quarter.
- $40K
AUD a year in support cost saved — based on the part-time helper hours the Founder no longer needed to schedule.
- +0%
faster catalogue updates across the six stores — measured against the manual reconciliation cycle that ran every Friday.
Numbers verified by the Founder against Shopify reports and the POS export. Anonymised under our standard case-study disclosure: vertical and outcome are real; retailer name, store locations, and POS vendor are not.
Chapter 04 · What we learned
The drafting was easy. The reconciliation was the hard part.
The reply pipeline itself took 14 days. The five weeks that followed were spent on stock — specifically, the times the POS said one thing and Shopify said another. A draft reply that quoted the wrong stock figure burned more trust than no reply at all. We added a single reconciliation pass before the model ever saw the ticket: if the two systems disagreed by more than a unit, the ticket routed to a human with both numbers shown side by side.
The lesson was about the order of operations. We assumed the model would be the hard part; in practice, the platforms beneath it were. The first week of pilot caught more stock-drift bugs than the prior six months of manual reconciliation. The Founder kept that reconciliation step running long after the support pipeline went live.
Settled handoff rate · ~12 reconciliation rules
Chapter 05 · In the client’s words
My team is on the floor again. The inbox is still there, and we still read it — but it is not eating the day. That is the bit I did not know we could buy back.
Founder · 6-store retail group · QLD (anonymised at the client’s request)
Curious whether your store has the same gap?
A free 45-minute audit. We look at your support volume, your POS, and where the floor team is losing the day. You leave with a one-page memo, whether we’d be a fit or not.