About the Client
The client is a Sydney custom furniture workshop serving high-end home owners. 6 staff: owner, designer, workshop carpenters, install crew. Leads come mostly from builders and interior designers; jobs run 8–12 weeks from quote to install.
Their Challenge
Everything was held together by memory + WeChat + email + Excel:
- Email flood: customers / suppliers / quotes / statements / receipts all piled in Gmail — 30+ a day, an hour to read
- Order status drift: customer asks 'where's my cabinet up to?' — owner walks to workshop, asks the carpenter, replies. 2 hours per ping.
- Quotes from gut feel: every quote recalculated from scratch, no easy way to look up similar past quotes
- Payment chasing: missed final payments, mismatched statements and invoices, no one tracking — happens multiple times a month
Why ManifoldX
The owner had quoted two consultancies: one priced $80K to redo everything in NetSuite over 6 months; the other wanted to start with a 'digital strategy assessment'. Both were too heavy, too expensive, too slow. We proposed adapting our existing Lark Base template — live in 6 weeks at under 1/10 the budget.
The Solution
A lightweight ERP centred on 'group chat as UI'. Three core components:
1. Lark Base — the data hub (13 tables)
Customers / Suppliers / Quotes-in / Quotes-out / Orders / Production tasks / Install jobs / Invoices / Payments / Statements / Modifications / Email archive / Contact log. All cross-referenced.
2. AI Copilot Bot — the interface
Mention the bot in any Lark group — 48 tool commands cover the most-used ops: query order status, update progress, draft quote, find old quotes, weekly install summary, statement reconciliation. 3,410 lines of code, runs on the client's own machine.
3. Gmail auto-sync — the input
With client authorisation, a daemon syncs new mail every 5 min via Gmail's standard IMAP interface; OpenAI extracts 11 fields (sender, email type, linked order, amount, key dates, attachment types…) and files into the right Lark Base table. ~30 emails/day, ~92% labelling accuracy; the 8% lower-confidence ones surface as confirmation cards in the group.
Tech stack
Working with us
Week 1: half-day on-site, 30 min each with owner, designer, workshop lead — drew up the existing flow. Week 2: delivered bot v1 (10 most-used commands) + 6 main tables. Weeks 3–4: live tuning, daily issues in chat. Week 5: Gmail auto-sync connected. Week 6: roles/permissions and mobile polish. Client didn't assign anyone to babysit — just a 30-min Wed evening review call each week.
I used to spend an hour every morning going through emails. Now I open Lark, read the bot's 'today's priorities' card — 5 minutes. The wildest part is it actually picks out the 3 emails out of 30 that need me personally. — Client owner (paraphrased from client interview)
Impact
- 70% faster quote turnaround: 3 days → 4 hours (bot pulls similar past quotes as a baseline)
- 12 hours/week saved for the owner: email triage + order lookup + reconciliation dropped from 15h/wk to 3h/wk
- 100% emails auto-filed: every Gmail message lands in Lark Base, no more 'where's that quote email'
- Avoided hiring an admin: owner had planned to hire an office manager next quarter; the system deferred that hire
What's next
Currently in monthly maintenance, adding a small feature each month: receipt OCR (staff drops receipts in WeChat, AI extracts and files), production scheduling board, customer follow-up reminders. This template has been abstracted into the general-purpose ManifoldX SMB Template — the next furniture client takes 2 hours to deploy with config swap only.