Case study · Live engagement
Goldenwood proved its own demand. In September the system to catch it went live, and the first order came from an AI answer.
Published · Updated
- Client
- Goldenwood Furniture & Interiors
- Industry
- Bespoke solid-timber furniture & joinery, since 1988
- Location
- Sydney, Australia
- Engagement
- Live engagement
- Built so far
- Storefront and catalogue, team portal and CRM, email on Lark, Google Ads, a live dashboard, Maps and social
- Revenue
~6×
In April 2026, with the owner running his own Google Ads to the first site we built for him, online revenue that month was about 6× the seven months before combined, by the owner's own count. This was before the current system existed.
- Leads
1storder from an AI answer
The first order through the new system came from a ChatGPT answer, not an ad.
- Time
~200estimated hours a month back
About one full-time person's month, spread across the office and the workshop: staff see their own jobs, pages open instantly, and posting, quotes, invoices and follow-ups run on their own.
Estimate: task time × monthly volume × team size
The problem
Goldenwood's owner ran their own test. In April 2026, with the owner running his own Google Ads to the first site we built for him, online revenue that month was about 6× the seven months before combined, by the owner's own count. This was before the current system existed. That result belongs to the owner's ads and that site. It proved the demand was real, and what the business could not yet do was catch, route and measure it.
Goldenwood is a maker whose work brings customers back years later for the next piece. Enquiries arrived through the website, Instagram and the phone, and often waited overnight for an after-hours reply. Customers lived in one system, orders in another and contacts in a third, and nobody had ever joined them. Marketing money went out and nobody could say what came back.
The first order through the new system came from a ChatGPT answer, not an ad. So we now track how often AI assistants name Goldenwood when buyers ask where to get furniture made. On 25 Sep 2026, ChatGPT named it in 2 of the 12 buying questions we check.
What we built
The foundation that turns enquiries into tracked, collected revenue. Everything below is built and running on our own infrastructure today, wired into the tools they already use.
Portal screens marked “real interface” are the running system, captured on a copy that holds only made-up records. The chat, dashboard and board are drawn.
01Marketing and sales
More of the right enquiries, and proof of which channel brought each sale.
AI search (GEO): buyers already arrive from ChatGPT answers.
- Websites that turn visits into enquiries
- Google Ads for residential and commercial buyers
- SEO, and visibility in AI search (GEO)
- Social media, posted without staff time
- Your Google Business Profile, run as a shop front
- Knowing which channel brings each sale
Storefront & catalogue
We rebuilt the storefront on Next.js and moved the real domain onto it on 10 September. Every address from the old site still opens a working page, so links and search rankings carried over. All 146 catalogue pieces carry the maker's own timbers, categories and descriptions, and each restyled photo was checked side by side against the maker's original. Customers can attach a photo of the piece they want, because that is how most of them ask.
Google Ads
Ads came back on 12 September, rebuilt to count only real enquiries as results. Trade buyers, meaning builders, developers and office fit-outs, got their own campaign on 23 September, so the homeowner test stays clean. Each ad enquiry can be traced from the search that produced it to its record in the CRM.
Maps & social
We rebuilt the Google Business Profile as a shop front, with products, photos and replies to reviews. Instagram and Facebook have posted every morning since 14 September, and LinkedIn since 21 September, with no staff time.
02Operations
Every enquiry caught, quoted and billed, with nothing typed twice.
- A CRM and a portal your team works in every day
- Quoting and staged invoicing, in step with your accounting system
- One clean customer and product record
- Email and team chat in one place
Team portal & CRM
Customer records from three disconnected systems now live in one CRM, with a portal the team works in. Deduplicating 200 raw records surfaced 174 unique customers, including 18 high-value repeat trade accounts that had gone quiet. Enquiries, quotes, orders and workshop tasks sit in one chain, and confirming an order drafts its Xero invoice. Every change is saved, read back and logged with who made it.
Email & chat on Lark
The company's email moved from Microsoft 365 to Lark on 10 September, next to the chat the owner already used. A new website enquiry is posted into the owner's Lark chat as it lands, with a clock on the reply. Staff can ask the CRM a question in Lark and get the answer in the chat.
03Management
How the business is doing, at a glance, without chasing anyone for it.
- An owner's dashboard
- AI agents that answer your team's questions from your own data
- Alerts on what is late or coming due
- Monthly reporting
Attribution & dashboard
Each website enquiry is saved with the signals it arrived with: the ad click, the campaign tag, the Business Profile link or the AI assistant that sent it. One private dashboard gives the owner a page per decision: money, ads, search and AI, leads, website traffic, social and the attribution chain.
04How we build
New work ships without breaking what already runs.
- Every change checked before it ships
- Testing that never touches live data
- Built into the tools you already use
- Your customer record stays yours
How the work is built
Every change is built on its own card and checked by a separate verifier before it ships. In the first six days on that board, about 250 build, check and research cards closed across the storefront, portal, dashboard, ads, search and social.
The storefront we built
We didn't just advise. We designed and built Goldenwood's website, the storefront the demand was landing on, and run it on our own infrastructure.
View the live site
What the revenue shows so far, and what is still being measured.
In April 2026, with the owner running his own Google Ads to the first site we built for him, online revenue that month was about 6× the seven months before combined, by the owner's own count. This was before the current system existed.
In the first month of the new website, CRM and contract (September 2026), Goldenwood booked about 10× the sales of the same month last year.
What feeds the number
Every website enquiry is saved with its source, ad enquiries are traced from the click to the CRM record, and stage invoices tie each order to Xero.
Before and after
| What changed | Before | Now in place |
|---|---|---|
| Storefront | Old Framer site | Next.js storefront on the real domain; every old link still works |
| Customer records | 200 raw rows, three systems | 174 clean customers, one CRM and portal |
| Microsoft 365 | Lark, beside the team chat and enquiry alerts | |
| Google Ads | Off since June | Live since 12 September, homeowner and trade, judged on enquiries |
| Attribution | None | Each enquiry saved with its source signals; the first order traced to an AI answer |
| Catalogue | Not checked piece by piece | 146 pieces checked against the maker's own record |
Everything above runs on our own infrastructure. What the system produces is measured every month against a day-one baseline: new revenue, new leads and hours saved. The first conversation and onboarding are free. Once we start building, you pay a monthly fee, plus a share of the revenue the system can measure, or one flat monthly fee if you prefer.
See if Margin fits your business
We take a small number of engagements and qualify every one. Two minutes to apply, or get in touch and we reply personally within two business days.