Jewelry and Ecommerce
An AI back office for a downtown LA jewelry auction house.
Jewelry District Auctions runs live jewelry auctions on eBay Live and TikTok Shop from downtown Los Angeles and is one of the top jewelry sellers on eBay Live. Henrik, the owner, needed the work around the auctions handled: buyer emails, custom pieces for the loose stones he sells, recap clips from every show, and the books. I built it in about nine days, deposit to live, and trained his team at the office.
Jewelry Auction House, Downtown Los Angeles, CA
9 days
From Deposit to Live, With an Office Training Visit
Timeline
About nine days from deposit to live, with an office training visit, tuned since
Investment
Fixed price build plus a monthly retainer
The problem
The auctions ran great. Everything around them was done by hand.
A live auction business makes its money during the show and spends the rest of the day cleaning up after it. Every show creates buyer emails about shipping, payment, and sizing. Some winners buy a loose stone and want it turned into a ring or a pendant, which means a chain of messages between the buyer, the team, and the jeweler. Every show is also hours of footage with great moments inside it that nobody has time to find. On the money side, eBay pays out in batches, so the bookkeeper had to match each payout back to its sale by hand to know what had landed, and a payout that got returned could sit unnoticed. Henrik has a team, and too much of its day went to repeated work instead of the auctions.
What was going wrong
- ✕Hundreds of buyer emails, answered one at a time by whoever was free
- ✕Buyers who won a loose stone and wanted it set, handled through back and forth messages with the jeweler
- ✕Hours of eBay Live footage from every show with the best moments buried inside
- ✕An inventory sheet kept by hand that lagged behind what had actually sold
- ✕eBay payouts matched to sales by hand, with no quick way to spot one that got returned
How I built it
Automate the repeated work. Keep a person on anything a customer sees.
Phase 1: Inbox autopilot
Connected the shop's inbox and built a drafting system that reads each new buyer email, checks it against the shop's written policies, and saves a reply as a draft. Internal mail and automated notices get labeled and skipped. A person reads, edits, and sends.
Phase 2: Loose stone to custom jewelry concierge
When a buyer wins a loose stone, the team sends a link. The buyer picks a setting, the system generates three AI design concepts for that stone, and the request goes to the jeweler with everything attached.
Phase 3: Auction recap clips
A scheduled job pulls each eBay Live show's replay, finds the biggest sales, and cuts a clip of the last 30 seconds of bidding plus eBay's own sold screen, biggest sale first, ready to post.
Phase 4: Reports, inventory, books, and training
A daily report, an inventory import with alerts, a payout reconciliation sheet for the bookkeeper, and a team ChatGPT workspace. Training day at the office, then tuning from real traffic.
Under the hood
What is actually under the hood.
Results
What changed for the business.
The team opens one hub. Buyer emails arrive with a reply already drafted, so answering is reading and clicking instead of writing from scratch. A won stone turns into three design concepts and a jeweler order through one link, with no setup messages. Every show produces its own recap clips. The bookkeeper gets a sheet where every eBay sale is already matched to its payout, and a payout that got returned is flagged instead of found later. The whole system went from deposit to live in about nine days, and I trained the team at their office.
9 days
Deposit to Live
Build started mid October 2026, live with an office training visit about nine days later
3
Design Concepts Per Stone
Generated in minutes once the buyer picks a setting, then routed to the jeweler
30 sec
Of Bidding Per Clip
Plus eBay's own sold screen, biggest sale first, cut automatically from every show
Every
eBay Sale Matched to Its Payout
On the bookkeeper's sheet through the eBay Finances API, with returned payouts flagged
What this proves
- 1Automate the parts that repeat: drafting replies, cutting clips, matching payouts. Leave the decisions to the people who know the business
- 2Keep approval on anything a customer sees. A draft a person sends in ten seconds beats an automatic reply nobody read
- 3A live auction business has its best marketing sitting in its own replays, and cutting it automatically is the difference between posting daily and posting never
- 4Measure the system by hours handed back to the team, not by how clever it is
What I learned
- The inbox first drafted replies to shipping notices and internal mail. The fix was teaching it to label those and never draft, and that is now the first rule it checks
- eBay's logged time for a sale can be minutes off. The clip tool now finds the sold screen on the replay itself instead of trusting the log
- The bookkeeper's own spreadsheet layout was the right spec. I kept the existing columns and added the checks beside them instead of handing over a new tool
The services behind these numbers
More case studies
More results, with the numbers attached.
Find out what your phone is costing you.
Fifteen minutes. I'll show you where calls are leaking, what it costs, and what I'd build. Free either way.
No contracts, no pressure. Just a real conversation about your business.