Every second-hand club came with a haggle.

Every offer was a conversation.

Good Lies buys, refurbishes and resells pre-owned golf clubs from its Melbourne base and through its own eBay store. Buying well is the business. Offer too much and the margin's gone; offer too little and the seller walks.

Every club that came in needed someone who knew the market to identify it, judge its condition, look up what it sells for and settle on a number. Then, often, defend that number. The brief had two goals: win more sellers, and cut the time spent haggling.

• Pricing ran through one person's time and judgement.
• Partner pro shops priced trade-ins from a monthly price-book document.
• Every quote is a lead, and nothing was capturing the sellers who asked and walked away.

AI identifies the club. Rules set the price.

The obvious build was to show an AI some photos and ask what the club is worth. Quick to demo, impossible to defend. Ask twice and you can get two numbers, and nobody can say why.

So the work was split. Claude's vision model does what it's good at: identifying the club from the photos and checking the seller's own condition grade against them. The price comes from a deterministic engine, built on live eBay market data blended with Good Lies' own price book. Same club, same condition, same offer, every time.

When the AI isn't sure, or a club falls outside the rules, the quote goes to a person instead of a guess. The seller is never stuck, and the business never pays for a wrong answer delivered with confidence.

The model never outputs the price. This is what makes the number defensible. From the proposal, July 2026

The build

• A five-step seller flow. Club type, brand and model from a searchable catalogue, guided photos, a seven-grade condition check, then an emailed offer.
• Claude vision on every photo set, identifying the club and flagging when the photos and the seller's grade disagree. Any failure routes to review, so no seller hits a dead end.
• A deterministic pricing engine. eBay market comparables, filtered for outliers and blended with Good Lies' price book, then condition, margin guards, floors and rounding.
• Every offer reproducible. Pricing rules are versioned and each quote keeps a snapshot, so any offer can be rebuilt exactly if a seller queries it.
• Cash or store credit for sellers, and a single trade figure for partner pro shops.
• A pro-shop portal for counter quotes, price checks and monthly statements, replacing the price-book document.
• An admin workbench. Review queue, inspection log, a pricing-rules editor that previews a change on a real club, and price-book upload.
• Market evidence captured, not scraped. Staff screenshot Facebook and Gumtree listings, and the AI fills in the fields for them to check.
• Emailed quotes, expiry reminders and one-tap re-quotes, backed by more than 2,600 automated tests.

An offer in minutes, and a reason behind every number.

The engine now gives an automatic offer for 2,113 of the 2,114 clubs in the catalogue. Anything it can't price confidently still reaches a person, with the reason attached.

Before launch, a test day ran 44 real clubs through the pro-shop counter flow. 32 priced automatically. The rules were tuned on what that day showed, and on the tuned rules 42 of the 44 price automatically.

Partner pro shops go live in October 2026, with the public seller flow following, Melbourne first.