Home / Cases / Used-Car Sourcing AI Assistant
Used Car Dealer · Sydney

Used-Car Sourcing AI Assistant: 3 hrs of daily market research, distilled to a 5-min brief

A Sydney used-car dealer who used to spend 3 hours a day reviewing public market listings. We built an AI assistant that lets him batch-import the listings he reviews, runs valuation + risk analysis, and pushes a bilingual daily brief to his phone.

IndustryUsed car wholesale / retail Size3-person dealer Timeline2 weeks StatusLive
3h→5min
Daily screening
~60h/mo
Owner hrs saved
2×+
Monthly hit rate
$2.3K
Project total cost

About the Client

A Sydney mid-size used-car dealer focused on refurbished resales. 3 staff: owner (sourcing + sales), mechanic, detailer. Profit hinges on sourcing — whoever spots 'cars 15% below market' first wins.

Their Challenge

Why ManifoldX

The owner had reviewed dealer-tier tools — all priced per dealership per month at $300+/mo. Our quote was a one-time $2,300, fully owned, no recurring fees, no third-party platform accounts required.

The Solution

1. Desktop AI assistant — data organisation

A local desktop tool running on the owner's own machine. He browses public used-car listings as usual; whenever he sees something interesting, he drops the link / screenshot / text into the assistant's 'today's candidates' zone. The tool auto-structures into 26 fields (price, km, year, spec, location, seller type, photo count, description length, etc.) into a local SQLite database — data never leaves his machine.

2. AI valuation engine

OpenAI does three things per candidate: (1) compare against the model's historical sold prices and score 'market deviation'; (2) detect 'rare specs' (rarity of year+km+colour+transmission combos); (3) flag risk (does the description contain 'accident / repaired / private import' etc.).

3. Bilingual daily brief

At 6am the owner's phone gets a WeChat push: today's 'top 5 worth looking at' — each with an AI one-liner explaining why and a recommended bid range. The #1 car auto-creates a 7am calendar reminder to inspect.

Tech stack

macOS 桌面应用 SQLite (本地) OpenAI GPT-4o WeChat Webhook Google Calendar API Node.js

Working with us

Week 1: shadowed the owner's morning routine for 3 days, captured every judgment rule that makes a car 'worth looking at' — this became the AI valuation prompt. Week 2: delivered the desktop assistant + brief. Week 3: ran on real data, deployed on his own machine.

I used to wake at 6 and leave home at 9 — 3 hours glued to the screen. Now I wake at 6, glance at the brief on my phone, and I'm on the road by 6:30 to inspect the top pick. — Client owner (paraphrased from client interview)

Impact

What's next

On a $150/month retainer covering occasional fixes. Next phase: support for more public data sources plus a 'seller-urgency' AI module (price drop magnitude, listing age, 'must sell this week' style description signals).