The Island Vehicle Arbitrage Engine
You described it in about seven minutes on the lanai. This is what it looks like as a system: eleven modules that find the spread, prove it, chase it, and hand you only the decisions that need a human. Nothing here is theory. Every piece is buildable with tools that exist today.
- Prepared for
- Joe Gowans
- Market
- Mainland → Maui
- Source
- Conversation, 8/28
- Status
- Concept spec
Intake Summary
- Operator
- Joe Gowans — Maui, HI. Currently in due diligence on a business acquisition. Has lot space and street frontage.
- The core thesis
- Certain vehicles carry a real price gap between the mainland and Hawaii. Buy low over there, ship it, sell high here. “I can buy that for fifty thousand, and I'm convinced I can sell it here for seventy.”
- Evidence so far
- Manual, anecdotal searches across Auto Trader, Facebook Marketplace and Cars.com. One worked example (a Hummer). A second read on lifted Tacomas, where the local premium is really paying for install labor.
- Known cost anchor
- Roughly $1,500 to ship a vehicle over. Your number, from the conversation.
- Inventory model
- Hold on the lot to start, on a busy street. Pre-sell selectively once volume justifies it. Auctions as a second supply channel.
- The competitive read
- Dealers already work Facebook Marketplace and move within minutes of a listing going up. Speed is the moat, not taste. “As soon as those cars come up on the Facebook Marketplace, they go out and they buy them.”
- Why it isn't built yet
- Doing it by hand across every model, every market, every day is thousands of hours of work. That is exactly the shape of problem software is for. “Somebody manually would have to do that. It would take thousands of hours.”
- What you want from it
- A system that scans continuously, filters to only the real spreads, verifies the vehicle is clean, opens the conversation with the seller, and escalates to you when there's a decision worth your time.
The Thesis, Restated
There is a repeatable, measurable price gap between what a specific vehicle sells for on the mainland and what that same vehicle sells for on Maui. If that gap reliably clears shipping, recon, tax and holding cost, then the only thing standing between you and the money is the speed and the coverage of the search.
The gap exists. You found it by hand, on real listings, on more than one model. That's the hard part, and it's done.
Coverage and speed. Every model, every market, every hour, instead of whatever you can eyeball on a Saturday.
Taste and judgment. Which cars Maui actually wants, what to pay, what to walk from. The machine never gets a vote on that.
The Machine, End to End
Read it left to right. Thousands of listings enter at stage one. By stage four you're looking at a handful a day that survived a real margin test and a clean title check. Stage five is a phone notification with a yes or a no on it. Everything after your yes is drafting, sending, tracking and follow-up that you never touch again unless the seller says something a machine shouldn't answer.
The Modules
The Spread Model
- All-in cost—
- Carry & overhead—
- Return on capital—
- Annualized turn—
Why this matters more than it looks. Move the "days on the lot" slider. A car that sits for four months can turn a great-looking spread into a mediocre one. That is exactly the variable a human eyeballing Facebook at night cannot track, and exactly the one the system watches on every unit you own.
Human vs. Machine
| Job | Who runs it | How it actually works |
|---|---|---|
| Watching every listing, everywhere | Machine | Continuous monitored searches across the marketplaces and auction feeds, deduped by VIN. |
| Pricing the Hawaii side | Machine | Rolling local comp set built from what's actually listed and what actually sold here. |
| Scoring the spread | Machine | Landed cost model plus a confidence score. Only cars over your threshold surface. |
| VIN, title and history check | Machine | Decode, brand and lien check, accident and recall pull, before you ever see it. |
| Deciding to make the offer | You | One notification, the car, the numbers, the risks. Yes or no. This is the whole job. |
| Writing and sending the offer | Machine, you approve | Drafted in your voice from the listing's own details. You tap send, or set a band where it sends itself. |
| The back and forth with a seller | Machine, you approve | Handles the routine replies, escalates the moment it hits price, condition or anything it shouldn't answer. |
| Negotiating the real number | You | Never automated. The system briefs you, it doesn't close for you. |
| Transport booking and tracking | Machine, you approve | Quotes pulled, booking prepped, milestones tracked to the port and to the lot. |
| Listing it for sale here | Machine, you approve | Photos organized, copy written, cross-posted. You approve the price. |
| Which cars Maui wants | You | The system reports the demand signal. You decide what it means. Your read is the edge. |
Data Sources
| Source | Status | Notes |
|---|---|---|
| NHTSA VIN decoder | Open, free | Government API. Decodes year, make, model, trim, engine, plant from any VIN. Zero friction. |
| Federal recall database | Open, free | Open recalls by VIN. Cheap insurance against buying a headache. |
| Title & brand history | Paid, per lookup | Salvage, flood, lemon and odometer branding through a licensed data provider. Per-VIN cost, only run on cars that already scored. |
| Auto Trader / Cars.com | Monitored search | Structured, stable, well-behaved. The reliable backbone of the mainland scan. |
| Facebook Marketplace | Needs a decision | The richest source and the one with real rules. There's an operator-assisted way to do this that stays inside the lines. Worth a five-minute conversation before we design it. |
| Dealer auctions | Licensing first | Wholesale channels and wholesale pricing benchmarks. Access depends on dealer credentials, which is a business question before it's a software one. |
| Hawaii live comps | Buildable now | What's listed on-island right now, what's been sitting, and what came off the market. This becomes proprietary the day we start collecting it. |
| Ocean transport rates | Quote-driven | The two major Hawaii carriers publish quoting. Rates get pulled per vehicle rather than assumed. |
| Your own sold history | Grows over time | The most valuable dataset in the whole system, and the one nobody else can copy. Every car you sell makes the scorer smarter. |
Build Sequence
Kill the thesis or confirm it, on real data
Before a line of product code. Pick three models you believe in. Pull the real mainland and Hawaii listing data on all three, run the landed-cost math on every pair, and look at the actual distribution instead of anecdotes.
- Real spread per model, with the range, not just the best case
- How often a car that looks good survives the full cost stack
- How fast the good ones disappear, which tells us how fast the system has to be
Scanner, comps, landed cost and the daily brief
The always-on half of the machine. It watches, it prices, it scores, and every morning it hands you a short list instead of a search bar.
- Continuous scan with VIN-level dedupe so the same car never hits you twice
- Hawaii comp engine and the landed-cost model behind the scorer
- Daily brief: the cars worth your attention, ranked, with the math shown
Verification and outreach
Where speed becomes the moat. The system does the title work and drafts the opening message so an offer goes out in minutes, not the next evening.
- Automatic VIN, title, lien and recall verification on every scored car
- Offer drafting in your voice, with your approval gate and an optional auto-send band
- Seller thread handling with a hard escalation rule back to you
Inventory, transport and resale
Everything after the buy. What's on the water, what's on the lot, what's aging, and what should be repriced this week.
- Inventory board from purchase through port to sold, with days-held on every unit
- Transport quoting and milestone tracking
- Resale listing autopilot and a repricing nudge when a unit starts to sit
Honest Flags
Once you're buying to resell at any volume in Hawaii, you're in licensed-dealer territory, which touches everything from auction access to how you can advertise. This is the first question to answer, and it's a phone call, not a build.
Facebook is the best source and the least automation-friendly. There's a legitimate operator-assisted design that keeps a human in the loop and stays inside the terms. It's a design constraint we build around on purpose, not something to quietly ignore.
The $1,500 anchor is a fine planning number and a dangerous scoring number. Rates move by vehicle size, season and carrier. The model quotes per unit instead of assuming.
The math above is per unit. The business is the turn. Cars on a lot are cash you can't redeploy, and the spread only compounds if the inventory moves. Worth modeling before scale, not after.
Hawaii's isolation is exactly why the spread exists. It's also why a wrong buy sits. Narrow demand means a great car at a great price still needs a buyer on this island. Phase one is designed to measure that, not assume it.
You already noticed dealers moving within minutes. If this works, others get there too. The durable advantage isn't the scanner, it's your own sold-price history feeding the scorer. That's the asset. Start collecting it on day one.
What I'd Need From You
The idea was always the hard part. You already did that.
You said it yourself on the lanai: anyone can write code now, but you can't teach the creative read that spots the gap in the first place. That's the part you brought. This document is just what that idea looks like once someone builds the machine around it.
Phase one is small, fast and honest. It either confirms the spread on real data or it saves you from finding out the expensive way. Either answer is worth having.