Vehicle auction deal-finder and valuation agent
Collects vehicle lots, estimates a maximum profitable bid, and emails a ranked shortlist.
The challenge
Buying vehicles at auction is a numbers game. The hard part is not finding listings, it is knowing quickly what a car will really cost once fees, transport, repairs and tax are counted, and what you could realistically sell it for.
The solution
A decision-support system for a used-car resale business. Two n8n workflows collect vehicle lots and comparable listings, using the official eBay API and manual entry where a site's terms prohibit scraping, then log everything to Google Sheets.
Claude estimates landing cost, demand and a maximum profitable bid for each lot. Kill-switch, stock-limit and capital guardrails run in workflow code, every decision is audit-logged and a ranked digest is emailed daily.
What it does
- Respects each source's terms: no scraping where it is not allowed
- Guardrails enforced in code, not left to the AI
- Full audit log of every recommendation
- People decide: no automatic bidding
How it works
- CollectEvery six hours, one workflow gathers vehicle lots and comparable listings, and removes duplicates.
- RecordEverything is written to a shared Google Sheet that also holds the configuration, the stock book and an audit log.
- EstimateEach day a second workflow asks Claude to estimate landing cost, demand and a maximum profitable bid for each new lot.
- Check the guardrailsCode, not the AI, checks the kill switch, the stock limit and the capital available after a contingency buffer.
- Rank and sendLots that pass are ranked and emailed as a daily digest, each with its reasoning.
- DecideA person reviews the shortlist and places any bid. The system does not bid.



Design decisions
Terms were checked before building
The terms and robots rules of each source were reviewed first. Where automated access is not permitted, lots are entered by hand and flow through the same valuation.
Guardrails live in code
Spending limits and the kill switch are enforced in the workflow itself, so a mistake in the AI's answer cannot authorise spending.
Every decision is explained
Each recommendation stores its cost breakdown, the comparables used, its risk flags and its reasoning in an audit log.
No bidding code exists
The system recommends only. There is no bid-placement logic anywhere to switch on by accident.
Built with
Two n8n workflows, the Anthropic API for the valuation step, the official eBay Browse API for comparable listings, Google Sheets as the tracker, and email for the daily digest. Margin targets, stock limits and capital are configurable without touching the workflow.
Got a process that eats your week?
Tell us where enquiries, shifts, quotes or follow-ups slip through, and we will show you what could be automated and what should stay with your team.
Book a free callRelated guides
More projects
Images are system design diagrams, not screenshots of client systems.
