A 5,000-car sale read overnight and cut to the few hundred worth your morning — each one annotated, each one with a max bid, sitting in your watch list before anyone is in.
A mid-week wholesale sale runs three to five thousand cars. Your buyer has maybe two hours before the first lane opens. Even at twenty seconds a car — glance at the row, check the grade, decide — that is twenty-eight hours of looking. Nobody is lazy here. The maths simply does not work.
So what happens instead is that the list gets worked top-down until the sale starts, and the rest is never looked at. The cars that would have made money are distributed randomly through five thousand rows, which means most of them are in the part nobody reached.
Overnight, while the sale is still being loaded, the whole run list is pulled — every lot, not the first page and not a sample. Then your buy box is applied to all of it at once: years, mileage bands, makes and models, condition grade floor, title rules, lane, geography, whatever you told me matters.
What survives gets the lookups your buyer would have done one car at a time, nine tabs deep: title brand check, Carfax, AutoCheck, Manheim MMR, book values, and the seller's own announcements and disclosures. Then each car gets a note written in the format your desk already reads, the ones that fail your rules are dropped with the reason kept, and the rest are pushed into the auction's own watch list ranked by margin.
By six in the morning the work is done. Your buyer opens the laptop and the list is already there — not a spreadsheet to interpret, the watch list they were going to build by hand.
The filters are not a fixed product feature set — they are whatever you buy on. Some desks care most about grade and announced frame damage. Some will not touch a car over 90,000 miles regardless of grade. Some want Northeast cars excluded in winter, some want a specific trim level only, some run a different rule for trucks than for sedans.
All of that is configuration, not code. You do not log into anything to change it and you do not pay for a revision — you tell me the rule has changed and it changes for the next sale.
A filtered list still needs reading. What makes a watch list usable at 7 AM is that every car on it already explains itself: title and owner count, what the history report said, the MMR with its range, the grade and whether frame was announced, the recon estimate, and then the number that matters — the maximum bid that still holds your margin.
Written once, in your wording, the same way every time. The useful side-effect is auditability: when a car turns out badly, the note says what was known before the bid, so you can tell the difference between a bad call and bad luck.
Nothing is deleted silently. Every car that comes out of the list is logged with why — branded title, three accidents on Carfax, recon estimate over your ceiling, bidding already above MMR, odometer discrepancy declared.
That matters for two reasons. It lets you check the rules are doing what you think rather than asking you to trust them, and it means that when you want to loosen a rule, you can see exactly how many cars it was costing you.
Most dealers want two things: the cars in the auction platform's own watch list, so the buyer logs in and they are simply there, and a ranked email or Google Sheet as the readable version. Both, usually.
Beyond that it can go wherever you work — Slack or Telegram for instant alerts on a car that scores unusually well, a shared sheet the whole desk can open, a direct write into your DMS or inventory database, or a CSV into whatever tool your analyst already uses.
This runs on the auction accounts you already hold and are licensed to use, and the history-report subscriptions you already pay for. I do not resell auction data, share credentials between clients, or create access you do not have. A system built on borrowed access dies the first time somebody audits it, and it takes your account with it.
It also does not bid. Most platforms prohibit automated bidding outright, and the downside of a bug in a bidding bot is that you own a car nobody wanted. The system finds, checks, annotates and ranks. A person decides.
Manheim, ADESA, ACV Auctions, OPENLANE, BacklotCars, SmartAuction, EDGE Pipeline, Copart, IAA, Your own dealer portal
Python, Selenium, Playwright, Pandas, Cookie management, Proxy rotation, Cron / scheduling, SMTP
The run list itself is fast — a five-thousand-lot sale is read in minutes. The slow part is the per-VIN lookups, because those run at a rate that does not hammer your accounts. A couple of thousand VINs fully enriched overnight is routine, and since your filters have already removed most of the sale before that stage, that is normally the whole qualifying list with room spare.
Yes, on the platforms that have a watch list, using your own logged-in account. That is the version most dealers want, because the buyer does not have to do anything with a spreadsheet — they log in and the cars are already flagged. You also get the ranked email as the readable copy.
They usually are, and that is the point of a custom build rather than a product. Different rules per segment, recon ceilings by your own labour rates, margin floors that vary by make, exclusions on specific trims or regions — all of it is configuration. You tell me the rule in plain English and it applies from the next sale.
$500 a month per auction site. That covers every sale on that site, Monday to Friday — if it runs three sales a day, that is fifteen sales a week, all included. There is no build fee, no per-car charge and no per-seat charge: the same price covers a 900-car sale and a 6,000-car sale, and hosting, site changes and fixes are part of it.
That is the normal way in. Tell me the auction and your buy box, and I will triage one actual upcoming sale and send you the watch list it produces. If the calls look wrong, we fix the rules then — before you have paid for a month. A one-off pull of a single sale is $150, and it comes off your first month if you go ahead.
Almost certainly. Most of the trade buys on a platform nobody outside their region has heard of — a private dealer-only marketplace, a regional auction's own site, a lender's repo portal, an in-house system with a listings module bolted on. Obscure portals are usually easier than famous ones, because nobody has ever bothered to defend them. Send me the name and a couple of screenshots and you will have a straight answer within a day.
It breaks, like every scraper eventually does — anyone who tells you otherwise is selling something. The difference is that you hear it from an alert the same morning rather than from three quiet days of an empty watch list, and fixes on systems I built are covered by the monthly, not quoted as a new project.
No, and I will not build that. Most auction platforms prohibit automated bidding, and a bug in a bidding bot means you own a car you never wanted. This finds, checks, annotates and ranks. The bidding stays with your buyer.