Private-party cars, overnight, in the CRM you already run

We work with independent lots and dealer groups whose used-car supply still depends on private sellers. The person we sit with is usually the inventory manager or the buyer who opens Facebook Marketplace at 6 a.m., then Craigslist, then Cars.com, then a local forum that only they remember. They copy a VIN or a plate, guess a wholesale number, and type a first message before the seller goes to work. Cyriac Zeh founded Nidrosoft in San Diego to build that night job into software the desk already owns. If your only problem is a slow OEM allocation, call the franchise. If your problem is cars that sell on Marketplace before your BDC wakes up, keep reading.

A normal Monday on a mid-size used lot looks like this. The manager prints a gap list: trucks under a price, SUVs in a mileage band, anything that aged past forty-five days and needs a replacement. Two people split the marketplaces. They message twenty sellers. Eight reply. Three are already pending. Two want a number that does not work. One books an appraisal. Meanwhile a car the desk wanted at 7:10 is gone by 8:00 because another lot sent a cleaner first text. That race is the work. We map it before we talk about models.

The CRM handoff is where most homemade scripts die. A seller thinks they are talking to a person. The buyer needs the year, mileage, title status, payoff, and a phone number that is not a Marketplace proxy. The desk needs a row in the pipeline: new, waiting, appointment, appraisal, bought, dead. If the conversation lives only in a phone, the next shift cannot see it. If it lives only in a spreadsheet, the appointment never hits the calendar. We treat the pipeline object as the product. Nexuvo already runs that object for dealers who let the night job write into it.

We are the wrong seat when you want a consumer car-shopping app, a new dealer website, or a chatbot on the service drive that cannot see inventory. We are the wrong seat when you will not own the marketplace accounts and you want us to guess passwords. We are the wrong seat for a group that only wants a board slide about “AI on the lot.” If you want ChatGPT to write Facebook ads, buy seats. If you want outreach that runs while the lot is closed, and a human who still approves the buy, write cyriac@nidrosoft.com.

Finding the car is a search problem with bad inputs. Sellers write “runs great,” skip the VIN, and post the same truck on three sites with three prices. Photos are night shots in a driveway. Mileage is “around.” Title is “clean I think.” A buyer who has done this for ten years can smell a rebuilt title from the wheel gap. A junior BDC cannot. The desk already has rules: no salvage, no flood, no seven-passenger vans over a mileage line, yes to one-owner trucks in a zip radius. Those rules rarely live in software. They live in the manager’s head and in a text thread with the closer. We write them down as filters before anyone sends a message.

Scoring is wholesale math, not vibes. The desk looks up a book number, subtracts recon they can already see (tires, windshield, a quarter panel), subtracts a margin, and decides whether a first offer is worth typing. If the payoff is underwater, the conversation is a different conversation. If the seller is a dealer posing as a private party, the time is wasted. Buyers already do this in a second tab. The failure is that they do it after they have already written a long message, or they skip it when the night gets busy. A score on the row before outreach is the difference between a full pipeline and a polite inbox.

Messaging has a clock. Marketplace threads go stale in hours. Craigslist emails bounce. Cars.com leads arrive with a form the seller filled in once and never checked again. The first text that works is short, names the car, and asks for a VIN or a time, not a paragraph about the dealership’s values. Follow-ups need a gap: too fast looks like a bot the seller will block, too slow and the car is gone. Most desks do follow-up when someone remembers. That is why the night job matters. Cars list after dinner. The desk that waits until 9 a.m. is buying leftovers.

Appraisal and buy are the human gate. A number on a screen is not a check. Someone still has to see the car, run the VIN, confirm the title, and decide. Transport, payoff, and the deal jacket come after. We do not automate the signature on a purchase. We do automate the work that never should have required a person to be awake: the hunt, the first message, the nudge, and the row in the CRM. If your store already has a strong appraisal process and a weak hunt, that is the Nexuvo-shaped problem. If your hunt is fine and recon is the fire, say so. We will not pretend a messenger fixes a body shop.

Once a seller replies, the thread becomes a deal or a dead end in a few moves. They name a price. The desk names a range. Someone asks about title, keys, and whether the car is local. A good buyer already has the next question ready. A tired night of copy-paste does not. We keep the score on the row so the morning person can see why the first offer was what it was. If book and recon have moved since 1 a.m., they change the number before they drive. That is the opposite of a bot that keeps arguing a stale wholesale.

Appointments fail for boring reasons. The seller gives a gate code and then goes to work. The buyer sends a junior with no authority. The address is a storage lot that is closed on Monday. We put the constraints on the file: window, location type, who can write a number, and whether a transport quote is needed if the car sits forty miles out. A calendar hold that the CRM can see beats a text that only one person received. If your store already runs tight appointments and loose sourcing, we will not rebuild the calendar. We will fill it.

The deal jacket is where retail stores still go paper-heavy. Title, ID, odometer, payoff letter, bill of sale. AI does not make a DMV clerk faster. What we can do is refuse to mark a car “bought” in the pipeline until the required scans sit on the row, and we can draft the seller checklist before anyone leaves the lot. That is document hygiene, not a magic closer. Stores that skip it spend Thursday hunting a title that was “in the car.” We have no interest in a scope that hides that hunt behind a generated email.

DMS write-back is a later conversation. Some groups want the acquired unit in the inventory system the same day, with a stock number and a cost. Some still key it. We do not force a DMS project to justify the hunt. Nexuvo’s first win is cars in the pipeline and hours off outreach. If your CDK or Reynolds project is already on fire, finish that with the people who own it. We can talk about a write-back after the night job is real and a human still approves the cost that hits the book.

Recon is a queue with parts, keys, and a bay. A car arrives, someone writes a list, the shop orders a windshield, and the car sits behind a truck that has been “almost done” for nine days. Photos for the web wait on the wash. The aged-unit meeting on Friday is a spreadsheet of excuses. We can draft a recon checklist from the appraisal notes and we can flag a unit that has had no status change in a set number of days. We cannot sand a quarter panel. If recon is your only fire, hire a shop manager. If recon is the second fire after a thin pipeline, say that in the audit and we will rank it second.

Photo days are a production line. Background, wheels turned, interior shots, a walk-around that does not cut the roofline. Portals punish a dirty car and a vertical phone clip. Desks already know this and still miss it when the lot is slammed. A build here is a required shot list and a hold on syndication until the list is complete, the same habit we use on real estate listings with Nomerlo’s public-gallery side in mind. The object is different. The refusal to publish a bad first frame is the same. A model that writes a description from the VIN is optional. The shot list is not.

Aged units are a pricing and attention problem. A car that sat sixty days has a story: wrong color, a missed recon item, a price that was brave in week one. The meeting already happens. What is missing is a weekly packet that shows cost, book, photos, and the last price change in one place so the manager does not rebuild it from the DMS and a group text. We will build that packet. We will not promise a model that “knows” the right price for your market. Your closer already has an opinion. The packet should make that opinion faster to defend.

Service-lane and BDC chat are adjacent and usually a distraction in the first scope. A customer asking whether a loaner is ready is a real message. It is also a different system, a different approval, and a different risk if the bot invents a promise. Nidrosoft will take that work after the inventory night job is owned, or we will send you to a shop that only does service chat. We would rather lose that slice than let it eat the audit. The lot feels units. Chat impressions do not pay the floorplan.

Nidrosoft is the San Diego studio of Cyriac Zeh. He ships end-to-end AI product work under the name The AI Architect. Before the studio he worked at Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. The automotive work is Nexuvo: a live acquisition system, not a concept car. We keep 16 public systems in the archive and we have shipped 125 products across the studio. We name those numbers so you know the shop builds software for a living. We do not name fake rooftops. If a dealer group is not on /work, we will not hint that they are.

Other Nidrosoft systems show how we treat queues and approvals, even when they are not cars. Eviction Wizard holds a legal file from notice to writ with a human on the filing. TokenTra shows AI spend by feature so a team can cut 10 to 30 percent of wasted tokens. Protectron drafts EU AI Act documents. Those products matter to a dealer group that is about to bolt a model onto the CRM and has no idea what the bill will look like, or that sells into Europe and just got a questionnaire. They are not automotive features. They are proof we already operate meters, documents, and human gates.

Guidera, Octana, Helia, Rizzer, Night Owl, and Nourishr are consumer and media systems in the same archive. We will not tell a used-car manager that a travel companion or a dating coach is relevant to their wholesale. We mention them only so you can see the studio ships finished products, then we return to the lot. If you want the consumer work, it is on /work. If you want cars, stay on this page and ask about Nexuvo and the CRM you already pay for.

Embed means we sit with the buyer at the hour they actually hunt. If that is 6 a.m. on Marketplace, that is when we watch. We see which filters they trust, which sellers they skip, and which messages they will not send. We take the CRM login they already use and we watch a lead die because the phone number never left the marketplace thread. We do not start with a tour of model vendors. The first week is access, shadowing, and the list of boards you already work. The longer write-up lives on /how-we-work.

Diagnose means we cost the hunt. How many listings the desk opens in a day. How many messages go out. How many get a reply. How many become appointments. How many become buys. We also cost the misses: cars the rules would have wanted that nobody messaged because the night got busy. Hours and missed units both go on the map. We score whether we can reach the marketplace accounts and the CRM. A beautiful score with no write path is a report, not a build. The audit ranks the queues that have both pain and a door.

Scope means you buy a result. For most lots the result is a night queue that drafts and sends outreach inside your rules, and a morning pipeline a person can run. We write what the system may say, what it must never say, and when a human has to take the thread. We write the buy decision as a human step. We refuse a scope that says “automate the dealership.” If you cannot name the board, the CRM, and the person who appraises, you are not ready to buy. That sentence saves both of us a quarter.

Build means we ship into your accounts. Your Marketplace identity, your CRM, your book tool if we are allowed to see it. Your people sit with us while the first live sequence goes out to a small slice of the radius, then the slice grows. Someone on the desk has to be able to pause a sequence, edit a template, and explain a score on a Friday. That person is part of the build. We do not leave a black box that only Nidrosoft can touch. Twelve-plus years of product work has made us allergic to that move.

Own means the credentials and the rules stay yours. We maintain what you ask us to maintain. We stay out of the rest. If a platform changes its inbox rules, the owner of the account is you, and we adapt the sequence with you. Continuous work is more units, cleaner scores, and fewer dead threads, measured on the same two numbers we already publish: inventory acquired and hours off outreach. If those numbers do not move, we say so and we stop expanding the slice.

Do not hire Nidrosoft to scrape a marketplace you do not have the right to use. You own the accounts. You own the relationship with the platform. We build the queue and the sequence on access you provide. If your plan is a hidden farm of accounts and a shrug when a listing site blocks you, we will walk. The night job is only useful if it still exists in six months. A ban that takes down the buyer’s personal Facebook is a store problem we will not create for a demo.

Do not hire us to hide a title problem. Salvage, flood, odometer, a seller who is not the owner: those are human calls and often legal ones. A sequence that pressures a seller past those questions is not a product we will ship. The same no applies to payoff games and to any write that moves money without a person. We are not a lender, an auction house, or a title clerk. If you need those, hire them. If you need the hunt to stop living in three tabs, we can talk.

Do not hire us for a new car-shopping consumer app unless that is the actual product and you have the appetite for a full build. Nexuvo is a dealer tool. Guidera is a travel companion. Those are different objects. A group that wants “our Turo” or “our Carvana” is asking for a company, not a night queue. We can say no in the first call. That no is cheaper than a six-month prototype that the lot never wanted.

We are a poor fit for a single-point lot that wants a magic number of cars with no one who can appraise, and for a public group that only wants a keynote. Nidrosoft’s main seat is an operator with a repeating hunt and a decision maker in the room, often 30 to 250 people, sometimes smaller when one pipeline pays for itself. A two-person independent can still write. We will tell you if Nexuvo as it stands is enough, or if you should keep hunting by hand for another season.

An AI audit is two weeks in the work. We sit through a hunt, we export a week of messages if you can share them, and we walk the CRM pipeline from new to bought. We write down the filters you already trust and the ones you only think you trust. We note which boards die after 10 p.m. and which ones the junior staff never open. The deliverable is a ranked map: night outreach first, recon packet second, photo hold third, or whatever your numbers say. The page for that service is /services/audits.

Access is the gate. Marketplace identities, CRM, and a sample of sold jackets with seller data redacted if counsel wants that. We use your accounts and your policies. We do not ask you to paste seller threads into a consumer chat product so we can “see what GPT does.” If a platform login is personal to an employee who is leaving, we say that in week one because the night job will die when they do. That is an ownership finding, and it is worth the audit by itself.

We score failure the way the lot scores it. A missed truck that fit the rules is a missed unit. A message that promised a number nobody can honor is a reputation hit and a wasted drive. A jacket that marks bought without a title scan is a Friday emergency. Those outrank a slightly prettier first text. Sometimes the first recommendation is a filter list and a shared inbox, with Nexuvo as the second step. That is a successful audit. You still own the map and you can take it to another shop.

At the end you get a build order in plain language. If the first line is “run Nexuvo on this radius and these rules,” we will write that. If the first line is “stop letting the night person hunt from a personal phone,” we will write that even if it does not need us. The audit is the product. The build is a separate buy. Start at /contact or send cyriac@nidrosoft.com a redacted week of the hunt and the name of the CRM.

Real estate transfers the overnight queue and the public-facing packet. Eviction Wizard holds dates and documents. Listing desks fight syndication lag the way lots fight portal photos. We will not tell a dealer that a notice form will buy them cars. We will tell them we already built a night job that leaves a human on the commit, and we already refused to publish a bad first frame. Those habits show up on the lot as sequences and as a photo hold.

E-commerce transfers catalog hygiene. Variants, feeds, and an inbox full of “is this still available” are cousins of a stock number and a Marketplace thread. We do not have a flagship storefront product, and we say that on the e-commerce page. The transfer is operational: a row that is allowed to go public only when required fields exist, and a reply that can see the row. Manufacturing transfers the floor board: the aged unit meeting is a shift meeting with VIN labels.

FinTech transfers the meter. If the group starts sending model calls from every desk, TokenTra is how you see the bill by feature and cut 10 to 30 percent of wasted spend. Protectron is how you answer an EU AI Act questionnaire if that market is real for you. We are not your bank and we will not move floorplan money. We will put a human on any write that looks like a purchase. Professional services transfer the workshop: a half-day with the buyer desk on their own boards, when the company is not ready to buy a build.

Write cyriac@nidrosoft.com with the boards you work, the CRM you use, and a redacted sample of messages from a real week. Name the person who can say yes to a buy. Cyriac Zeh reads that mail from San Diego. We will tell you whether the next step is Nexuvo as it already runs, a two-week audit, or a no. We do not need a brand film or a rooftop count. We need to see how the hunt works when nobody is performing for a visitor.

If you already want the ranked map, book the audit on /services/audits. Two weeks, your accounts, your people, a build order you can take elsewhere. If you already know the night job is the fire, say so in the first line. If recon is the fire, say that too so we can tell you we are the wrong first shop. Honesty on that point is cheaper than a kickoff that politely ignores the body shop.

A workshop is available when the desk does not share a story about the hunt. Corporate AI training at Nidrosoft uses your Marketplace filters and your actual dead threads, not a toy catalog. Leadership sees what a sequence can send and what a person still has to sign. The playbook stays. A workshop is enough when alignment is the problem. It is not enough when the hours are already counted and the CRM is already named. We will say which side you are on.

Nidrosoft is twelve-plus years of product work in a studio that still answers mail. 125 products. 16 public systems. The automotive proof is Nexuvo: three times the inventory, eighty percent less manual outreach, Facebook and Craigslist and Cars.com while the lot is closed. If that is the desk you run, write us. If it is not, use the sibling pages or close the tab and go back to the three tabs you already have. Both are honest endings.