Concept & technology

VinReach: AI follow-up for anonymous dealer-website shoppers

Most dealer-website shoppers leave without filling in a form; VinReach follows up with them about the exact vehicles they viewed.

Built by Magnitude's founder Private beta Published

The short answer

VinReach is AI remarketing software for car dealers, in private beta as of September 2026. One script tag identifies consenting anonymous shoppers, an authorized crawl reads the dealer's live inventory, and each shopper gets a one-to-one email about the exact vehicles they viewed. Sold VINs are matched back to those messages. No DMS or CRM integration is required.

At a glance

What is VinReach?

Product
AI remarketing software for car dealers
Status
Private beta, Founding Dealers cohort
Company
VinReach, Inc. (Delaware), built by Magnitude's founder
Install
One async script tag
Inventory
Authorized crawl of the dealer's site
Website
vinreach.ai
Built on
Cloudflare Workers and R2 · Next.js · Postgres with row-level security · Temporal · Claude

The problem

Why do most dealer-website shoppers never become leads?

Dealer websites typically turn only a few percent of visitors into form leads. That is an industry framing behind VinReach's positioning, not a VinReach measurement; everyone else shows up as traffic, not as a lead anyone can call.

How it works

What does VinReach do, step by step?

  1. Identify

    A pixel records vehicle-page views; with consent, identity resolution links them to a person.

  2. Crawl

    An authorized crawl of the dealer's site picks up vehicles, price changes and sold flags.

  3. Generate

    AI drafts a one-to-one email about the vehicles viewed, in the dealership's voice.

  4. Book

    Replies go to the dealer's BDC, the team that books appointments.

  5. Prove

    Sold VINs are matched back to the messages sent, each with an evidence label.

Architecture

How is VinReach built?

VinReach architecture by layer Two inputs feed VinReach: the shopper's browser, through a pixel installed with one script tag, and the dealer's public website. Browser events pass through the edge layer on Cloudflare Workers, which scrubs data, hashes the IP address and reads the Global Privacy Control signal, and then into identity resolution, which matches visits within one store. The inventory engine crawls the dealer's website, extracts vehicles, checks VINs and emits price-drop and sold events. Identity and inventory events both start journeys, which run as Temporal workflows behind a send gate. The AI layer drafts emails with Claude (with a fallback model), triages replies and tunes subject lines and send times. Emails go to the shopper's inbox, and replies go to the dealer's BDC. Underneath, Postgres with row-level security holds each dealer group's data, and attribution matches sold vehicles from the inventory engine back to the sends. SIGNALS IN VINREACH LAYERS Shopper's browser pixel: one script tag vehicle-page views 01 EDGE Cloudflare Workers R2 photo CDN scrub, hash IP, read GPC 03 IDENTITY Per-store matching exact + scored links borderline → human review Dealer's website public vehicle pages prices & sold flags 02 INVENTORY Authorized crawl extract · VIN check · diff price-drop & sold events 04 JOURNEYS Temporal workflows send gate: opt-outs, caps exit on opt-out or reply 05 AI LAYER Claude drafts reply triage → people subject bandit · send time Dealer's BDC replies & handoffs Shopper's inbox one-to-one email AI disclosure by default 06 DATA + ATTRIBUTION Postgres with row-level security per dealer group sold vehicles matched back to sends, with a confidence label sold events sends VinReach architecture by layer Two inputs feed VinReach: the shopper's browser, through a pixel installed with one script tag, and the dealer's public website. Browser events pass through the edge layer on Cloudflare Workers, which scrubs data, hashes the IP address and reads the Global Privacy Control signal, and then into identity resolution, which matches visits within one store. The inventory engine crawls the dealer's website, extracts vehicles, checks VINs and emits price-drop and sold events. Identity and inventory events both start journeys, which run as Temporal workflows behind a send gate. The AI layer drafts emails with Claude (with a fallback model), triages replies and tunes subject lines and send times. Emails go to the shopper's inbox, and replies go to the dealer's BDC. Underneath, Postgres with row-level security holds each dealer group's data, and attribution matches sold vehicles from the inventory engine back to the sends. SIGNALS IN Shopper browser + pixel Dealer site vehicle pages 01 EDGE Workers R2 hash IP · GPC 02 INVENTORY Crawl extract · diff VIN check 03 IDENTITY Per store exact + scored review band inventory events 04 JOURNEYS Temporal workflows send gate: opt-outs, caps exit on opt-out or reply 05 AI LAYER Claude drafts + triage checked against the vehicle record subject bandit · send time Inbox one-to-one email Dealer's BDC replies & handoffs 06 DATA + ATTRIBUTION Postgres + row-level security isolated per dealer group sold vehicles matched to sends sends
Simplified from the VinReach codebase, September 2026. Dashed boxes sit outside VinReach.
  • Edge. Cloudflare Workers validate, scrub and bot-filter every event before storage; R2 serves vehicle photos.
  • Inventory. Structured data first, then platform-specific selectors, then LLM extraction; VINs must pass the check-digit test.
  • Identity. Matching stays inside one store; weak links are never written, borderline ones go to human review.
  • Journeys. Durable Temporal workflows that exit on opt-out or reply, behind a send gate that fails closed.
  • AI. Claude drafts emails (with an automatic fallback model), checked against the vehicle record; replies about price, complaints or legal matters go to people.
  • Data. A Next.js app on Postgres, with row-level security isolating each dealer group.

Guardrails

What privacy and attribution guardrails are built in?

Engineering mechanisms, not a statement of legal compliance.

  • No raw IP addresses. The edge keeps a coarse network bucket and a hash whose salt rotates daily.
  • Global Privacy Control. The browser signal is read at the edge and travels with every event.
  • Append-only consent ledger. Entries are hash-chained, and an opt-out reaches every running journey.
  • AI disclosure. AI-drafted messages carry a disclosure by default.
  • Labelled attribution. Each sale is marked as dealer-reported or inferred from inventory, and an inference counts only when a quorum of independent sold signals agree.

Status

Where is VinReach today?

September 2026

Private beta with a Founding Dealers cohort. No customer results are published yet; they will be added only once the dealers involved sign off. VinReach, Inc. is a Delaware company built by Magnitude's founder.

FAQ

Frequently asked questions

How does VinReach identify anonymous shoppers?

A pixel added with one script tag records which vehicle pages a shopper views. Identity resolution links those visits to a person within a single store. Optional third-party identity resolution is off by default and needs the shopper's analytics and marketing consent.

Does a dealer need a DMS or CRM integration?

No. VinReach reads inventory by crawling the dealer's own website, with the dealer's authorization captured during onboarding, and replies go to the dealer's BDC. If a dealer has its own inventory feed, the feed takes precedence over the crawler.

How does VinReach decide a sale was influenced?

When a viewed vehicle sells, VinReach matches the sale to recent viewers and labels it as reported by the dealer or inferred from the vehicle leaving the website. An inferred sale counts only when a quorum of independent sold signals agree.

Work with Magnitude

Marketing a dealership?

Magnitude builds paid media and conversion systems for automotive dealers and real estate developers.

Talk to Magnitude VinReach Founding Dealers