FleetSnap: AI inbox and claims assistant for Turo hosts.
Turo hosts with 5 or more cars stop running their business through one calendar and a Notes app. FleetSnap takes over the inbox, the financial analysis, and the damage-claim path. Same fleet, dramatically less Sunday-night triage.
Live at fleetsnap.ai.
Screenshot from fleetsnap.ai, captured 2026-05-06.
The problem
A Turo host with one car has a hobby. A Turo host with five or more cars is running an operation. The work that wasn't there at one car explodes at five: messaging guests, scheduling cleaners, tracking damage claims, reconciling earnings, chasing supplements, answering "is the car available?" at 11pm.
The hosts who scale past 5 cars do it on willpower. They wake up to 30 unread guest messages and triage by urgency on a phone. They keep finances in a spreadsheet that gets touched once a quarter. Every claim they file by hand is 45 minutes they could have spent on a car they actually own.
The ICP is fleet-sized: Turo hosts with 5+ cars who have already proven the unit economics and now need the operational layer that turns a portfolio into a business.
What Downshift built
One assistant across three host workflows: inbox, money, claims. Nine named surfaces, each tied to one job a fleet host has to do today.
AI inbox with suggested replies
Every Turo guest message gets a Claude-generated suggested response in the host's voice. The host approves, edits, or rewrites. The 30-message backlog becomes a 5-minute review.
Concierge VA layer
On the Concierge plan, a human VA layers on top of the AI suggestions. Hosts hand the inbox over entirely; the assistant only escalates the edge cases.
Turo earnings analysis
Pulls Turo earnings data and answers the questions hosts actually ask: which cars are profitable, which months were down, what insurance and platform fees are doing to the take.
Financial statement analysis
Upload statements; the assistant categorizes, totals, and flags. Quarterly tax-prep stops being a weekend.
AI damage detection (vision)
Legacy from RepairSnap's vision pipeline. OpenAI GPT-4o detects damage from before/after photos; Claude generates itemized cost ranges with location-aware labor rates.
Claim and supplement workflow
Damage report → estimate → supplement request, all from one screen. Stops the back-and-forth that loses fleet hosts weeks per claim.
Tier-sized pricing for fleets
Pro plan: $99.99/mo base + $6 per vehicle. Concierge plan: $199.99/mo base + $20 per vehicle (5-vehicle minimum). Pricing scales the way a fleet does.
Legacy damage estimate plans
For hosts who only need damage estimation, four legacy tiers (Starter $19.99 / Manager $59.99 / Pro $99.99 / All-Star $199.99) keep the lighter use case priced fairly.
Pre-rendered for AI search
Static HTML snapshots pre-built for AI and search bot crawlers via the project's prerender step, so the marketing pages are answerable by Cursor / Perplexity / Claude / search engines without JS.
Screenshot from fleetsnap.ai, captured 2026-05-06.
The architecture
Full-stack TypeScript on a Neon-Postgres + Railway substrate. Two AI providers split by job: Claude on conversation and quotes, GPT-4o on vision.
Frontend
- React 18 + TypeScript
- Vite + TanStack Query
- Wouter routing
- Static HTML snapshots via prerender step (for AI/search bots)
Backend
- Express + TypeScript
- Drizzle ORM
- PostgreSQL (Neon serverless)
- Server timeout 300s for image-processing requests
AI surface
- Anthropic Claude for inbox suggested replies and repair quotes
- OpenAI GPT-4o for legacy damage detection (vision)
- Same vision-AI lineage as RepairSnap
Ops
- Railway hosting (push → main → staging → production)
- Stripe subscriptions across two pricing systems
- SendGrid transactional email
- PostHog + LogRocket analytics
Real outcomes
Numbers below are from the production codebase pricing config and from fleetsnap.ai. No fabricated user counts.
Fleetsnap Pro base monthly. Plus $6 per vehicle.
Concierge plan base monthly. Plus $20 per vehicle. 5-vehicle minimum.
Vehicles in the fleet ICP, where the operational pain becomes acute.
Pricing systems wired in production: legacy damage estimate plans + Concierge fleet plans.
Host workflows the assistant covers: inbox, money, claims.
AI providers split by job: Claude on conversation + quotes, GPT-4o on vision.
Server timeout for image-processing requests, sized for vision-AI rounds.
Legacy damage-estimate tiers ($19.99 / $59.99 / $99.99 / $199.99) for hosts who only need estimation.
Sources: Concierge / legacy pricing tiers and the 5-vehicle Concierge minimum come from ~/Github/FleetSnap/fleetsnap/CLAUDE.md (server/concierge-pricing.ts + Stripe price-id table). Tech stack and 300s timeout come from the same file. Per-host MAU and revenue are not on the public marketing page and are shared on request.
Screenshot from fleetsnap.ai, captured 2026-05-06.
What's in flight
FleetSnap's near-term focus is the inbox + financial analysis pair. The legacy damage-estimate flow keeps the lighter use case alive. Production hardening continues with a deliberate staging-first deploy path (push to main → push to staging → manual production cutover), so a Turo host's live inbox is never the test environment.
Want this for your venture?
Downshift deploys a pod of senior engineers into the venture, with one of them in the technical co-founder seat, and ships production-grade AI products. Most go live in two to four months. No equity required.
See more shipped Downshift ventures on the wins page, or read the RepairSnap, FairStack, and Rivalize case studies.