How to Build an App Like Carvana: Online Car Sales from Search to Delivery

Most people building a Carvana-style platform are not trying to compete with Carvana nationally. They are a dealer group with 5,000 vehicles in inventory who want to stop losing online buyers to Carvana. Or a fleet remarketing operation that needs a self-service buying portal for commercial buyers. Or a rental car company selling aged-out vehicles without sending them to auction. The problem is the same across all of them: buyers want to complete the full transaction online, and no off-the-shelf product handles the full stack.

A core platform costs $80K-$140K and takes 16-20 weeks. Here is what that covers:

The four hard problems are: financing (use Dealertrack or RouteOne, not Stripe), digital paperwork (state-specific contracts via DocuSign), 360-degree photos (Spincar is faster than a custom build for V1), and delivery logistics (requires a driver mobile app with condition assessment). The VIN is the primary key for everything.

Who builds this instead of buying Carvana's model?

Four specific categories of business build their own digital retail platform rather than feeding inventory to Carvana or paying CarGurus for leads.

Regional dealer groups with 10+ locations and thousands of vehicles in inventory already have the product. What they lack is the buying experience. A branded digital retail platform lets them capture online buyers without paying Carvana a $1,500-$3,000 lead acquisition fee per vehicle sold. At 500 vehicles sold per year, that is $750K-$1.5M annually going to a competitor who also happens to be buying their inventory at auction.

Fleet remarketing operators managing and reselling fleet vehicles for corporations, rental companies, or government agencies need a self-service channel for commercial buyers. Procurement teams prefer to research and purchase without talking to a salesperson. Fleet transactions are often $20K-$80K per vehicle, and the buyer's decision process is more analytical than emotional. A Carvana-style platform with detailed vehicle history, condition documentation, and online financing fits this buyer better than a traditional dealer website.

Rental car companies selling aged-out vehicles have known condition history, documented service records, and a captive supply chain. Hertz launched Hertz Car Sales as a direct digital retail channel for this reason. When a vehicle ages out of the rental fleet, it goes into the sales platform with full transparency, rather than to a wholesale auction where the company captures less margin.

Independent dealer platforms aggregating inventory from multiple independent dealers need a consistent buying experience across all listings. The dealer pays a SaaS fee for access to the platform; the platform owns the buyer relationship. This is the multi-dealer marketplace model, and it requires a dealer onboarding layer that a single-dealer build does not.

According to Cox Automotive's 2024 Car Buyer Journey Study, 79% of car buyers want to complete at least one step of the purchase process online. The demand is not the question. The question is whether your specific inventory, buyer type, and margin structure justify a custom build over buying leads from an aggregator.

How does the business model work?

Carvana made $14.4 billion in revenue in 2024. Their model is straightforward: they buy vehicles at wholesale, recondition them, photograph them, and sell them at retail. The gross profit per unit (GPU) is the difference between what they paid for the car and what they sold it for, minus reconditioning and logistics costs. Carvana's GPU was approximately $5,500 per vehicle in 2024, per their annual report.

When you build your own platform, you have three monetization models to choose from depending on your position in the market.

Direct retail (dealer group model): You own the inventory. You buy at wholesale, sell at retail, and the platform is your distribution channel. The platform cost is justified if it improves your sell rate or average selling price versus traditional channels.

Commission marketplace (multi-dealer model): You aggregate inventory from dealers who pay to list. You charge a per-transaction fee (typically 1-3% of the sale price) or a flat fee per unit sold. At 1,000 vehicles per month at an average $25,000 sale price and a 2% fee, the platform generates $500,000 per month in revenue. The platform cost is a fixed overhead that becomes highly profitable at scale.

Lead generation (lighter build): You do not process the transaction. You generate a qualified buyer who has been through the financing calculator, viewed the vehicle, and initiated a purchase. You sell that lead to the dealer for $500-$1,500. This is a lower-risk model with lower margins, but it is a valid starting point if you want to prove demand before building the full transaction layer.

The financing integration is not just a product feature. Finance and insurance (F&I) profit is the margin engine of any automotive retailer. According to the National Automobile Dealers Association (NADA) 2024 Annual Report, the average dealer earns $1,575 in F&I gross profit per used vehicle sold. A platform that captures this F&I income rather than routing it to a brick-and-mortar finance office changes the unit economics of the business significantly.

What features do you need at each stage?

V1 — what you need to open the doors ($80K-$140K, 16-20 weeks)

Skipping the lender network API and using "call us to arrange financing" instead delays most transactions by 1-2 days and loses buyers to competitors who process instantly. That lost conversion is the real cost of skipping it.

V2 — growth features (add at 100+ transactions per month)

V3 — scale features (above $50M GMV or 2,000+ monthly transactions)

At this volume, manual state paperwork expansion becomes a bottleneck. A compliance management layer with Darwin Automotive or Reynolds & Reynolds integration handles per-state form sourcing programmatically rather than requiring legal review for each expansion. Budget $50K-$80K plus ongoing vendor fees.

Multi-state title management, particularly for electronic title states versus physical title states, requires a title operations workflow. This is primarily a people-and-process problem, not just a software problem. Budget 2-3 operations headcount before the software investment makes sense.

Real-time lender decisioning with multiple lender integrations (rather than a single lender network) becomes relevant when you want to route applications to specific lenders based on credit profile or vehicle type.

Build vs. Carvana: when does a custom platform win?

Keep using Carvana or CarGurus when:

You have fewer than 100 vehicles in inventory and no brand recognition. Aggregator platforms give you buyer traffic you have not earned yet. The lead cost hurts less than the platform cost at low volume.

You sell through multiple channels (auction, dealer trade, retail) and do not want to own the buyer experience end-to-end. Aggregators are a distribution channel, not a competitive threat at this stage.

Your sales cycle requires in-person negotiation. High-trim, specialty, or modified vehicles where buyers want to inspect before committing do not convert well in a fully online flow.

Build your own when:

Your monthly F&I gross exceeds $100K and you are currently routing it through a physical finance office or losing it to aggregators. A custom platform captures that margin directly.

You have a repeating supply (fleet, rental, lease returns) where standardized reconditioning and known vehicle history let you offer the transparency online buying requires. Random auction purchases are harder to document to the standard that converts online buyers.

Your buyer is a commercial or fleet procurement team. These buyers prefer a self-service, documented process over negotiating with a salesperson. The Carvana consumer model maps almost perfectly to this buyer type with minor adjustments.

The payback period for a $110K platform build is roughly 60-90 vehicles per month at $1,500 net improvement per unit (from eliminated lead fees or captured F&I). Most dealer groups at 200+ vehicles per month see payback inside 18 months.

How does the buying flow actually work?

"The unit economics of digital auto retail only work when you eliminate the information gap that forces buyers to visit a lot. 360-degree photography and transparent pricing close that gap." -- Karen King, VP Digital Retail, Cox Automotive (Automotive News, 2023)

The eight-step flow is the skeleton of the platform. Each step has a specific engineering piece behind it, and the cost consequences are worth understanding before you prioritize.

Search filters by year, make, model, price, mileage, body style, color, and features. A search backend that handles faceted filtering well saves 3-4 weeks compared to building on a basic SQL query layer. The monthly payment estimate shown in search results is calculated client-side and does not require a backend call.

Vehicle detail page with 360-degree spin, static photos, vehicle history report link, and feature list from VIN decode. According to Dealer Inspire's 2023 Digital Retail Report, vehicle detail pages with 360-degree spin viewers convert at 2.4x the rate of pages with static photos only. Spincar handles the spin; the integration is an embed URL returned after photo upload.

Financing calculator is client-side math: monthly payment given price, down payment, interest rate, and term. Interactive sliders update the payment in real time. This sets buyer expectations and reduces drop-off on the credit application step.

Credit application collects name, address, employment, income, and SSN last four digits (for a soft pull). This submits to Dealertrack or RouteOne via API. Lender decisions come back within minutes. The buyer selects a financing offer. The platform does not make lending decisions.

Trade-in collects VIN, mileage, and condition. The platform queries Black Book or KBB via API and returns an offer. Seven-day expiration is standard. The offer deducts from purchase price at checkout.

Checkout confirms financing offer, trade-in value, and total. Stripe processes the down payment deposit. The actual loan is originated by the lender; they pay the dealer directly.

Paperwork via DocuSign for the buyer's state: retail installment sale contract, title transfer documents, and power of attorney for DMV filing. Sourcing the correct state forms is a legal problem, not a software problem. Work with a compliance vendor for each state you launch in. Budget 4-6 weeks of legal review per state.

Delivery scheduling lets the buyer pick a date and time window. The platform assigns a driver. The driver app handles everything from here.

VIN decoding: the foundation everything else depends on

Every vehicle in your system is identified by its VIN. The VIN is a 17-character code that encodes manufacturer, model, year, trim, engine, and assembly plant. Decode it automatically when a dealer uploads a vehicle, and never ask them to type year, make, model, and trim manually.

The NHTSA free API (vpic.nhtsa.dot.gov/api) decodes VINs and returns year, make, model, and trim. It is government-maintained, works for US vehicles, and costs nothing. Response time of 300-500ms is fine for an admin upload flow.

Paid VIN decode services like VinAudit or DataOne Software return richer data: engine specs, fuel type, transmission, MSRP, standard and optional equipment, and recall history. Pricing is $0.05-$0.20 per decode. At 500 uploads per month, that is $25-$100, well worth the enrichment.

The VIN also connects to vehicle history. Carfax and AutoCheck sell per-report access ($8-$15) or dealer subscription packages. Displaying the report link on the vehicle detail page is table stakes for buyer confidence.

Auto financing: why Stripe is the wrong answer

Dealertrack's 2024 State of Automotive Financing report found that 62% of car buyers apply for financing before visiting a dealership or clicking "buy." The financing integration is not a nice-to-have feature added after launch. It is the product.

Stripe handles payment processing. It does not originate auto loans. The mistake most teams make early: they see the checkout step and assume Stripe handles the full transaction. It handles the deposit. The $22,000 loan on a $25,000 car is originated by a lender and paid directly to the dealer.

Two lender networks cover most of the US market. Dealertrack (owned by Cox Automotive) connects dealers to 1,500+ lenders: national banks, regional credit unions, and captive finance companies like Honda Financial and Ford Motor Credit. RouteOne covers a similar network with slightly different lender composition. Integration with either requires a dealer agreement; if you are building for a dealer group that already has one, integration is straightforward. Multi-dealer platform operators need a software integrator agreement.

For buy-here-pay-here dealers who originate their own loans, the financing architecture is different. The dealer is the lender. You need a loan management system that tracks payment schedules and handles collections. This is a separate product category, not a feature on top of the standard platform.

Trade-in valuation and what moves the offer

The trade-in offer is the difference between a buyer who completes the purchase and one who abandons checkout to negotiate with a CarMax. Get it right.

Three valuation data sources exist. Black Book provides weekly auction-based wholesale values: what dealers actually pay at auction for a given year, make, model, trim, mileage, and condition. API access costs $5K-$20K per year. Kelley Blue Book is the consumer-facing standard, slightly higher than wholesale, more recognizable to buyers. NADA Guides covers commercial and RV markets well.

The calculation: query the API with VIN, mileage, and condition. Take the wholesale value. Apply a reconditioning estimate of $500-$1,500 for a standard used car. The trade-in offer is wholesale minus reconditioning. Display it instantly.

The offer needs an expiration. Seven days is standard. The physical vehicle gets picked up by the delivery driver when the new vehicle is delivered. The driver completes a condition assessment at pickup and flags any discrepancy between the submitted condition and actual condition. If the condition is worse than submitted, the offer adjusts. Build this adjustment logic before you launch a return policy, not after.

State-specific paperwork: the compliance problem

The paperwork layer is where many teams underestimate scope. Each US state has its own retail installment sale contract form, title transfer documents, and DMV filing requirements. Some states, like Arizona and Virginia, allow fully electronic titles. Most still require physical documents to be mailed.

DocuSign handles the electronic signature layer. The buyer receives a DocuSign envelope with the applicable documents for their state and signs on their phone or computer. Signed documents are stored and accessible from both buyer and dealer accounts.

Sourcing the correct forms is a legal problem. Work with an automotive compliance vendor: Darwin Automotive and Reynolds & Reynolds both offer document management services for dealers that handle per-state complexity. Budget 4-6 weeks of legal review per state before launch.

Start with 3-5 states where your inventory concentration is highest. Expanding to all 50 states is a compliance project, not a software project. The software configuration per state takes a day; the legal review takes weeks.

Delivery and the logistics layer

The buyer scheduled a delivery window. A driver loads the vehicle and drives to the buyer's address. This requires a driver-facing mobile app. Cross-platform mobile (a single React Native codebase for iOS and Android) saves $30K-$50K compared to building two native apps separately. We build cross-platform here unless there is a specific performance reason not to.

The driver app shows the day's delivery queue, navigation to the delivery address, a pre-delivery inspection checklist, and a digital signature capture for the delivery receipt. At delivery, the driver inspects the vehicle against the listing condition (photo documentation of current state), walks the buyer through the vehicle, and captures their signature.

If the buyer is trading in a vehicle, the driver takes custody of the trade-in and completes a condition assessment. The assessment photos go back to the platform to confirm whether the submitted condition was accurate.

The 7-day return adds a second logistics flow. This is not just a policy statement. The return requires a buyer-initiated return request in the app, a driver pickup assignment, a condition assessment at pickup, and a refund trigger that reverses the Stripe deposit and unwinds the financing. Build this as a first-class feature, not an afterthought. Teams that announce a return policy before building the return logistics flow create serious operational problems at the first return.

Driver location sharing between the driver app and the buyer-facing tracking page can run on Firebase Realtime Database. It is a lightweight, real-time solution that does not require building a custom WebSocket layer for V1.

What goes wrong in builds like this

Three failure modes appear consistently across digital automotive retail builds.

Dealertrack onboarding takes longer than expected. The API is well-documented. The dealer agreement process is not. Dealers need business verification and sometimes an in-person compliance review before the integration is live. Start the dealer agreement process in week one of the build. Teams that start it in week twelve face a 4-8 week delay that blocks the financing integration entirely.

State paperwork scope is discovered mid-build. A team builds the DocuSign flow assuming one generic contract, then discovers each state requires different forms with different fields and different signing order requirements. If you audit the states you plan to launch in before development starts, this is a 1-week planning exercise. Discovered mid-build, it is a 3-4 week delay plus rework.

The return policy is announced before the return logistics exist. Marketing puts "7-day return, no questions asked" in the headline. Engineering has not built the return pickup flow, the condition assessment workflow, or the refund trigger. The first return creates a manual operations emergency. Budget the return flow as a full feature block before you publish the policy.

How RaftLabs fits into this

We have scoped and built digital automotive retail platforms across dealer groups, fleet remarketing operations, and commercial vehicle marketplaces. The financing integration and delivery logistics are the two pieces where scoping conversations matter most. Both have dependencies (dealer agreements, state compliance review) that affect timeline in ways that are not obvious until you have been through the process.

If you are evaluating this build and want to understand the right architecture for your specific market, inventory size, and buyer type, book a 30-minute scoping call. We will tell you what is realistic, what is not, and where the timeline risks sit.