Introduction

Buying a used car from a private seller is one of the most stressful financial decisions most people make. You don't have time to become a mechanic. You don't have the expertise to spot hidden problems. And you definitely don't have the budget to hire an engineer to inspect every car you're interested in.

The result? Buyers either overpay for cars with hidden mechanical issues, or they walk away from good deals because they're too uncertain.

We're building CarInspect—a mobile app that puts a pocket mechanic in your phone. Using real-time audio analysis, model-specific checklists, and crowdsourced training data, we're making it possible for anyone to confidently inspect a used car before they buy.

The Problem: Why Private Car Buying is Broken

The Knowledge Gap

  • 85% of private car buyers have no mechanical knowledge (source: Swedish Consumer Agency)
  • Professional pre-purchase inspections cost 2,000-5,000 SEK and require scheduling
  • Many buyers skip inspection entirely, hoping for the best

The Trust Gap

  • Sellers may not disclose known issues (intentionally or unintentionally)
  • Buyers have no way to verify claims like "engine runs great"
  • Hidden costs (timing belt replacement, gearbox fluid leaks) can cost 10,000-50,000 SEK after purchase

The Time Gap

  • Finding a car, scheduling viewings, and getting inspections takes weeks
  • Buyers need to make quick decisions or lose the car to another buyer
  • No time for professional assessment

The Market Opportunity

  • Sweden: ~500,000 private car sales annually
  • Europe: ~15 million private car sales annually
  • Average car price: 150,000 SEK (Sweden)
  • Market size: 75 billion SEK (Sweden alone)

The Solution: CarInspect

Core Concept

CarInspect is a mobile app that combines three powerful tools:

  1. Real-Time Audio Analysis – Detect engine knock and other mechanical issues instantly
  2. Model-Specific Checklists – Know exactly what to look for on your specific car
  3. Crowdsourced Training Data – Continuously improve accuracy with user contributions

How It Works: Carl's Story

Carl is a 35-year-old software engineer. He found a 2010 Volvo V40 on Blocket.se that fits his budget. But he's nervous about hidden mechanical issues. He downloads CarInspect.

Step 1: Car Selection
Carl opens the app and enters: 2010 → Volvo → V40. The app instantly displays a checklist of common issues for this model:

  • Check timing belt (critical for this year)
  • Inspect gearbox fluid (known issue)
  • Listen for valve noise
  • Check for rust spots

Step 2: Sound Analysis
At the seller's location, Carl positions his phone near the engine bay and records a 45-second audio clip. The app captures real-time waveform data and uploads it to our server.

Within 10 seconds, the server analyzes the audio and returns:

  • ✅ Engine Status: 🔴 WARNING
  • ✅ Finding: Engine knock detected (78% confidence)
  • ✅ Recommendation: "This engine shows signs of pre-detonation. We recommend professional inspection before purchase."

Step 3: Inspection Checklist
Carl completes the checklist:

  • ✅ Timing belt looks worn
  • ✅ Gearbox fluid is dark (might need change)
  • ✅ No obvious rust spots
  • ✅ Engine knock detected by app

Step 4: Report Generation
Carl exports a PDF report with all findings. He shares it with his mechanic friend, who confirms: "Yeah, that knock is real. Budget 15,000 SEK for timing belt replacement."

Result: Carl negotiates the price down by 15,000 SEK. He buys the car with realistic expectations and avoids overpaying.

Key Features

1. Sound Analyzer (Engine Bay)

  • Real-time audio capture with waveform visualization
  • Server-side ML analysis for knock detection and anomalies
  • Confidence scoring (e.g., "78% confident engine knock detected")
  • Plain-language results (no technical jargon)
  • Metadata capture: Phone model, GPS, timestamp, engine RPM (future)

2. Car Model Database

  • 20,000+ car models with known issues and common problems
  • Year-specific guidance (e.g., "2010 V40 has timing belt issues")
  • Visual checklists with photos and diagrams
  • Severity indicators (Critical / Important / Minor)
  • Offline access (checklists work without internet)

3. Inspection Profile

  • Consolidated report with all findings
  • Photo attachments from checklist items
  • User notes and observations
  • Condition rating (1-5 stars, auto-calculated)
  • Export options: PDF, email, cloud sync

4. Sound Library & Database

  • Crowdsourced audio samples tagged by car model and condition
  • Side-by-side comparison: Healthy vs. problematic engines
  • Community ratings ("Is this knock?" voting)
  • ML training data for continuous improvement
  • Privacy-first: Anonymized, opt-in contributions

5. User Accounts & Sync

  • Inspection history across devices
  • Saved favorites (car models for quick access)
  • Cloud backup of reports
  • Optional login (works without account initially)

Business Model

B2C (Consumer) - Primary Revenue

  • Freemium model:
    • Free: 3 inspections/month, basic features
    • Premium: Unlimited inspections, advanced reports, priority support
    • Pricing: 99 SEK/month or 799 SEK/year
  • Target: 100,000 users in Year 1, 500,000 by Year 3
  • Conversion: 5-10% to premium
  • Revenue Year 1: ~5-10 million SEK

B2B (Dealership) - Secondary Revenue

  • Dealership licenses: 5,000-10,000 SEK/month per dealership
  • Features: Branded reports, bulk inspections, API access
  • Value prop: Certify used cars, build buyer trust, reduce returns
  • Target: 100 dealerships by Year 2
  • Revenue Year 2: ~10-15 million SEK

Data Monetization (Future)

  • Anonymized audio library licensing to automotive manufacturers
  • Predictive maintenance insights for insurance companies
  • Market data (car condition trends by model/region)

Technical Architecture

Mobile App (iOS/Android)

  • Framework: React Native or Flutter
  • Audio capture: Native microphone API
  • Local preprocessing: Noise reduction, compression
  • Cloud sync: Firebase or custom backend

Backend Infrastructure

  • API Server: Node.js/Python (FastAPI)
  • Audio Processing: Python (librosa, scipy)
  • ML Model: TensorFlow/PyTorch for knock detection
  • Database: PostgreSQL (metadata), S3 (audio files)
  • Deployment: AWS or Google Cloud

ML Pipeline

  • Audio preprocessing: Spectral analysis, MFCC extraction
  • Knock detection classifier: CNN or XGBoost
  • Training data: 1000+ labeled samples per car model
  • Accuracy target: 85%+ vs. professional mechanic assessment

Security & Privacy

  • HTTPS/TLS for all data in transit
  • Encryption at rest for audio files
  • GDPR compliance: Data deletion, consent management
  • Anonymization: No personal identifiers in audio

Development Roadmap

Phase 1: MVP (Months 1-6)

Goal: Proof of concept with engine knock detection

Deliverables:

  • Mobile app (iOS + Android)
  • Audio capture & preprocessing
  • Basic ML model for knock detection
  • 10 car models in database
  • User authentication
  • Basic reporting

Team: 2 mobile engineers, 1 ML engineer, 1 backend engineer, 1 product manager

Cost: ~2.5 million SEK

Timeline:

  • Month 1-2: App architecture, audio pipeline setup
  • Month 2-3: ML model training (using public datasets)
  • Month 3-4: Integration, testing
  • Month 4-5: Beta testing with 100 users
  • Month 5-6: Launch on App Store/Play Store

Phase 2: Expansion (Months 7-12)

Goal: Scale to 10,000 users, expand car model database

Deliverables:

  • 500+ car model

Appendix A: Technical Deep Dive

Audio Processing Pipeline

Step 1: Capture

  • User records 45-60 second audio clip via phone microphone
  • Real-time waveform visualization shows audio is being captured
  • Metadata captured: timestamp, GPS location, phone model

Step 2: Preprocessing

  • Noise reduction (spectral subtraction to remove background noise)
  • Normalization (peak or RMS normalization to 0dB)
  • Resampling to 16 kHz (standard for audio ML)
  • Segmentation into 1-second windows for analysis

Step 3: Feature Extraction

  • MFCC (Mel-Frequency Cepstral Coefficients): 13-40 coefficients capturing human-perceived frequency
  • Spectral features: Centroid, rolloff, bandwidth (identify frequency content)
  • Temporal features: Zero-crossing rate, RMS energy (detect transients like knock)
  • Onset detection: Identify sudden changes (characteristic of knock)

Step 4: ML Model Inference

  • Input: Feature vector (50-100 dimensions)
  • Model: CNN (Convolutional Neural Network) trained on 1000+ labeled samples
  • Output: Knock probability (0-1) + confidence score

Step 5: Post-Processing

  • Temporal smoothing (reduce false positives from single frames)
  • Confidence adjustment based on model uncertainty
  • Result formatting: "Knock detected (78% confidence)"

ML Model Architecture

Input (1, 50)
      ↓
    Conv1D(32 filters, kernel=3) → ReLU → MaxPool(2)
      ↓
    Conv1D(64 filters, kernel=3) → ReLU → MaxPool(2)
      ↓
    Flatten
      ↓
    Dense(128) → ReLU → Dropout(0.3)
      ↓
    Dense(1) → Sigmoid
      ↓
    Output: Knock probability (0-1)

Training Details:

  • Dataset: 1000+ labeled audio samples (healthy vs. knock)
  • Split: 70% train, 15% validation, 15% test
  • Optimizer: Adam (learning rate 0.001)
  • Loss: Binary cross-entropy
  • Epochs: 50
  • Batch size: 32

Target Metrics:

  • Precision: 90%+ (minimize false positives)
  • Recall: 80%+ (catch real knocks)
  • F1 Score: 85%+
  • AUC: 0.90+

Data Privacy & Security

Audio Data Handling:

  • Audio files are encrypted in transit (HTTPS/TLS)
  • Stored encrypted at rest (AES-256)
  • User can delete their audio anytime
  • Opt-in sharing for ML training (not mandatory)
  • No personal identifiers stored with audio

User Data:

  • GDPR compliant (right to deletion, data portability)
  • Clear privacy policy (what data we collect, how we use it)
  • No third-party data sharing without consent
  • Regular security audits

Appendix B: Competitive Analysis

Market Competitors

Competitor Offering Limitations
Blocket.seCar listings + basic inspection toolsLimited to listings, no audio analysis
Professional mechanicsFull inspection (2,000-5,000 SEK)Expensive, slow, requires scheduling
OBD-II scanner appsEngine error codesOnly shows codes, no audio analysis
YouTube tutorialsDIY inspection guidesRequires expertise, time-consuming
Insurance companiesPre-purchase inspection (some)Limited availability, expensive

Our Competitive Advantages

  1. Audio Analysis: Unique technology for detecting knock and anomalies
  2. Accessibility: Designed for non-technical users (plain language, simple UI)
  3. Speed: Results in 10 seconds vs. hours for professional inspection
  4. Cost: Free/cheap vs. 2,000-5,000 SEK for professional
  5. Crowdsourced Data: Improves over time with user contributions
  6. Model-Specific: Tailored guidance for each car's known issues
  7. Privacy-First: Opt-in data sharing, anonymized

Appendix C: FAQ

For Investors

Q: Why audio analysis? Why not just use OBD-II diagnostics?
A: OBD-II only shows error codes (which may not be set yet). Audio analysis detects mechanical issues before they trigger error codes. Combined, they're powerful. We'll integrate OBD-II in Phase 3.

Q: What's your moat? Can't competitors just copy this?
A: Our moat is data. The more users we have, the more audio samples we collect, the better our model becomes. This creates a virtuous cycle that's hard to replicate. Plus, we're building brand trust early.

Q: What's your path to profitability?
A: B2C freemium (99 SEK/month premium) + B2B dealership licensing (5,000-10,000 SEK/month). We break even in Month 18-20 and hit 181.5M SEK net profit by Year 3 (conservative).

Q: Who are your customers?
A: Primary: Private car buyers (non-technical, want confidence). Secondary: Dealerships (want to certify cars, build trust). Tertiary: Car enthusiasts (want to know their car's health).

Q: What's your go-to-market strategy?
A: Launch with early adopters (car forums, Reddit), then scale via app store optimization and partnerships with Blocket.se, insurance companies, and dealerships.

Q: What are the risks?
A: Model accuracy (mitigated by training on diverse data), user adoption (mitigated by strong product-market fit), competition (mitigated by data moat), regulatory (GDPR compliant from day 1).

For Users

Q: Is this app a substitute for a professional inspection?
A: No. This app is a screening tool to help you make informed decisions. For major purchases, always get a professional inspection.

Q: How accurate is the knock detection?
A: Our target is 85%+ accuracy vs. professional mechanic assessment. We're training on diverse data and continuously improving.

Q: What if the app says my car is fine but it has problems?
A: This is possible (false negative). That's why we recommend professional inspection for cars with high mileage or unknown history. The app is a first-pass screening tool.

Q: Can I use this app offline?
A: Checklists work offline. Audio analysis requires internet (to send audio to our server for processing).

Q: What happens to my audio data?
A: By default, your audio is deleted after analysis. You can opt-in to share it (anonymized) to help train our model. You can delete your data anytime.

Q: How much does it cost?
A: Free tier: 3 inspections/month. Premium: 99 SEK/month or 799 SEK/year for unlimited inspections.

Q: Which cars does the app support?
A: MVP launches with 10 popular models (Volvo V40, VW Golf, etc.). We'll expand to 500+ models by Month 12 and 20,000+ by Year 2.

Q: Can I use this for my own car?
A: Yes! Many users will use this to monitor their car's health over time.

For Dealerships

Q: How does the dealership program work?
A: You get a dealership license (5,000-10,000 SEK/month), branded reports, bulk inspection tools, and API access. You can certify your used cars and build buyer trust.

Q: Can I white-label the app?
A: Not in MVP, but we're open to partnerships in Phase 2+.

Q: What's the ROI for dealerships?
A: Reduced returns (fewer buyers discovering hidden issues after purchase), faster sales (buyers feel confident), premium pricing (certified cars sell for more).

Appendix D: Success Metrics & KPIs

User Metrics

  • Downloads: 50,000 (Year 1) → 500,000 (Year 3)
  • Active Users: 10,000 (Year 1) → 150,000 (Year 3)
  • Retention: 40%+ monthly active users
  • Engagement: 70% of users complete full inspection

Product Metrics

  • Knock detection accuracy: 85%+ vs. professional assessment
  • Analysis time: <10 seconds (end-to-end)
  • App uptime: 99.5%+
  • Audio library size: 1000+ samples per car model

Business Metrics

  • Premium conversion: 5% (Year 1) → 10% (Year 3)
  • Premium ARPU: 99 SEK/month
  • Dealership adoption: 10 (Year 1) → 100 (Year 2)
  • Customer acquisition cost: <50 SEK
  • Lifetime value: >500 SEK

Financial Metrics

  • Break-even: Month 18-20
  • Year 1 revenue: 5.9M SEK (conservative)
  • Year 3 revenue: 193.5M SEK (conservative)
  • Gross margin: 70%+ (software business)
  • Net margin: 40%+ (Year 3)

Appendix E: Regulatory & Compliance

GDPR Compliance

  • Data collection: Explicit consent for audio sharing
  • Data retention: Users can delete data anytime
  • Data portability: Users can export their inspection history
  • Privacy policy: Clear, transparent, updated regularly

Liability & Disclaimers

  • Clear disclaimer: "This app is not a substitute for professional inspection"
  • Liability waiver: We're not responsible for missed issues
  • Insurance: Professional liability insurance (E&O)
  • Terms of service: Clear terms covering usage, limitations, liability

Audio Rights & Licensing

  • User grants license: To use audio for ML training (opt-in)
  • Anonymization: No personal identifiers in audio
  • Revocation: Users can revoke sharing anytime
  • Attribution: Users credited in "Sound Library" (optional)

Appendix F: Roadmap Details

Phase 3: Scale (Months 13-24)

  • Months 13-15: OBD-II Integration
  • OBD-II scanner integration
  • Real-time engine diagnostics
  • Correlation with audio analysis
  • Expanded anomaly detection
  • Months 15-18: Marketplace Integration
  • Blocket.se API integration
  • AutoTrader integration
  • One-click inspection for listings
  • Price recommendations
  • Months 18-21: International Expansion
  • UK market entry
  • German market entry
  • Norwegian market entry
  • Localization (language, car models, regulations)
  • Months 21-24: Advanced Features
  • Predictive maintenance (estimate time to failure)
  • Insurance partnerships
  • Dealership dashboard v2
  • API for third-party integrations

Appendix G: Team & Hiring Plan

Current Team

  • [Your Name]: Product & Strategy
  • [Co-founder]: Engineering Lead

Year 1 Hiring

  • 2 Mobile Engineers (iOS + Android)
  • 1 Backend Engineer
  • 1 ML Engineer
  • 1 QA Engineer
  • 1 Product Manager (part-time initially)
  • 1 UX/UI Designer (part-time initially)

Year 2 Hiring

  • 1 Data Engineer (for database expansion)
  • 1 Business Development (dealership partnerships)
  • 1 Marketing Manager
  • 1 Customer Support Lead

Advisors

  • Automotive Expert: Former mechanic, 20+ years experience
  • ML Expert: PhD in audio processing, published researcher
  • Business Advisor: Former CEO of automotive company

Appendix H: Exit Strategy

Potential Acquirers

  1. Blocket.se / Schibsted: Automotive marketplace, natural fit
  2. AutoTrader / Cox Automotive: Used car marketplace
  3. Insurance companies: Allianz, If, Folksam (predictive maintenance data)
  4. Automotive OEMs: Volvo, Scania (training data, predictive maintenance)
  5. Fintech companies: Klarna, Tink (financing integration)

Exit Timeline

  • Year 2-3: Acquisition likely (if growth targets met)
  • Valuation: 500M-2B SEK (based on revenue multiples)
  • Return: 80-320x for early investors

Contact & Next Steps

Interested in learning more?

  • Email: [your-email@example.com]
  • LinkedIn: [your-profile]
  • Website: [coming soon]
  • Demo: [link to prototype or video]

Let's build the future of used car buying together.

CarInspect: Your Pocket Mechanic for Confident Car Buying