From Pain Point to MVP: I Built StatementSync in One Week

Originally published at chudi.dev

"I spend 10 hours a week just typing numbers from PDFs into spreadsheets."

That's what a freelance bookkeeper told me. She processes 50+ bank statements per month. Each one takes 10-15 minutes of manual transcription. The tedium is real.

I built StatementSync to fix this. One week from idea to production.

The Pain Point

Bookkeepers work with bank statements constantly. The workflow:

1. Client sends PDF bank statement
2. Open PDF, open spreadsheet
3. Manually type each transaction
4. Check for errors
5. Repeat 50+ times per month

The tools that exist either:

• Require proprietary software (QuickBooks, Xero)
• Charge per file ($0.25-1.00 per statement)
• Have terrible accuracy (OCR garbage)

For someone processing 50+ statements monthly, per-file pricing adds up fast. $25-50/month minimum, scaling with volume.

Validation Before Code

MicroSaaSBot's validation phase scored StatementSync before I wrote a single line of code:

Problems scoring below 60/100 get killed. No code written. This prevents building products nobody wants.

The validation phase confirmed:

• Real people have this problem
• They're already paying for solutions
• Current solutions have clear weaknesses to exploit

The Week

Day 1-2: Deep Validation

MicroSaaSBot's Researcher agent dug deeper:

• Competitive analysis (TextSoap, HappyFox, manual OCR tools)
• Pricing research ($0.25-1.00 per file is standard)
• Feature gap analysis (batch upload, bank-specific parsing)

Key insight: Flat-rate pricing would be a massive differentiator. Heavy users hate per-file fees.

Day 3: Architecture

MicroSaaSBot's Architect agent designed the system:

Frontend: Next.js 15 (App Router)
Auth: Clerk (handles signup, OAuth)
Database: Supabase PostgreSQL
Storage: Supabase Storage (PDFs, exports)
Payments: Stripe (subscriptions)
PDF Parsing: unpdf (serverless-compatible)
Hosting: Vercel

Critical decision: Pattern-based extraction instead of LLM inference.

LLM extraction would cost $0.01-0.05 per statement in API calls. Pattern-based extraction costs nothing at runtime. For a flat-rate product, this is the difference between profit and loss.

Day 4-6: Implementation

MicroSaaSBot's Developer agent built:

Day 4: Auth flow, database schema, file upload

// Prisma schema
model User {
  id                  String   @id @default(cuid())
  clerkId             String   @unique
  email               String
  subscriptionTier    Tier     @default(FREE)
  conversionsThisMonth Int     @default(0)
  lastResetAt         DateTime @default(now())
}

model Conversion {
  id              String   @id @default(cuid())
  userId          String
  originalFileName String
  status          Status   @default(PENDING)
  extractedData   Json?
  excelPath       String?
  csvPath         String?
}

Day 5: PDF parsing engine

async function extractTransactions(pdfBuffer: Buffer): Promise<Transaction[]> {
  const pdf = await getDocument({ data: pdfBuffer }).promise;
  const text = await extractText(pdf);

  // Pattern matching for supported banks
  const bank = detectBank(text);
  const parser = getParser(bank); // Chase, BofA, Wells, Citi, Capital One

  return parser.extract(text);
}

Day 6: Export generation, Stripe integration, dashboard

Day 7: Deployment

MicroSaaSBot's Deployer agent:

• Configured Vercel deployment
• Set up Supabase production
• Connected Stripe webhooks
• Ran smoke tests

Live by end of day.

The Technical Challenge

Problem: pdf-parse doesn't work on Vercel serverless.

pdf-parse has native dependencies that fail on Vercel's serverless runtime. I discovered this at 2 AM when the production build crashed.

Solution: Switch to unpdf.

unpdf is built for serverless from the ground up. No native dependencies, works perfectly on Vercel. The switch took 2 hours but saved the deployment.

If you're processing PDFs on Vercel, Netlify, or any serverless platform, use unpdf. Not pdf-parse. Save yourself the debugging.

The Product

StatementSync today:

Free Tier:

• 3 conversions/month
• Single file upload
• 7-day history

Pro Tier ($19/month):

• Unlimited conversions
• Batch upload (20 files)
• 90-day history
• Priority support

The $19/month flat rate is the differentiator. Process 50 statements? Same price. Process 200? Same price. Heavy users save money. Light users get simplicity.

Results

What I'd Do Differently

1. Start with one bank - Supporting 5 banks day one was overkill. Start with Chase (most common), add others based on demand.

2. Skip the dashboard MVP - Users just want to upload and download. The fancy dashboard came before proving the core value.

3. Launch before Day 7 - Could have deployed a working version by Day 5 and iterated publicly.

The Lesson

Building fast doesn't mean building sloppy. It means:

• Validate before you code - Kill bad ideas early
• Architecture matters - Pattern-based vs LLM extraction was the key decision
• Launch before perfect - Iteration beats planning

MicroSaaSBot compressed weeks of work into days by handling the tedious parts automatically. I focused on the decisions that mattered.

StatementSync is proof that AI-assisted development can ship real products, not just demos.

Related: Serverless PDF Processing: unpdf vs pdf-parse | Portfolio: StatementSync