Building BANKTRUST: Trust-First Bank Statement Exports for Accounting Firms

In accounting, speed is useful — but trust is non-negotiable.
If the numbers don’t reconcile, nothing else matters.

BANKTRUST started from a simple but uncomfortable question:

“Why are accountants still retyping bank statements in 2025?”

Not because better tools don’t exist —
but because most automation tools ask users to trust results they can’t verify.

BANKTRUST is our answer to that gap.

This is the story of how we’re building a trust-first bank-statement-to-ledger pipeline for bookkeeping and accounting firms handling multiple clients every month.

The Problem: Automation Without Confidence

For many accounting teams, monthly bank statements are still painful:

• PDFs arrive in different formats, layouts, and quality
• OCR tools extract data, but:
  • totals don’t always match
  • balances drift
  • edge cases silently fail
• The result is a dangerous tradeoff:
  • Trust the machine blindly
  • or retype everything by hand

Neither option scales.

On the firm side, this shows up as:

• Hours lost reconciling “almost right” exports
• Senior staff double-checking junior work
• CSV files that look clean but can’t be trusted without manual review

The core issue isn’t parsing.
It’s confidence.

The Core Idea: Trust Must Be Visible, Not Assumed

Early on, we made one foundational decision:

BANKTRUST would never hide uncertainty.

Instead of pretending every parse is perfect, we designed the system to:

• Show what it’s confident about
• Flag what needs attention
• Prove that totals reconcile — or clearly show when they don’t

In other words:

Trust isn’t a claim. It’s a measurable output.

That idea shaped everything that followed.

Building the Backbone: Statements, Transactions, and Trust Metrics

We started with the boring, essential questions:

• Can we always reconcile opening and closing balances?
• Can we prove where every number came from?
• Can an accountant review results without leaving the tool?

1. Statement-Centric Ingestion

In BANKTRUST, everything starts with a statement, not raw rows.

Each uploaded PDF produces:

• A statement record with:
  • start date
  • end date
  • currency
  • opening & closing balances
• A full transaction list
• A calculated variance (amount + percent)
• A confidence score that reflects parse quality

If the math doesn’t close, the system doesn’t pretend it does.

2. Trust Signals, Not Black Boxes

Instead of “success / failure,” each statement surfaces:

• Confidence percentage (e.g. 98.5%)
• Variance indicators
• Anomaly flags at transaction level:
  • balance gaps
  • suspect parses
  • outlier amounts

This lets accountants answer the real question:

“Can I rely on this export — and if not, where should I look?”

The API: Built for Firms Who Automate Everything

BANKTRUST isn’t just a UI tool.
It’s designed to slot into existing workflows.

So we exposed a clean, minimal API:

• POST /api/v1/statements
• Auth via per-firm X-API-Key
• Multipart upload:
  • PDF file
  • client_id

The response mirrors the UI:

• statement summary
• confidence + variance
• full transaction list with anomaly metadata

No “magic success.”
Just structured data firms can build around.

API documentation is public at:

👉 https://banktrustapp.com/docs/api

Shipping the Product Surface: Landing, Docs, and Early Access

Once the core system was working, we focused on making BANKTRUST presentable and honest.

What we shipped:

• Public landing page
  • Clear positioning: ledger-ready exports you can trust
  • No pricing pressure — early access only
• Early-access waitlist
  • Email-only, manually curated cohort
• API reference docs
  • Concrete examples, not marketing fluff
• Product walkthrough video
  • Short, silent screen demo showing:
    • upload → trust view → export

All deployed on a lean stack:

• Next.js (App Router) on Vercel
• Supabase for auth, data, and storage
• Minimal dependencies, maximum clarity

What We Learned (So Far)

A few lessons reinforced themselves quickly:

1. Accountants don’t want “smart” — they want explainable
   Confidence beats cleverness every time.

2. Variance is not an error — it’s information
   Surfacing mismatch early builds trust faster than hiding it.

3. Good UI doesn’t replace good math
   The product only works because reconciliation is first-class.

4. Early access works best when it’s human
   We deliberately chose manual onboarding to learn from real firms.

What’s Next for BANKTRUST

BANKTRUST is live, but early.

Next steps include:

• Parser upgrades
  • More layouts
  • Better edge-case handling
• Google Sheets export
  • OAuth-based, one-click delivery
• API key management UI
  • Create, rotate, revoke keys in-app
• Clear pricing tiers
  • Once usage patterns are real, not hypothetical

All guided by the same rule:

If trust drops, the feature isn’t done.

Why This Matters to RUKMAYA

BANKTRUST reflects how we like to build at RUKMAYA:

• Start from real operational pain
• Design around human verification, not blind automation
• Keep the stack lean so the workflow stays legible

If you’re building in fintech, accounting, or any system where numbers matter, the lesson is simple:

Don’t just automate the output.
Design for the moment someone asks, “Can I trust this?”

• 🟡 See how we build trust-first products
• 🟢 Visit BANKTRUST
• 🔵 Talk to us about your own workflow-heavy product

— The RUKMAYA Team

Share this post