The Document Scan That Unlocked Paywork's Core Features

The Document Scan That Unlocked Paywork's Core Features

How I replaced manual input with OCR to increase verification completion by 1.5x

How I replaced manual input with OCR to increase verification completion by 1.5x

Shipped

APP/WEB

B2C SaaS

Timeline

Jun 2023 - July 2023

Jun 2023 - July 2023

My Role

UX Research

User Testing

Amplitude Analysis

UX Research

User Testing

Amplitude Analysis

Team

1 Product Designer(Me)

1 Product Owner

3 Front-end Developers

2 Back-end Developers

1 Product Designer(Me)

1 Product Owner

3 Front-end Developers

2 Back-end Developers

Tool

Figma

Amplitude

Jira

Confluence

Figma

Amplitude

Jira

Confluence

Overview

Paywork is a business document management platform for small business owners in Korea. Users can create and manage essential business documents in one place, including estimates, transaction statements, and tax invoices.


To access core paid features like tax invoice issuance, users must first complete business authentication. However, the manual input required during this process caused a significant drop-off before users could reach those features. This project introduced OCR to automate data entry in the authentication flow, reducing friction and increasing the number of users who could successfully complete verification.

Key Objectives

Amplitude analysis revealed an overall authentication completion rate of just 41.4%.

The steepest drop-off occurred at the business information input step, where users had to manually fill in 9 fields across two screens. Six of those fields (business name, registration number, representative name, address, business type, and category) were already printed on their business registration certificate.


Re-entering information they already had in hand was identified as the primary cause of drop-off.

Existing Paywork Business Authentication Flow

Existing Paywork Business Authentication Flow

Amplitude Business Authentication Funnel Analysis

Amplitude Business Authentication Funnel Analysis

Amplitude Business Authentication Funnel Analysis

May 2023 ~ June 2023

May 2023 ~ June 2023

Strategy

The goal was to eliminate manual re-entry entirely.

Round 1. Third-party business registry API (Rejected)

I first explored an API that auto-fills fields from a registration number alone.

Testing revealed two blockers: sole proprietor data (the majority of Paywork’s user base) was incomplete due to the API's reliance on credit lookup systems, and the per-query pricing model added unsustainable cost at scale. we ruled it out.

I first explored an API that auto-fills fields from a registration number alone.

Testing revealed two blockers: sole proprietor data (the majority of Paywork’s user base) was incomplete due to the API's reliance on credit lookup systems, and the per-query pricing model added unsustainable cost at scale. we ruled it out.

Round 2. OCR (Selected)

OCR resolved both issues. It extracts data directly from the user’s uploaded certificate image, so sole proprietor fields are never missing and there is no per-query cost. It was the option that satisfied both accuracy and cost requirements.

Design Solution

I designed an OCR authentication UX flow where uploading a business registration certificate auto-fills all required fields in one step.


To account for recognition errors, I added a review and correction step so users can verify and edit extracted information before submitting. Each screen was restructured around a single action to reduce perceived effort at every step.

Impact

Paywork's core users are small business owners in their 50s and 60s who needed to send invoices quickly.


For this group, having to manually type critical business information was not just friction, it was a real barrier before they could use any paid feature. By introducing OCR to remove that barrier, we saw the following results.

1.5x

increase in verification completion rate

41.4% → 62.8%, measured over one month post-launch using Amplitude (July to August 2023).

5x

faster completion time

Average completion time dropped from 2m 42s to 30s, measured using Amplitude over the same period.

THANK YOU
FOR VIEWING

THANK YOU
FOR VIEWING

THANK YOU
FOR VIEWING