COSAP: An affordable SAP alternative for Korean SMBs.

Role: Primary PM · cross-market requirements (Korea + Singapore)
Company: SkyLab Group
Period: 2025
Stack: Multi-tenant SaaS, Toss Payments, Lago, RAG pipelines

Core insight

The bet

Small and mid-sized companies rarely lack software. They lack something that catches the error before it posts. COSAP is built around that idea: an AI-powered for Korean , positioned as an affordable alternative to SAP, covering accounting, legal compliance and payroll in one multi-tenant product.

Thao owned

modules I led product design across: 8
client markets: Korea and Singapore: 2

Source: Thao's CV (SkyLab, COSAP)

Why it matters

The problem

The company's launch material frames three leaks. Finance: closing the books takes 10+ days a month. Payroll: about 1 in 5 runs contains an error. Leadership: internal fraud can take a year to notice. Those are industry benchmarks the company cites (EY, APQC, ACFE), not measurements of mine.

Decision

The tension

SMBs will not rip out the ERP, accounting and HR tools they already run, and a new system means weeks of training. So the product stance became a layer above those systems, with Telegram as the everyday surface. Toggle the two pillars below.

The decision

Replace the company's ERP, or sit above it?

Chosen: A layer above existing systems

Sit above those tools and catch errors before they post, with Telegram as the everyday surface.

Source: COSAP company introduction deck v7, p.4 and launch deck v3, p.8 (confidential decks; paraphrased).

Regulated market

Compliance before the save

Korean and Singaporean clients carry different labor rules. Rather than audit policies afterwards, the product maps each employee to a country and blocks a non-compliant policy at the moment it is written, citing the law. Try it.

Cross-market

One work item, two languages

A Vietnamese employee justifies a Korean card expense in Vietnamese; the Seoul approver reads it in English. The company's line is that most apps translate the menu, while COSAP translates the work.

Two phones showing the same expense, one in Vietnamese and one in English

COSAP unified portal inside Telegram showing pending approvals and modules

Company figures

What it did in the field

Two measurements the company reports. They describe the product at client sites, not my personal outcome, and the client is not named.

Forge at a manufacturing client: before vs after

Company figure

Before ForgeWith Forge

Purchase-order errors per month12.4−83%
Time to flag an anomaly14.3 days
On-time delivery73%
Inventory turnover4.2×

Each row is scaled to its own larger value, so rows in days, minutes and percent can sit side by side. 82% of error-prone purchase orders were blocked automatically.Source: COSAP company introduction deck v7, p.18 (company figure, reconciled by COSAP against ERP and audit logs). Client name withheld.

Month-end close (hours)

Company figure

72 hBefore (3 days)

4 hWith COSAP

-94.4%

From about three days to about four hours, benchmarked across 12 SMB deployments in Q1 2026.Source: COSAP company introduction deck v3, p.14 (company figure); v7, p.8 confirms the 3-day baseline.

The public site also advertises 60% less review time, risk detection within 24 hours and 40% fewer inquiries. Those are marketing claims from cosap.ai.

Ownership

What was mine

Mine
Primary PM for cross-market requirement gathering across Korean and Singaporean clients, and product design across 8 modules.
The team's
The AI engine, integrations and Forge deployments were built by the engineering team and its industry partners.
The company's
The figures above, the launch decks and the public site. I cite them; I do not claim them.

Results

purchase-order errors per month at a manufacturing clientCompany figure: −83%
month-end close, from about three daysCompany figure: ~4 h
modules I led product design acrossMy CV: 8

Source: COSAP company introduction deck v7, p.18 and launch deck v3, p.14 (company figures, client name withheld); Thao's CV (SkyLab, COSAP). The company figures describe the product at client sites, not my personal outcome.