sgkstudio.
AI Automation

AI Workflow Automation
get out of repetitive work

Spreadsheet cleanup, ERP entry, sending emails, document review — hiring someone to handle this kind of repetitive work costs ₩33M+ a year (median of job postings, 410 postings). Automation takes over that role, and the build starts at ₩3.5M.

Tell us the work you want automated and we will sort what is feasible from what isn't, for free, in writing. No obligation to proceed.

Simple repetitive-task RPA

Work that macro-level automation already covers

We automate rule-based repetitive tasks — spreadsheet processing, web scraping, sending emails.

  • Replacing Excel VBA / Python scripts
  • Automated web data collection and monitoring
  • Automated email and messaging notifications

AI workflow automation

Work that needs document classification and data extraction

We connect document classification → data extraction → DB entry → email sending into one pipeline on n8n / Dify.

  • Document OCR plus LLM data extraction
  • Integration across multiple systems (ERP, email, spreadsheets)
  • Only the steps that need a decision go to a person

LLM agent + RPA integration

Work that needs both judgment and execution

An LLM reads context and makes the call, RPA executes it. This covers multi-step processes too.

  • Classifying, responding to, and escalating customer inquiries
  • Drafting and reviewing quotes and contracts automatically
  • Complex work spanning multiple systems
What you can check

We left it open so you can check, instead of asking you to take our word

The hardest part of choosing an automation vendor is knowing whether the people can actually do it. So we publish what we built and ran. All three below are things you can open and count yourself.

9articles written and shipped automatically

Choosing the topic, drafting, charts, checks, and deployment all run in one pass at 02:20 every night. On days a draft fails the quality bar nothing ships; on days it passes, several do. It started on 1 September and stands at 9.

Count the articles
8deliverables produced by running it

Quote sending, auto-replies to enquiries, CV screening, and leave requests were actually run, and the resulting logs and screens are published as they came out. We opened them so you can judge the output before signing anything.

See the output
the commandhow each number is counted, alongside it

Commit counts, running automations, and decision records are shown together with the command that counts them. Run it yourself and you get the same number. That is the difference from printing a big figure with no source.

See how it is counted

The 8 deliverables were built against fictional companies (an interior firm, a cafe). They are separate from the automation that actually runs our own operations, and anything we deliver is rebuilt for your environment.

Tell us the work you want automated and we will sort what is feasible from what isn't, in writing
Scope

What we automate

We gather scattered inputs, let AI judge them, and write the result back into your systems. Four engineering standards turn the work from a 'demo' into a 'running system', and all four are included in what we deliver.

01 · Connect

Gather scattered inputs

We bring inputs arriving through different channels into one place and normalize them.

  • Email, messaging, web forms
  • Excel, Google Sheets, ERP/CRM
  • PDF and image documents (OCR)
02 · Process

AI understands and decides

We extract the needed values from unstructured material and answer from evidence.

  • Extraction and classification (unstructured → structured)
  • RAG answers grounded in your documents
  • Confidence gate — if unsure, hand off
03 · Execute

Write back into your systems

The decision lands in the actual workflow systems and triggers notifications.

  • Automatic entry into ERP, CRM, spreadsheets
  • Auto replies and draft generation
  • Reports, dashboards, alerts
Design

Exception map

The points that need human judgment are defined during design, before anything is built.

Verify

Quantified acceptance

Pass/fail is judged by the test pass rate agreed before work starts, not by feel.

Operate

Management console

A screen for non-developers to correct wrong answers and update knowledge ships with the build.

Report

Monthly quality report

Response and processing quality is tracked numerically and reported every month.

Process

From contract to delivery, usually 2–6 weeks

Each stage's output is the starting condition for the next. Acceptance is judged by the pass rate on test cases agreed before we start, not by feel.

  1. 01Free assessmentTell us what you want automated through the online assessment tool or a short email, and we will separate what's feasible from what isn't, in writing. No obligation to proceed.Free
  2. 02Work analysisWe observe the actual workflow to fix scope and exception handling (the points that need human judgment) in a written document.1–3 days
  3. 03ContractFixed scope, an explicit out-of-scope list, and quantified acceptance criteria (pass rate on test cases, response time) are locked into the contract.half upfront
  4. 04BuildWe onboard your data and business rules onto our standard core, checking direction against interim demos.5–15 days
  5. 05Acceptance (UAT)Acceptance requires passing tests built from real inquiries and real workflow cases. The pass record is kept in writing.half balance
  6. 06HandoverSource code handed over via GitHub, plus an operating manual for non-developers, a walkthrough video, and one training session.1–2 days
  7. 07OperationMonitoring, knowledge updates, prompt tuning, and a monthly performance report — no minimum commitment, cancel any time.Monthly

The finished system's full source code is handed over to your GitHub — you are not locked into us. LLM API and server accounts are set up under your name from the start.

Pricing

Pricing

Every project is a one-time build fee plus a monthly maintenance fee. The build fee varies with scope and complexity; maintenance covers ongoing operation, monitoring, and stabilization after the build ships.

One-time build fee

Covers the full process — assessment, design, development, testing, and delivery. Priced by scope and complexity.

Monthly maintenance fee

Covers server, LLM API, and vector DB infrastructure operation, stabilization, and incident response. Contracted separately after the build ships.

We provide an exact quote through a free consultation.

Need workflow automation?

Send us the workflow and we will assess whether it's automatable for free, in writing. Every consultation runs remotely over email.

Contact: contact@sgkstudio.co.kr