WEBHOOK / FORMCF7 · n8n triggerLLM CLASSIFICATIONClaude API · reply genGMAIL DRAFTGAS · draft savedSLACK NOTIFYchannel: #inquiryFig. 01.A — system schematic

— No. 01 / Automation

Inquiry Automation Pipeline

inquiry-automation · 2024 — present

受信した問い合わせをLLMで分類・返信文を自動生成し、Gmail下書き保存とSlack通知まで一気通貫で自動化。GASとn8nの2パターンで実装し、要件に応じた使い分けを提案できるよう設計した。An end-to-end pipeline that classifies inbound inquiries, drafts a reply with an LLM, saves it to Gmail, and notifies the team in Slack. Built in both GAS and n8n to demonstrate flexible tooling decisions.

Type業務自動化ワークフローOperational automation workflow
Role設計 · 実装 · 運用Design · Implementation · Ops
StackClaude API · Google Apps Script · n8n · Slack API · Gmail API
Latency≈ 3 — 5 sec / inquiry
Status本番稼働中In production
Claude APIGASn8nSlackGmail
01 / Context

Background

中小企業のBtoB問い合わせは1日10〜30件。担当者が1件ずつ分類し、過去返信を探し、文面を整える時間が積み上がる。定型対応に費やす時間を削り、担当者は判断と提案に集中する ─ そのための仕組みとして設計。SMB BtoB teams handle 10–30 inquiries per day. Most of the time goes into triage, finding past replies, and polishing wording. Reduce the routine work so the team can focus on judgement and proposals — that was the brief.

02 / Pipeline

Pipeline stages

N° 01受信フックの正規化Normalize inbound hookCF7 / n8n trigger → JSON payload
N° 02LLM分類 · 返信文生成LLM classification & replyClaude API · category + draft
N° 03Gmail下書き保存Save Gmail draftGmail API via GAS
N° 04担当者へSlack通知Notify owner on SlackSlack incoming webhook
03 / Outcome

Outcomes & decisions

  1. 分類・返信文作成にかかる手作業を大幅に削減。担当者は確認・送信のみに集中できる構成。Significantly reduces manual triage and drafting time — the team focuses on review and send only.
  2. 人手による「最終確認」工程は残し、誤送信リスクを排除。A human review step is preserved — drafts are never sent automatically.
  3. GAS版は0コスト、n8n版は分岐拡張性。要件で選択可。GAS variant is zero-cost; n8n variant scales branching logic. Pick by need.
  4. プロンプトはカテゴリごとに分離 ─ 後から追加・差し替えできる構成。Prompts are split per category, so they can be swapped without redeploy.

— Next case

YouTube AI Pipeline

次へNext