Real Case: AI Support for LINE Integration. Winning 12 Orders in One Month

Recently, I hear a lot of stories about connecting Slack or Notion to Claude and saying "You can manage customers and projects with AI." In fact, we do this internally and have introduced it to our clients. It is incredibly convenient for customer management.

However, after supporting one specific client, my perspective changed completely.

Regional companies often use LINE groups for internal communication. Especially in industries like renovation, painting, and construction where many people are on-site. Craftsmen, sales staff, office workers, and even the president all communicate via LINE.

Actually, my father's construction company also uses LINE groups for all internal communications.

※ Note that phone interactions are not fully captured either.

Unlike Slack or Notion, LINE is difficult to integrate with AI. You cannot easily feed group chats into an AI later.

So, when these companies talk about AI, they often stop at "We use LINE, so maybe later."

But I think this area is actually very promising. Many companies struggle here, yet few have addressed it.

When we tried it, we found you can do it while keeping the LINE group structure.

If you add one official LINE account as a "recorder" inside the group, it can capture incoming messages in real time.

The most common clients for this support were regional renovation and painting companies.

Group chats are automatically recorded, AI extracts necessary information into a customer ledger, and monthly reports are generated automatically. Field staff just write in LINE as usual.

We used three tools: Official LINE Account, Google Sheets, and Claude API. The running cost is about ¥1,200–¥2,700 per month.

Here, I will explain the entire process with diagrams.

What was the original state?

Inquiry channels were scattered: flyers, referrals, website, and LINE.

Customer interactions were mixed across phone, email, and LINE, managed by individual staff. Office staff manually entered data into a customer sheet, but information held privately by salespeople wasn't included.

Looking deeper, the same project was being manually entered in three places:

1. Customer management sheet
2. Monthly report form (30 self-reported questions)
3. Project-specific budget sheets

Normally, one might suggest introducing Slack or a CRM to consolidate this.

But adding a new tool just creates a fourth place to write. Three becomes four.

First step: Reading the LINE groups

Before choosing tools, we had them let us read their internal LINE group chats.

We found that office staff were already reporting in a nearly fixed format:

"Phone inquiry / Flyer response / Date / Name / Address / Phone / Request / @Person in charge Site survey scheduled for [Date] [Time]."

All necessary info was already in LINE. The only missing piece was the effort to copy it to the ledger.

So the strategy became: "Stop asking people to write new things; instead, mirror what they already wrote in LINE behind the scenes."

System Overview

The flow is:

1. Write in the LINE group as usual.
2. The "recorder" official LINE captures each message line-by-line.
3. Chatter is filtered out; AI extracts names, sources, status, and amounts from remaining messages.
4. Confident data goes to the customer ledger; uncertain items go to a "To Confirm" list.
5. Monthly reports are generated at month-end.

Note: Building the system (programming, data cleanup, testing, manuals) was done with Claude Code. The daily sorting engine runs on Claude API. Keeping these separate helps clarity.

5 Steps to Implementation

Honestly, nothing flashy.

1. Create a recording-only Official LINE and invite it to the group

Since existing official accounts were linked to other broadcast tools, we created a new one for recording.

Two common pitfalls:

• Group participation settings are off by default.
• Only one official account can join a single group. We moved the notification account to another group first.

2. Log incoming messages as raw data

Using GAS (Google Apps Script), we simply write each message line-by-line without processing.

Note: Past history before adding the recorder cannot be retrieved.

3. Use Claude API to sort necessary info

Chatter like "Meet tomorrow at 9" is filtered out before hitting the AI, reducing costs.

From remaining messages, AI extracts name, address, phone, source, work type, survey date, status, and amount. "32 man" becomes ¥320,000; "30th 14:30" becomes a datetime.

Items where AI hesitates (e.g., two people in one message) go to "To Confirm" rather than the main ledger to avoid errors.

We tested with 23 real patterns to ensure accurate sorting before going live.

4. Clean up existing ~870 rows

Old sheets had duplicates and inconsistencies. We merged 5 exact duplicates and flagged 105 ambiguous rows for client review.

5. Automate monthly reports

Reports match the existing meeting format. Preliminary reports on the 28th, final on the 1st. Includes counts and win rates by channel.

Also added auto-status updates after survey dates and Friday alerts for stalled projects.

Kickoff to production took about 6 weeks.

Insights from Cleanup

The biggest surprise during cleanup:

80% of rows were stuck at status "Site Survey."

Surveys happened, but whether quotes were sent, orders won, or lost wasn't recorded in the ledger. That data lived in monthly forms or budget sheets.

Thus, historical data couldn't show which channels (flyers, referrals, web) led to wins.

What if we switched to Slack/New Tools?

We could have mandated a switch to Slack or a new CRM. But that would just create a fourth writing location, likely leading to abandonment.

Stuck "Survey" rows would accumulate again. Budget allocation for flyers would remain intuitive. Counts exist, but revenue attribution doesn't.

Having numbers but missing the key metric is wasteful.

Where to Start

Order of operations:

1. Map who writes what and where currently.
2. Consolidate writing locations. Prefer leveraging existing habits over creating new ones.
3. Introduce AI afterwards. Build a human-review path for uncertain AI outputs first.

Data Gaps Between Roles

Field staff reported properly, office staff kept ledgers, presidents reviewed reports. Everyone did their part.

But the gap between "After Survey" and "Outcome" fell between field and office roles.

This mirrors gaps between Sales and Marketing regarding conversion rates and channel ROI.

Rules given to field staff were minimal:

1. Write one line in LINE when an order is won or lost (e.g., "Mr. X Order Won ¥59,000").
2. Send one message per customer.

This alone builds channel-specific win rates over time.

AI isn't left unchecked. For the first month, humans review data 3-4 times/month. Then every 2-4 weeks rules are refined. Non-LINE interactions (like phone calls) still require manual entry.

Action Plan for Tomorrow

Write down these 4 points for your company:

• Where/who writes new inquiries?
• Where/who writes survey/meeting schedules?
• Where/who writes won/lost orders?
• Where/who writes amounts?

If any info is written in 2+ places, that's your automation entry point.

If any info is written nowhere, that's your blind spot.

Regional companies using LINE groups have huge growth potential.

AI speeds up tasks, but sales depend on conversations. Capturing conversation outcomes in data makes the system meaningful.

Self Introduction

━━━━━━━━━━━━━━

Free consultation available for those struggling with lead gen and AI implementation.

DM "Wall Bounce" if interested.

▼ Free download: "Sales Materials Template Currently Winning Orders"

https://x.com/meguron12/status/2102518411377160594

▼ Follow for more.

https://x.com/intent/follow?screen_name=meguron12

━━━━━━━━━━━━━━