Using AI to Grow Your FAQ: What Role Division Was Decided Before Introducing a LINE Bot?

Hello, this is Minemura.
Last time, I touched uponevaluating tools when creating a LINE bot.
This time, I will talk about thinking through which parts of the overall design to entrust to AI.

When communicating with contractors via LINE, the same questions are repeated over and over. Having a person answer them every single time is not just a hassle, but also leads to risks.

"Why are staff members responding to these questions every time when they don't really need a human to answer them?" I had been ignoring this sense of discomfort, but the burden on operations and the seeds of trouble were growing little by little.

Although I decided to introduce AI to solve this, simply automating the initial response is not enough.
What I am thinking of is positioning AI as a "new staff member who grows the FAQ."

Current Flow

Currently, the flow is just: LINE inquiry → human checks and answers → share to Slack. Because of this, the following problems tend to occur.

• Confusion on the front lines due to not being able to answer immediately

• Repeatedly answering the same questions

• Information is not being accumulated

When I calculated it—

• 5 minutes per inquiry

• 20 inquiries a day, 100 a week → 400 minutes every month (over 33 hours!?)

• About 70% of the content consists of inquiries that have occurred in the past

In other words, humans are spending over 3 hours every month on "questions that could be solved if they were just in the FAQ."

Flow After AI Introduction

After introducing AI, the LINE inquiry will be checked against the FAQ by the AI to provide an initial response and accumulate the FAQ. This will enable immediate responses, automatic answers to similar content, and the accumulation of answers. I think the tasks requiring human intervention will be limited to things like escalations or deciding on new FAQ registrations.

• Notify Slack if the AI cannot determine the answer

• AI organizes what humans have answered as FAQ candidates

• Trigger FAQ registration with a Slack stamp

AI's Role

1. Primary Response (FAQ Matching)
   　Immediate replies to frequently asked questions.

2. Learning (Reflecting staff answers into FAQs)
   　Automatically organizes escalated questions and their answers.
   　Uses generative AI to format them into "FAQ-style text."

3. Slack Notifications
   　- Escalation notifications
   　- Reminders for unanswered questions
   　- Requests to confirm FAQ addition candidates

4. FAQ Registration via Stamp Detection
   　When you want to save something as an FAQ, simply add a stamp in Slack to make it a registration candidate.

Human's Role

• Cases requiring judgment (new or exceptional questions)

• Approval of FAQ candidates

• Adjustments when the BOT's response is unnatural

AI as a New Staff Member

I decided to treat the AI as a new staff member named "Arai-san" (Newcomer + AI).

• At first, it's awkward and the answers are insufficient.

• A senior staff member needs to check and correct them.

• But it grows little by little as answers accumulate.

Training a new staff member also takes effort at first.
However, once that person grows, the effort is significantly reduced, and the entire team becomes stronger.
The same goes for Arai-san (the new AI); while it takes time to set up the initial system and organize the FAQs, it will continue to work as a "staff member who reduces effort" thereafter.

The Future of Hiring Arai-san (New AI) vs. Not Hiring

• If you don't hire Arai-san
  　・Unresolved issues carry over to the next day, leading to trouble.
  　・FAQs don't grow, and you keep answering the same questions.
  　・Psychological burden on people accumulates.

• If you hire Arai-san
  　・BOT responds immediately → The team can act right away.
  　・FAQs grow, and the number of questions itself decreases.
  　・People can focus on judgment and approval.

Looking at this contrast, I feel that "incorporating AI into the role of developing FAQs" is highly significant.

Summary

Entrusting AI with tasks is not just about "providing answers."
Learning from the content answered by humans and growing the FAQ is the role best suited for AI.

Just as you would explain the scope of work to a new staff member, you should also decide "how much to entrust" to Arai-san (the new AI).
By clarifying roles, AI will grow into a reliable team member.

I am convinced that by organizing this before implementation, you can prevent hesitation and rework when actually operating the tool.