Gadget EC support｜Insight work and AI agents cut CPA in half
Overview
| Item | Detail |
|---|---|
| Industry | Gadget EC |
| Price | About ¥10,000–¥50,000 |
| Result | CPA cut in half through insight work and AI agents |
| Sales | Owned EC only |
| Scope | From marketing strategy through execution |
A niche gadget. Copy came from using it hard, then insight. CPA came down to half. That knowledge was also built into an AI agent, so people who were not specialists could still write copy at the same quality.
Marketing strategy
| Strategy | Detail |
|---|---|
| Who / What | Lock who we speak to, and what we deliver, first |
| How | Meta ads only |
| Focus | Because the category is niche, deep insight work |
Strategy and execution, both. Who and What were locked after using the product ourselves. How was Meta ads only.
Nobody on the marketing team had used this gadget deeply. Even a paid-media specialist did not move CPA. Leading with specs does not create a reason to buy. So we used it hard, and wrote copy from “what it is like to use.”
The material was not the maker’s claim. It was what we felt after using it. Where it snags. Why someone keeps going. Who you would recommend it to. That became the Who / What detail.
Delivery was Meta ads only. Video and stills. Design followed the rules. The work went into the copy.
Marketing tactics
| Tactic | What we did |
|---|---|
| Meta ads | Video and stills. Design followed the rules. The weight sat on copy |
| Building and using AI agents | Store the deep insight, so other people can write copy at the same quality |
Delivery was ads only. CPA came down through copy precision. That copy was not allowed to stay with specialists only.
Meta ads
Creative was video and stills. Design followed the rules. Novelty in look was not the point. Copy was.
The problem was that media skill alone did not move CPA. Nobody on the marketing team had used the gadget deeply. Even a specialist could not find language that landed. So someone who knew the gadget sorted Who / What, used it heavily, and wrote the copy from there.
Building and using AI agents
The point was not to stop at “only the specialist can write strong copy.” A marketer who knows the gadget and can dig insight does not scale a team by themselves.
What we did: store the data, then make it repeatable. The experience and the sorting behind the specialist’s copy went into an AI agent. Someone with the basics can pull “the real insight of a heavy user” out of the agent and use it in ads.
Not only the marketer who knows the gadget inside out. We built and shipped an agent so other people could write copy at a similar level.