Mercari is moving its internal documentation from Confluence to Notion. Kiko Ando wrote the checkpoint on 31 March 2026, with t-yama from the same Central Knowledge Management Committee.
The project is not finished. Ando says so in the first section. End of March was a milestone, not a shutdown of Confluence.
This is Mercari’s own engineering post. If you are on the same route, the Confluence → Notion playbook is the import checklist. Mercari’s post is what a large import looks like once you stop expecting it to be clean.
Why they started
They did not treat this as a tool swap. The project was a rebuild of knowledge management across Mercari.
Docs already sat in Confluence and in other tools. They wanted one structure that an LLM can read as context. They also wanted cheaper links to AI and to other tools, less duplicated documentation, and less of a split between teams that live in different products.
Ando lists one more reason that is easy to skip. If Mercari moves again later, a single store should make that move cheaper.
They started looking in July 2025. They talked to VPs, the PM Office, and risk and compliance. The hearing said some documents would stay behind, and that nothing else was a hard blocker. They started the real work from that.
The import was never going to be clean
Export and import dropped information and broke layout. When they surveyed it, they counted more than 50 gaps. Macros and unusual tables were in that set.
Ando treats that as normal. Confluence and Notion are different products. A no-effort, 100 percent import is not available. They decided to move on the assumption that people would fix pages by hand afterward.
IT kept testing the ugly cases. They wrote a guideline that names what to fix before import, what to fix after, and what “good enough” looks like.
The playbook makes the same split. Pages can move. Macros, marketplace apps, and full version history arrive as flattened content, or they stay behind. Mercari’s 50 is the size of their survey at the time. A new project needs its own count.
Finance docs left the company-wide track
Mid-project, Merpay and Mercoin hit a second wall. Some of their documents have compliance requirements that Notion’s standard features could not meet then.
They pulled those documents out of the company-wide migration. Merpay and Mercoin named their own owners. Those owners define the requirements and look for a solution on a separate track.
The first hearing had already said some documents would be skipped. This was a sharper version of that. A finance constraint showed up after the project had started.
What was done by the end of March
Aside from the documents they had excluded, and the ones on the separate finance track, the internal docs they still use were mostly imported into Notion.
They also finished a license cleanup. People who no longer need constant Confluence access could be taken off those licenses.
A lot of useful pages stayed in Confluence anyway. Some have no owner. Mercari still wants people to read them when they need them. They set up a way to grant temporary access and a way to grant lasting access.
The end state they want is a closed stack. Confluence stays. You request access when you need a page.
What they would redo
Ando splits the regrets into two moments.
If they could return to the Confluence rollout, they would have turned the freedom down. People could create spaces. People could issue API tokens. Rules existed, and the system did not enforce them. Individuals and teams built workflows that assumed those tokens. Once Notion was the plan, the same people asked for the same API access and the same system links.
When Mercari wrote tighter rules for Notion, some people said the new tool was less free. Ando thinks the old setup was too loose. Tightening a platform later creates pushback, and a long loose period makes that pushback louder. He would set the rules earlier.
If they could return to the start of this project, they would learn how Confluence was actually used before they argued about exceptions. A missing macro might be an edge case. They often could not tell, because they did not know which macros the company used, or how many of each. They also wanted update frequency by space, and which spaces had special requirements. Some of that became visible later. Decisions were slow while the counts were missing.
A custom migrator was tempting. It might cover some gaps. It also needs to be safe to run twice, and it can create new gaps. They did not have the people to take that risk. From outside the project, “why don’t you just build it” was a constant question.
Ando says a migration like this needs more than import skill. You need usage research, a budget of people’s time, a rule for what you will leave broken, and someone whose job is the project.
What I would copy
Write the gap list before you promise a clean import. Mercari found more than 50, then wrote the before-and-after fixes into a guideline. Plan the manual pass as part of the job.
Split documents that have a harder compliance bar. Merpay and Mercoin did not block the rest of the company. They got their own owners.
Count the macros, the active spaces, and the API tokens before you debate edge cases. Ando lost time because the totals were missing.
Keep a read path for pages you will not import. Mercari’s closed-stack plan is that path. Temporary access and lasting access are both explicit.
The Confluence → Notion playbook is the page-and-permission checklist. Use Mercari’s post for the part where the import is allowed to be imperfect, and Confluence stays as an archive you open on purpose.