工作中自主发起 · 自动化Self-initiated at work · automation
发票对账自动化工具Invoice reconciliation tool
Korea Foods Company Ltd（英国）· 自主发起Korea Foods Company Ltd (UK) · self-initiated
- 接手积压半年以上、覆盖英国 25 家门店的 Amazon 平台采购发票。手工处理完三个月的数据后，我发现对账规则是固定的，可以交给程序做，于是向上级提议并立项。
- 先把流程、判断逻辑、例外情况和优先级写清楚，再和 AI 多轮对齐需求，2 天交出第一版。用手工核对过的三个月数据检验，结果全部一致，还反查出人工核对时的一处商品分类错误。
- IT 提出财务数据不能上传外部服务后，我把它改成在本地离线运行的 Python 工具，加了四层校验和人工覆盖机制。月度对账从约 3 天缩短到半天至 1 天，剩下的时间主要用在人工复核上。
- 写了配套 SOP，并把工具和文档改成英文，交给英国同事独立使用。
- Took over Amazon purchase invoices for 25 UK stores that had been left unprocessed for more than six months. After reconciling three months by hand, I saw that the matching rules were fixed and could be automated, and proposed the project to my manager.
- Wrote out the workflow, decision logic, exceptions and priorities first, then worked through the requirements with AI over several rounds and delivered a first version in two days. Tested against the three months I had checked by hand, it matched every result, and it caught one product-category error in the manual work.
- When IT ruled that financial data could not be sent to external services, I rebuilt it as a Python tool that runs offline on a local machine, with four layers of validation and manual overrides. Monthly reconciliation went from about three days to between half a day and one day, most of it now spent on manual checks.
- Wrote an SOP, translated the tool and documentation into English, and handed both over to UK colleagues, so they could run it on their own.
工具由 AI 辅助编写，我负责规则、校验设计、测试和验收。属于公司内部资料，代码与数据不公开。The code was written with AI assistance; the rules, validation design, testing and sign-off were mine. Company-internal: code and data are not public.