Cross-border e-commerce agency: a governed digital employee and founder knowledge turned into software

Context

2 Open China Ecommerce helps European mid-market brands—many of them family businesses—sell on the Chinese platforms: Tmall Global, JD Worldwide, Douyin, Xiaohongshu. It is an expert-knowledge-intensive business spanning three time zones and two cultures, with a critical dependency on the knowledge accumulated by its leadership.

That is the classic shape of a service firm built around its founder: administration, onboarding and offboarding, business development, store diagnostics—processes that lived in the leadership’s head and calendar, with the ceiling on scale that implies. The brief was not to pilot AI. It was to turn that knowledge into operational capability of the organisation, with the security and control a firm handling confidential information for more than forty brands requires.

Collaboration

We designed and deployed the transformation on Claude, Anthropic’s AI—chosen for its reliability following complex instructions and its safety orientation—in four moves.

A digital employee on the payroll

An assistant with its own corporate identity now triages the administration inbox and runs complete administrative cycles on a schedule: invoicing, VAT, payroll requests. Not a chatbot—an agent with responsibilities, limits and recurring obligations.

Enterprise-grade governance

Autonomy thresholds: what it may do alone, what requires human approval. Mandatory approval on sensitive actions in connected systems. An auditable execution record. And—the differential—formal performance evaluation of the agent: a full audit of its executions, eleven documented lessons and an improvement plan. A digital employee managed with a documented performance review, the way a person is managed.

Know-how turned into software

Processes that depended on the leadership’s memory are now executable procedures the AI runs end to end, with human supervision at the sensitive points: full employee onboarding and offboarding, a business development programme with human approval on every send, and an independent store diagnostic methodology for the Chinese platforms that runs inside the back office itself and verifies every figure against the raw data.

Bilateral Europe-China infrastructure

Claude for global operations; Chinese models deployed from inside China for the local regulatory and commercial corpus, where Western tools do not reach well. An architecture that respects both regulatory realities in which the client operates—GDPR and the AI Act on one side, the Chinese framework on the other.

Outcome

Eight months in production. The digital employee triages some 200 emails a day and runs its administrative cycles on schedule with an auditable record—lifting roughly 70% of the administrative load off the leadership’s calendar.

Seven critical processes that used to depend on the leadership now run on the system, with human approval where an error would be expensive or irreversible. Independent store diagnostics on Tmall, JD or Douyin went from three days to four hours, every figure verified against back-office data. Confidentiality between client accounts is preserved by design, not only by policy.

The agent’s formal performance evaluation produced eleven documented lessons and an improvement plan—the basis of a replicable governance framework, and of the AI programme we now deploy with new clients. Nothing implemented demands an internal technical team or unaffordable bespoke development. The method—define the agent’s role, write its contract and its limits, measure before trusting, deploy in phases with continuous evaluation—is management applied over standard infrastructure.