订舱、运单、拼箱、清关、对账共用同一份数据。AI 在流程中完成起草、核对与字段填写,是否采纳由你确认,每次调用的成本系统均可追溯。 Bookings, shipments, consolidations, customs and accounting all draw on the same record. AI drafts, cross-checks and completes fields within the process; you decide what is accepted, and the cost of each call is on record.
记录中列明名称、可操作的动作、运行模式、风险等级、所用模型与审核责任人。新上线的智能体先以影子模式运行,准确率达标后再放开权限。The row lists its name, the action it may touch, the mode it runs in, its risk level, the model it uses and who reviews it. A new agent starts in shadow mode and is promoted only once its accuracy meets the threshold.
算法并不复杂:AI 填写且你未修改的字段、它指出且你认可的问题、它自动完成的动作,这三类各按一条工时定额折算。token 支出除以省下的工时,即为每省一小时的成本。Three things count as saved work: a field AI filled that you never edited, a problem it flagged that you accepted, an action it completed unattended. Each is worth a set number of minutes. The token spend divided by the hours saved gives the cost per hour saved.
询价到结算共六个阶段。每个节点可以挂载智能体、设置进入条件,也可以标记为不得绕过的红线。改动先存为草稿,发布前可与线上版本比对。Six stages from quote to settlement. Any node can host an agent, carry an entry condition, or be marked as a red line that cannot be skipped. Edits sit in a draft you can diff against production before publishing.
这一行的关键信息大多来自邮件:订舱确认、提单草稿、改航通知、催款。因此收件箱直接内置于系统,来信自动关联到对应单号,正文中的航次、截关、箱型可解析后填入单据。In this trade most of what matters arrives by mail: booking confirmations, draft B/Ls, schedule changes, payment reminders. The inbox is therefore built into the system — messages attach themselves to the right job, and the vessel, cut-off and equipment in the body are parsed into fields.
请带一票真实数据,30 分钟从建单推进到结算。AI 在流程中提出了什么、消耗多少成本,我们一并查看。Bring one real job. Thirty minutes from entry to settlement, and we review together what the AI proposed along the way and what it cost.