成交后没付款？AI审计跟进并推送Slack预警

原文要点

原模板每小时运行，抓取GoHighLevel指定pipeline/stage中closed-won且金额大于0的商机；先用Google Sheets的Notification Log按opportunity_id去重，再取联系人最近对话、消息历史和未完成任务，查Stripe客户charges计算成交后已付与未付。上下文里生成无跟进、跟进过期、客户未回、任务逾期、部分付款、应升级等确定性标记，交Gemini输出含severity/evidence/建议话术的JSON；需预警则发Slack并写回表格。配置要改pipelineId、stageId、locationId，Stripe查找原为固定email，需换成正确客户。

这次做成什么

做一个可审计的“成交未收款”每小时巡检练习：用合成商机、Stripe收款、对话和任务数据算欠款，生成确定标记，让模型只输出严格JSON，再决定是否发Slack并写Notification Log去重。

准备什么

• n8n可运行实例与Schedule Trigger可用
• GoHighLevel OAuth2凭据及locationId/pipelineId/stageId
• Stripe测试模式密钥与可查询charges权限
• Google Gemini或PaLM API凭据
• Slack Bot凭据与目标channel
• Google Sheets服务账号与Notification Log写入权限

具体怎么做

1. 建去重日志表

   在Google Sheets建Notification Log，列至少含opportunity_id、notified_at、severity、last_status；n8n用Google Sheets节点按opportunity_id lookup，存在则跳过，避免同一商机每小时重复提醒。

2. 拉取成交商机并过滤

   Schedule Trigger设每小时；HTTP请求GoHighLevel opportunities接口，查询参数替换为你的locationId、pipelineId、stageId；Filter节点保留status=closed-won且monetaryValue>0，再排除已在Notification Log出现的id。

3. 聚合三类证据

   对每个商机并行取：GHL conversations与messages按contactId取最新方向和时间；tasks按contactId取open且dueDate早于当前；Stripe先按客户email查customer，再list charges，累计created>=closeDate且paid=true金额，未付=金额-已付。

4. 计算确定标记再调AI

   Function/Code节点生成布尔flags：no_rep_follow_up、stale_follow_up、unanswered_customer_reply、overdue_tasks、partial_payment、escalation_due；把金额、天数、flags作为上下文发给Gemini，要求只返回规定JSON，不允许寒暄。

5. 分支输出与写回

   IF节点读取verdict.alert_needed；true则Slack postMessage到预警频道，文本含金额、证据和建议话术，随后append/update Notification Log；false只写last_status=ok不发消息。失败分支把HTTP状态写入同一表便于排查。

可直接复制的材料

你是B2B收款审计助手。只输出一个JSON对象，不要Markdown，不要解释。输入字段：opportunity_id, deal_value, paid_since_close, outstanding, days_since_close, last_customer_reply_at, last_rep_follow_up_at, open_overdue_tasks, flags。规则：severity只能是none|low|medium|high；outstanding>0且escalation_due=true为high；outstanding>0且unanswered_customer_reply=true或overdue_tasks=true至少medium；仅partial_payment=true为low；无欠款为none。alert_needed在severity为medium或high时true。evidence数组最多4条，必须引用输入字段。customer_message用中文，语气礼貌，包含应付金额和一个明确下一步。schema={"severity":"","alert_needed":false,"evidence":[""],"customer_message":"","internal_note":""}

{"opportunity":{"id":"opp_901","contactId":"ct_221","monetaryValue":12000,"currency":"USD","status":"closed-won","closedAt":"2026-10-01T09:00:00Z","email":"buyer@example.test"},"ghl_messages":[{"direction":"inbound","at":"2026-10-10T10:00:00Z"},{"direction":"outbound","at":"2026-10-03T08:00:00Z"}],"tasks":[{"id":"t1","status":"open","dueDate":"2026-10-08"}],"stripe_charges":[{"customer":"cus_test_1","amount":500000,"currency":"usd","paid":true,"created":1791158400}]

{"ghl":{"locationId":"REPLACE_LOCATION_ID","pipelineId":"REPLACE_PIPELINE_ID","stageId":"REPLACE_STAGE_ID","closedWonStatus":"closed-won","linkBase":"https://app.gohighlevel.com/v2/location/REPLACE_LOCATION_ID/contacts/"},"stripe":{"lookupBy":"contact.email","paidOnly":true,"since":"opportunity.closedAt"},"sheets":{"spreadsheetId":"REPLACE_SPREADSHEET_ID","sheet":"Notification Log","dedupeKey":"opportunity_id","columns":["opportunity_id","notified_at","severity","last_status"]},"slack":{"channel":"#finance-alerts"}}

输入与输出示例

本站虚构演示，非原作者运行结果；未声称已调用接口。

输入

商机opp_901成交额12000美元，10月1日成交；Stripe成交后仅一笔已付5000美元；客户10月10日来消息后销售未再跟进；任务t1截止10月8日仍open。

对应输出

{"severity":"high","alert_needed":true,"evidence":["outstanding=7000","unanswered_customer_reply=true","overdue_tasks=1","stale_follow_up=true"],"customer_message":"您好，您的订单尚有余款7000美元未结清。请今天回复是否按原链接付款，或需要我重发账单与分期方案。","internal_note":"通知财务并安排24小时内电话跟进"}

做完检查这些

• 同一opportunity_id第二次运行被Notification Log拦截且不重复发Slack
• paid_since_close只累计created不早于closedAt且paid=true的charges
• severity为medium或high时alert_needed=true且Slack与表格各有一条一致记录
• 模型输出能通过JSON.parse且不含Markdown代码块

容易卡住的地方

• Stripe原文用固定email查客户，接真实数据前必须改为联系人邮箱映射
• GHL的pipelineId/stageId/locationId错会拉到空列表或错管道
• 金额单位混用美元与分会直接导致欠款误判
• Slack发送成功但表格写回失败会形成重复提醒，需同一事务式检查