门诊可及性分析代理

概览

价值

功能说明

• 先检验预约提前期

  爽约率随着「提前多久预约」陡峭上升。这是最强的那根杠杆，而且完全在你手里。

• 找出人们到不了的时段

  一天里的第一个时段，或者一个要换两趟公交才能到的时段。那是排程事实，不是患者的过错。

• 把没人记为损失的产能数出来

  提前告知取消却从来没被重新排上的号，以及候诊名单还排着队、诊次却提前结束。

• 核对通知信是否及时送达

  某个人群的沟通落在了就诊日之后。看不见，而且完全可以修。

• 把每一处改动折算成号数

  这处修补能挽回多少，好让它能和其他选项公平地竞争。

• 绝不对患者作画像

  它不产出任何个人风险分，也绝不建议任何会让某人被降低优先级的做法。只看模式。

这会改变什么

运行方式

该智能体通过 PROMIVO 运行时在服务器端运行。每次运行都会逐步记录，每次工具调用在执行前都会进行权限检查。

1. 1输入按智能体的输入结构进行校验
2. 2模型claude-opus-5，并自动回退
3. 3工具extract_structured_data
4. 4权限自动执行 READ 级操作；更高的等待你批准
5. 5输出符合输出结构的结构化结果
6. 6日志记录 token、耗时和每一个步骤

演示

演示数据Illustrative sample output, abridged.

{
  "period": "H1 2026",
  "appointments": [],
  "suppressionThreshold": 10
}

{
  "escalate": true,
  "disclaimer": "An operational analysis of appointment records. No clinical judgement has been made, no individual patient has been scored, profiled or identified, nothing recommends deprioritising or charging any patient, no appointment has been changed, and cohorts below ten are suppressed.",
  "slotPatterns": [
    "The 08:00 slot is missed at 2.4x the clinic average. The first bus arriving at that site is 07:52."
  ],
  "cohortPatterns": [
    "One cohort's non-attendance is 2.1x the average. The communication timing finding below appears to account for most of it, which makes it a fixable operational cause rather than a patient one."
  ],
  "leadTimePattern": [
    {
      "leadTime": "0–7 days",
      "appointments": 2100,
      "nonAttendanceRate": 4.1
    },
    {
      "leadTime": "8–28 days",
      "appointments": 4800,
      "nonAttendanceRate": 9.8
    },
    {
      "leadTime": "29–84 days",
      "appointments": 3900,
      "nonAttendanceRate": 19.4
    },
    {
      "leadTime": "85+ days",
      "appointments": 1400,
      "nonAttendanceRate": 31.2
    }
  ],
  "escalationReason": "Non-attendance differs materially between cohorts and appears linked to communication arriving late, which is an equity of access finding; and 610 refillable cancellations went unfilled against a waiting list of 1,900.",
  "cohortsSuppressed": 4,
  "fixesWithATradeOff": [
    "Overbooking the 08:00 slot would recover roughly 120 appointments and would transfer waiting time to patients who did attend. Listed last for that reason."
  ],
  "communicationTiming": [
    "For 340 appointments, the letter was sent fewer than 3 days before the date. Those show a 27% non-attendance rate against 9.8% for the same lead time band overall."
  ],
  "suppressionThresholdUsed": 10,
  "fixesWithNoCostToPatients": [
    {
      "change": "Cap routine booking lead time at 28 days with a partial booking model",
      "appointmentsRecovered": 740
    },
    {
      "change": "Refill cancellations with more than 48 hours notice from the waiting list",
      "appointmentsRecovered": 610
    },
    {
      "change": "Send communication at least 7 days ahead for every appointment",
      "appointmentsRecovered": 90
    }
  ],
  "capacityLostWithoutAPatient": [
    {
      "note": "A waiting list of 1,900 existed throughout. Nobody failed to attend these — they were simply never offered to anyone.",
      "cause": "Cancellations with more than 48 hours notice, never refilled",
      "appointmentsLost": 610
    },
    {
      "note": "",
      "cause": "Sessions finishing early against booked capacity",
      "appointmentsLost": 180
    }
  ]
}

使用条件

使用场景

价格

常见问题

评价

客户运行过该智能体后，评价功能将开放。