需求计划对账代理

概览

价值

功能说明

• 量出每份预测自身的偏差

  连续八个季度乐观 18%，那不是预测，是一个带着已知修正系数、却没人去用的目标。

• 展示各份计划在哪里分歧

  按产品、按期间来看，而不是用一个把一切都掩盖掉的公司级总差额。

• 为分歧定价

  每一处分歧折算成缺货或积压要花多少钱。这就是能结束会议的那个数字。

• 指出在哪里该信哪份预测

  历史准确度因产品而异。同一个团队在一条线上可靠，在另一条线上就不可靠。

• 找出从不变动的预测

  每个周期都原封不动重复的数字是一个占位符，不是一份计划，它会被当作占位符对待。

• 不设定任何计划

  它没有通往计划系统的通道，也不锁定任何生产、采购或产能。

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.

{
  "horizon": "Q4 2026",
  "currency": "USD",
  "forecasts": [],
  "minimumTrackRecord": 6
}

{
  "escalate": true,
  "disclaimer": "An analysis of your own forecasts against your own actuals. No plan has been set or published, no production, purchase or capacity committed, no service level recommended, and bias-corrected figures are existing forecasts with measured error applied rather than independent forecasts.",
  "divergences": [
    {
      "period": "Q4 2026",
      "spread": "Sales 41,000 · Finance 36,000 · Operations 31,000",
      "product": "Product A",
      "unitsApart": 10000,
      "costIfWrongEitherWay": "Building to Operations and Sales being right costs 340,000 in stockouts. Building to Sales and Operations being right leaves 10,000 units held for roughly a year at 61,000. Both numbers land in departments that did not produce the forecast."
    }
  ],
  "biasBySource": [
    {
      "source": "Sales",
      "product": "Product A",
      "direction": "optimistic",
      "consistency": "consistent-correctable",
      "periodsMeasured": 8,
      "magnitudePercent": 18.4
    },
    {
      "source": "Operations",
      "product": "Product A",
      "direction": "conservative",
      "consistency": "consistent-correctable",
      "periodsMeasured": 8,
      "magnitudePercent": 9.1
    },
    {
      "source": "Sales",
      "product": "Product C",
      "direction": "unbiased",
      "consistency": "noisy-not-correctable",
      "periodsMeasured": 8,
      "magnitudePercent": 2.1
    }
  ],
  "biasCorrected": [
    {
      "note": "Sales corrected for its measured 18.4% optimism gives 33,500. Operations corrected for its 9.1% conservatism gives 33,800. Both existing forecasts, with their own measured error applied — not a new forecast. They agree once the bias is removed, which is the finding.",
      "product": "Product A"
    }
  ],
  "mostReliableBy": [
    "Product B: Operations has been within 4% for eight periods. Product C: Sales, where Operations has been erratic. Reliability is not uniform across the range."
  ],
  "escalationReason": "Product A forecasts are 32% apart with a priced consequence above threshold, and one source shows 18.4% consistent bias.",
  "unchangedForecasts": [
    "Product D's finance number has been identical for three cycles. That is a placeholder and should not carry equal weight in the meeting."
  ],
  "insufficientTrackRecord": [
    "Product E launched two quarters ago. No bias correction is applied to it."
  ]
}

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。