量身 AI← 案例库← Case library
案例 · AI 落地方案CASE · AI DEPLOYMENT PLAN · 区域公共交通 · 多伦多Regional transit · Toronto

安省都市联通 MetrolinxMetrolinx

从「票卡上的 AI」走到「路网上的 AI」From AI on the card to AI on the network
基于公开资料的分析,与该企业没有合作关系。Based on public information; no affiliation with the company.
视频制作中,上线后会出现在这里。Video coming soon — it will appear here.
一句话判断The verdict AI 在 PRESTO 客服里已见效;扩建规划同样需要它,并配上负责人与基线。AI is proven in PRESTO service; expansion planning needs it too, with owners and baselines.
起步顺序Order of work 先定基线,再在一条线路上与规划师并行做预测,最后跨机构复用。Baseline first, then one corridor forecast beside planners, then reuse across agencies.

公开事实Public facts 已核对Checked

  • 5M+
    PRESTO 服务的客户Customers served by PRESTOMetrolinx 公开资料Per Metrolinx materials
  • 11
    使用 PRESTO 的公交机构Transit agencies on PRESTO一套票务,多家运营方One ticketing system, many operators
  • 7,834
    2024 年员工数Employees in 2024Revelio Labs 统计口径Revelio Labs estimate
  • 3
    CTO 分享的 AI 应用,从试点到规模化AI uses its C T O described, from pilot to scale聊天机器人、卡片转换模型、客服主题聚类Chatbot, card-conversion model, call clustering

三块 AI 工作,一个先后顺序Three pools of AI work, one order 判断Judgment

  • PRESTO 客服与反欺诈PRESTO service and fraud(虚拟客服、语音认证与反欺诈已写入公开条款。Chatbot, voice authentication and fraud screening are already in public terms.)→ 先定咨询量与欺诈的基线。Fix baselines for contacts and fraud.
  • 扩建规划Expansion planning(为 GO 与地铁扩建做客流预测与排班。Ridership forecasts and crew scheduling for new GO and subway service.)→ 选一条线路,与规划师并行预测。Run one corridor forecast beside planners.
  • 跨机构数据Cross-agency data(十一家机构的票务数据,同一批客户。Ticketing data from eleven agencies, one customer base.)→ 先统一口径,再谈模型。Agree shared definitions before any model.
  • 基线Baseline → 与规划师并行试点Pilot beside planners → 跨板块复用Reuse across pools

专家组裁决Expert panel 依据公开观点的推演Inference from public views

Foteini Agrafioti

技术与业务战略对齐Strategy alignment

数据团队、PRESTO 与规划部门的权责写成一页纸;每个应用挂到服务目标上。One page of ownership across data, PRESTO and planning.

Erik Brynjolfsson

价值与度量Value & measurement

宣称收益前先定基线,并设一条对照线路。Fix a baseline and a comparison corridor first.

Ziad Obermeyer

算法在优化什么What it optimizes

反欺诈与语音认证:先写清真实目标,并按客户群分别看结果。For fraud checks, name the real target; compare by customer group.

Ethan Mollick

人怎么用How people use it

用真实排班案例测出预测会在哪里出错;规划师保持复核。Test forecasts on real cases to find where they break.

Andrew Ng

节奏与方法Pace & method

选一条三到六个月出数字的线路,并写明下一步复制到哪。One corridor, numbers in three to six months, then copy it.

Thomas Davenport

整家企业怎么变Whole-enterprise change

涉及十一家机构,这是流程变革而非加工具;每个应用点名流程负责人。Eleven agencies: process change. Name owners.

保留的分歧Disagreement kept · 一条线路 vs. 整个体系One corridor vs. the whole system

Andrew Ng:先在一条线路上赢,用它造势头。Win on one corridor first.

Thomas Davenport:十一家机构共用数据;流程要为全体设计。Design for all eleven agencies.

→ 只试点一条线路;共享数据口径从第一天起按全网来写。Pilot one corridor; write data definitions network-wide.

六位专家是谁、怎样思考、有什么铁律 →Who the six experts are, how they think, their rules →

三个目标Three priorities 建议Recommendation

1

证明一条线路的预测Prove one corridor forecast

一条扩建线路,有基线,规划师确认数字。One expansion corridor, a baseline, and planners who confirm the numbers.

2

有证据地扩大 PRESTO 客服Scale PRESTO service with proof

每次扩大前,先测问题解决率与欺诈拦截。Measure resolution and fraud catches before each widening.

3

统一跨机构数据口径Agree shared data definitions

跨机构票务数据的一套书面定义。One written set of ticketing definitions across agencies.

场景组合Scenarios 建议Recommendation

场景Scenario做法Approach状态Status
线路客流预测Corridor ridership forecast
提效Efficiency体验Experience
对一条扩建线路做客流预测,并与规划师自己的估算对比。Forecast demand on one expansion corridor and compare with the planners' own estimate.建议Proposed
排班决策辅助Crew scheduling support
提效Efficiency
工具为新增班次提出排班建议;每次调整由排班员批准。A tool suggests rosters for new service; schedulers approve every change.建议 · 人工批准Proposed · people approve
PRESTO 虚拟客服PRESTO virtual agent
体验Experience提效Efficiency
在已有咨询助手的基础上扩大规模,并测量解决率与转人工率。Scale the existing enquiry assistant with measured resolution and hand-off rates.已上线(公开)· 建议扩大Live (public) · scale-up proposed
跨机构反欺诈分析Cross-agency fraud analytics
降风险Risk增收Revenue
统一口径之后,整合各机构票务数据做反欺诈分析。Combine ticketing data across agencies, after shared definitions are agreed.建议 · 数据口径统一后Proposed · after data definitions

12 个月路线12-month roadmap 建议Recommendation

数据先行Data first

第 1–4 月Months 1–4
  • 咨询量与欺诈基线Baselines for contacts and fraud
  • 共享数据口径初稿Shared data definitions drafted
  • 选定一条线路One corridor chosen
◆ 基线签字Baseline signed

试点Pilot

第 5–7 月Months 5–7
  • 与规划师并行预测Forecast beside planners
  • 对照线路测量Comparison corridor measured
  • 边界测试记录Mollick-style edge test
◆ 规划师认可Planners trust it

上线Go live

第 8–10 月Months 8–10
  • 线路预测投入使用Corridor forecast in use
  • 一个车场试用排班辅助Scheduling support in one depot
  • 上线即监控Monitoring from day one
◆ 负责人签字Owner sign-off

扩展Extend

第 11–12 月Months 11–12
  • 复用到第二条线路Reuse on a second corridor
  • 机构数据复盘Agency data review
  • 年度重排优先级Next year re-ranked
◆ 进入下一年Next year begins

治理与合规Governance 建议Recommendation

  • 把每个应用对应到已负责 AI 监督的董事会委员会Map each use to the board committee that already oversees AI
  • 语音与反欺诈数据:向法律顾问确认隐私义务Voice and fraud data: confirm privacy obligations with counsel
  • 规划师与排班员批准,AI 只给建议Planners and schedulers approve; AI only proposes

三层指标Three-layer metrics 建议Recommendation

采用Adoption规划师每周使用预测的比例Planners using forecasts weekly · 助手处理的 PRESTO 对话数PRESTO chats handled by assistant
过程Process预测与实际客流的误差Forecast error versus actual ridership · 未改动即被采纳的排班Rosters accepted without edits · 转人工率Hand-off rate to staff
结果Outcome每条线路的规划工时Planning hours per corridor · 每千次出行拦截的欺诈Fraud caught per thousand trips

目标值由企业自定。Targets are the company's to set.

风险Risks 判断Judgment

没人认领的预测A forecast nobody owns

如果没有规划师为数字负责,它不会改变任何排班。If no planner owns the number, it will not change a roster.

各机构口径不同Agencies define things differently

同一个客流词可能有三种含义;先统一定义。One ridership word can mean three things; fix definitions first.

面向客户的错误Customer-facing errors

虚拟客服在票价上答错会损害信任;要测转人工情况。A virtual agent wrong about fares erodes trust; measure hand-offs.

前 90 天First 90 days 建议Recommendation

第 1–2 周Weeks 1–2点名流程负责人;盘点在用 AIName process owners; list live AI uses
第 3–4 周Weeks 3–4收集咨询量与欺诈基线Collect baselines for contacts and fraud
第 5–10 周Weeks 5–10选定线路;起草数据口径Pick the corridor; draft data definitions
第 13 周Week 13首次复盘First review
今天只需要一个决定:One decision today: 指定一条扩建线路及其规划负责人name one expansion corridor and its planning owner

我们可能错在:We could be wrong if: 规划师在真实案例上否定预测,或基线显示客流本来就很好预测。planners reject the forecast on real cases, or the baseline shows ridership was already predictable.

本页是基于公开资料的分析,与所述企业没有合作关系,不代表其官方立场或内部计划;企业数字多来自其自述或媒体转述;「专家组」是依据六位专家公开观点的推演,不是他们的原话,也不代表他们或其雇主的立场。资料整理日期:2026-11-14。This page is an analysis based on public information. It has no affiliation with the company and does not represent its official position or internal plans. Company figures come mostly from its own statements or media reports. The expert panel is an inference from six experts' public views, not their words, and does not represent them or their employers. Compiled 2026-11-14.

想给你的公司也做一份?Want one for your company?

同样的方法:公开事实、专属判断、12 个月路线和前 90 天。公司大小都可以聊,中文沟通。Same method: public facts, a company-specific judgment, a 12-month roadmap and the first 90 days. Any size of business; we can talk in Chinese or English.

Telegram @johnsunlei邮件 Email [email protected]