2026
Q4
季度 Impact Cases · 发布于Quarterly Impact Cases · Published 2026-12-18

让系统放大能力Systems That Multiply

更好的系统,让更好的工作方式可以不断复用。Better systems make better work repeatable.

本季度的案例把个人判断转化为团队能力:Incident Copilot 缩短故障恢复时间,发布信心地图让风险清晰可见,还有一系列无需从零开始、可供其他团队直接采用的实践模式。This quarter's cases turn individual judgement into team capability: an incident copilot that shortens recovery, a release map that makes risk visible, and practical patterns other teams can adopt without starting again.

06已确认案例validated cases
11可复用资产reusable assets
05覆盖职能functions represented
所有已确认案例All validated cases

认可成果,Achievement earns recognition.
让故事产生影响。Stories create impact.

每个经 TL 确认的案例都会进入本季度展页;页面呈现方式依据案例内容与可用素材确定。Every case confirmed by its TL appears on the quarter page. Page treatment reflects the available story and media.

会持续学习的 Incident CopilotThe Incident Copilot That Teaches Back

会持续学习的 Incident CopilotThe Incident Copilot That Teaches Back

Platform Reliability Guild

一套带有安全护栏的 AI 故障排查流程,减少了在 Runbook 之间反复搜索的时间,并把每次已解决事件沉淀成下一支团队更清晰的诊断路径。A guarded AI troubleshooting workflow reduced time spent searching across runbooks and turned every resolved incident into a better diagnostic path for the next team.

基线Baseline

值班工程师使用监控面板、日志和现有 Runbook 诊断事件,并在服务恢复目标内完成处置。On-call engineers diagnose incidents using dashboards, logs, and existing runbooks while meeting the service recovery target.

卓越表现Excellence

团队没有直接上线无约束聊天机器人,而是构建了经过评估的检索流程,加入引用核验、置信度提示和人工决策节点。The guild built an evaluated retrieval workflow with citation checks, confidence prompts, and a human decision point instead of deploying an unguarded chatbot.

影响Impact

在试点样本中,获得有效诊断的中位时间从 42 分钟降至 18 分钟,同时减少 Runbook 搜索成本,且没有引入任何自动生产操作。Median time to a useful diagnosis fell from 42 to 18 minutes across the pilot set, while runbook search effort dropped and no autonomous production action was introduced.

复用价值Leverage

评估集、提示词模式、安全检查表和事件后学习闭环,现已成为其他运营知识场景可复用的起点。The evaluation set, prompt pattern, safety checklist, and post-incident learning loop are now a reusable starter for other operational knowledge domains.

证据速览Evidence at a glance
  • 24 起事件的复盘样本24-incident retrospective sample
  • 诊断中位时间由 42 分钟降至 18 分钟42 → 18 min median diagnosis
  • 带引用回答接受率 87%87% cited-answer acceptance
  • 自动生产操作为 00 autonomous production actions
发布Release / 26.9.4有条件放行Conditional go
82信心度confidence
质量Quality变更风险Change risk回滚Rollback
人人都能读懂的发布地图A Release Map Everyone Can Read

人人都能读懂的发布地图A Release Map Everyone Can Read

Cindy Lau & Release Quality Circle

一张统一的信心视图,在每次发布决策前连通测试覆盖率、变更风险、回滚准备度和尚未闭环的证据。A single confidence view connected test coverage, change risk, rollback readiness, and unresolved evidence before each release decision.

基线Baseline

团队在发布检查点前完成计划测试并报告缺陷。Teams complete planned testing and report defects before the release checkpoint.

卓越表现Excellence

质量小组把分散信号转化成决策模型,与两个产品共同试点,并和工程师一起调整阈值,而不是再增加一张合规表单。The circle translated fragmented signals into a decision model, piloted it with two products, and refined thresholds with engineers rather than adding another compliance form.

影响Impact

发布后期的意外减少,高风险变更更早得到关注,试点团队每次发布还减少了约 6 小时的重复状态整理。Late release surprises fell, high-risk changes received earlier attention, and the pilot teams cut duplicated status preparation by roughly six hours per release.

复用价值Leverage

信心模型、证据检查表和会议引导指南,不依赖相同测试工具即可被其他团队采用。The confidence schema, evidence checklist, and facilitation guide can be adopted by teams without sharing the same test tooling.

证据速览Evidence at a glance
  • 2 个产品试点2 product pilots
  • 每次发布节省 6 小时6 hours saved per release
  • 后期风险升级减少 31%31% fewer late risk escalations
层叠需求区间展示未来数周的预测范围。 / Layered demand bands visualising a forecast range across upcoming weeks.不隐藏不确定性的需求预测Forecasting Demand Without Hiding Uncertainty

不隐藏不确定性的需求预测Forecasting Demand Without Hiding Uncertainty

Revenue Products & Data Science

预测体验不再给酒店团队一个看似精确的单点数字,而是清晰展示不确定性和可用于决策的范围。A forecast experience exposed uncertainty and decision ranges instead of presenting one deceptively precise number to hotel teams.

基线Baseline

提供每周需求预测并监控模型准确性。Provide weekly demand forecasts and monitor model accuracy.

卓越表现Excellence

团队将模型优化与用户研究、解释状态和决策阈值结合,使之真正符合规划人员的工作方式。The team paired model improvements with user research, explanation states, and decision thresholds that matched how planners actually work.

影响Impact

试点用户能更早应对高波动时段,在保留最终判断权的同时,减少了人工表格核对。Pilot users acted earlier on high-variance periods and reduced manual spreadsheet reconciliation while retaining final judgement.

复用价值Leverage

不确定性展示模式与研究指南,现已用于指导 TDC 内其他预测型体验。The uncertainty-display pattern and research guide now inform other predictive experiences across TDC.

证据速览Evidence at a glance
  • 8 位试点规划人员8 pilot planners
  • 核对时间减少 23%23% less reconciliation time
  • 解释有用性评分 4.6/54.6/5 explanation usefulness
来自实践From the practice

在技术变得不可见之前,先让取舍清晰可见。Make the trade-off visible before the technology becomes invisible.

高压决策实践手册A Playbook for Decisions Under Pressure

高压决策实践手册A Playbook for Decisions Under Pressure

Architecture Practice

12 张简明决策卡,帮助团队在设计讨论中选择集成、缓存、可观测性和故障处理模式。Twelve short decision cards help teams choose integration, caching, observability, and failure patterns during design conversations.

基线Baseline

架构师评审解决方案设计并记录重大决策。Architects review solution designs and record major decisions.

卓越表现Excellence

架构实践组把反复出现的建议整理成以案例为核心的短卡片,并在真实设计会议中测试,而不是发布一份冗长标准文档。The practice converted recurring advice into concise, example-led cards tested in live design sessions rather than publishing a long standards document.

影响Impact

团队更快形成清晰决策,评审者在下一轮评审中看到的重复设计缺口也明显减少。Teams reached clearer decisions sooner and reviewers saw fewer repeated design gaps in the next review cycle.

复用价值Leverage

卡片格式、引导脚本和贡献规则,让任何工程社区都能安全扩展这套手册。The card format, facilitation script, and contribution rules allow any engineering community to extend the playbook safely.

证据速览Evidence at a glance
  • 12 张决策卡12 decision cards
  • 9 场设计会议9 design sessions
  • 重复评审缺口减少 38%38% fewer repeated review gaps
64%校验提速faster validation
14已调查异常anomalies investigated
团队能采取行动的能耗信号Energy Signals Teams Can Act On

团队能采取行动的能耗信号Energy Signals Teams Can Act On

Hotel Systems & Sustainability Data

统一的数据契约让异常能耗模式更早暴露,酒店团队可以及时调查,而不必等到月底回顾。A shared data contract made unusual energy patterns visible early enough for hotel teams to investigate, rather than reviewing them after month end.

基线Baseline

维护可靠的能源数据流和月度报告。Maintain reliable energy-data feeds and monthly reporting.

卓越表现Excellence

团队统一了不一致的源数据定义,加入数据质量指标,并与负责调查异常的运营用户共同设计告警上下文。The team aligned inconsistent source definitions, added data-quality indicators, and designed alert context with the operational users who investigate anomalies.

影响Impact

试点更早发现持续的非营业时段能耗,也减少了行动前反复争论数据质量的时间。The pilot detected persistent out-of-hours consumption sooner and reduced time spent disputing data quality before action.

复用价值Leverage

数据契约、异常上下文模式和校验规则,还可扩展到用水及其他运营信号。The contract, anomaly context pattern, and validation checks can support water and other operational signals.

证据速览Evidence at a glance
  • 5 家试点酒店5 pilot properties
  • 调查 14 个异常14 anomalies investigated
  • 数据校验提速 64%64% faster data validation

一次抵店,更少交接One Arrival, Fewer Handoffs

Guest Journey Product Trio

重新设计的抵店前异常流程,减少了宾客到店前处理常见档案不匹配问题所需的团队数量。A redesigned pre-arrival exception flow reduced the number of teams needed to resolve common profile mismatches before a guest reached the property.

基线Baseline

按计划交付抵店前改进并达到服务支持目标。Deliver planned pre-arrival improvements and meet service support targets.

卓越表现Excellence

三人小组梳理完整交接链,观察实际支持工作,并同时调整界面和背后的职责边界。The trio mapped the whole handoff chain, observed support work, and changed both the interface and the ownership boundary behind it.

影响Impact

试点中,常见异常从四次交接缩减为一条引导式解决路径,减少了等待和重复解释。Common exceptions moved from four handoffs to one guided resolution path in the pilot, reducing delay and repeated explanation.

复用价值Leverage

跨系统交接地图与异常设计检查表,可以复用于其他宾客旅程节点。The cross-system handoff map and exception-design checklist can be reused for other guest journey moments.

证据速览Evidence at a glance
  • 典型交接由 4 次降至 1 次4 → 1 typical handoffs
  • 解决速度提升 19%19% faster resolution
  • 3 项可复用旅程工具3 reusable journey tools
把学习继续带向前Carry the learning forward

下一步,你的团队会What will your team
复用什么?reuse next?

打开 Impact WallOpen the Impact Wall 返回计划介绍Back to the program