Contact center analytics on the Wind (trigger) · Wave (topic) · Channel taxonomy, indexed by the questions an operation must answer. Every view names its probe question and its COPC anchor; every number traces to a contact record you can open, read, and export. Fictional brand, synthetic data.基于 風（触发）· 浪（话题）· 渠道 三轴分类法的联络中心分析，按运营必须回答的问题编排索引。每个视图都标注其探针问题与 COPC 条款锚点；每个数字都能追溯到一条可打开、可阅读、可导出的联络记录。品牌为虚构，数据为合成。
Wind — trigger風 · WindIdentifiable business trigger vs baseline demand有風（可识别业务触发）与無風（基线需求）
Wave — topic浪 · WaveInteraction topic and driver, taxonomy plus discovered clusters交互话题与驱动，既有分类法加发现聚类
Channel渠道 · ChannelVoice, chat, WeCom, auto-populated from system metadata语音、在线、企微，由系统元数据自动填充
Q0 · Is today under control, and where is it not?Q0 · 今天是否受控，哪里失控？
COPC CX R8.0 · Exhibits 1 to 3 metric families · Exhibit 1: satisfaction and dissatisfaction as independent variablesCOPC CX R8.0 · 附表 1 至 3 指标族 · 附表 1：满意与不满意须作为独立变量
Built to be scanned in ten seconds at shift start. Every status pairs a word with the color, and CSAT and DSAT hold separate rows because an averaged score hides a growing bottom box.为开班十秒扫读而设计。每个状态都用文字配合颜色，不单靠颜色传达；CSAT 与 DSAT 分列两行，因为平均分会掩盖不断增长的末框。
Intraday view日内视图
Q0 · Are today's actuals still the day we staffed for?Q0 · 今天的实际量还是我们排班时假设的那一天吗？
COPC CX R8.0 · 3.7.4 Real-time management, adherence at the intervalCOPC CX R8.0 · 3.7.4 实时管理，时段级遵时
Bars are actual offered volume to the now-marker; the dashed line is the forecast. Volume diverges at 13:00, service level bleeds from 13:30: the reforecast trigger fires on the top panel before the pain reaches the bottom one.柱为截至"现在"标记的实际来量，虚线为当日预测。13:00 量开始偏离，13:30 服务水平失血：重预测的触发点在上图，痛感才到下图。
Q9 · How long will these customers actually wait, and does the SL target match that tolerance?Q9 · 客户到底愿意等多久，SL 目标是否匹配这条容忍度？
COPC CX R8.0 · Appendix 1: abandon and speed-of-answer targets must be mathematically consistent · Cleveland: seven factors of caller toleranceCOPC CX R8.0 · 附录 1：放弃率与应答速度目标须数学一致 · Cleveland：来电者容忍度七因素
Share of queued callers gone by each wait time. Tolerance is local (Anton: New York abandons three times faster than Phoenix), so the curve is fitted, never imported. If the curve is flat at your SL threshold, the threshold is a costume.各等待时长下排队来电的累计放弃比例。容忍度是本地化的（Anton：纽约放弃速度是凤凰城的三倍），曲线必须实测拟合而非照搬。如果曲线在你的 SL 阈值处是平的，这个阈值就是装饰。
Median abandoner leaves at 96 seconds.放弃者的中位离开时点为 96 秒。
AHT decompositionAHT 分解
Q0 · Which component of handle time is moving, and who owns it?Q0 · 处理时长的哪个分量在动，归谁管？
The stack is the definition drawn. Total looks stable, up 5% on the year, while hold climbs 58% (a systems and knowledge problem) and wrap falls (a tooling win). The single-line AHT trend would have shown none of it.堆叠即定义本身。全年总量看似稳定，仅升 5%，而保持时长涨 58%（系统与知识库问题）、话后处理下降（自动小结上线）。单线 AHT 趋势什么都看不出来。
Talk通话Hold保持Wrap (ACW)话后（ACW）
Absenteeism control chart缺勤控制图
Q17 adjacent · Is absence in control, or are we interrogating noise?Q17 关联 · 缺勤是否受控，还是我们在审问噪声？
COPC CX R8.0 · 3.5.3 absenteeism managed to show it is in control · XmR limitsCOPC CX R8.0 · 3.5.3 缺勤须管理至"受控"状态 · XmR 控制限
The Standard's own wording is statistical; almost nobody draws the chart the phrase comes from. Limits computed from the data separate special cause (week 7, Chinese New Year, investigate the week) from common cause (fix the system, stop the weekly witch-hunt).标准原文就是统计语言，却几乎没人画出这句话对应的图。由数据算出的控制限把特殊原因（第 7 周，春节，查那一周）与共同原因（改系统，停止逐周问责）分开。
The table a certification audit reads first, kept as a standing monthly state instead of a pre-audit scramble. Filled dots are months at target; the right columns apply the Standard's two rules.认证审计第一眼看的表，改为按月常备状态而非审前突击。实心点为达标月份；右侧两列执行标准的两条规则。
Month at target当月达标Missed未达
Trigger decomposition, the Wind axis drawn触发分解，風轴落图
Q1 · What share of volume did the business schedule itself, and from which trigger family?Q1 · 本月量有多大比例是业务自己排出来的，来自哪个触发族？
Triggered demand (campaigns, policy changes, releases, logistics events) stacked over baseline demand. Triggered demand is calendar-coordinatable; marketing knows the blast date, WFM finds out at 9am. Hover a month for its triggered share.有風需求（营销、政策变更、发版、物流事件）堆叠在無風基线之上。有風需求可以日历协同：营销知道群发日期，WFM 早上九点才知道。悬停任一月份查看触发占比。
Baseline, no trigger無風基线Campaign营销活动Policy change政策变更App releaseApp 发版Logistics event物流事件
Semantic cluster map语义聚类图
Q4 · What is growing that has no taxonomy label yet?Q4 · 什么在增长而分类法还没给它名字？
A 3,000-record projection of the 12,000-contact sample. Hover or tap a cluster for share, growth, trigger mix, root cause, and live exemplars; each exemplar links to its full record in the explorer. The red cluster is the point: fast-growing demand the taxonomy had not named.12,000 条样本中抽 3,000 条嵌入投影。悬停或点击簇查看份额、增速、触发构成、根因与实时例句；每条例句链接到浏览器中的完整记录。红色簇是本图的意义：分类法尚未命名的高速增长需求。
Fulfillment履约Product产品Account and payment账户与支付Relationship关系Emerging, unlabeled新兴未命名Unclustered未聚类
Decisions priced in contacts用联络量为决策定价
Q1 · When a business decision ships, what did it cost in contacts and sentiment?Q1 · 一个业务决策上线，代价是多少联络与情绪？
Lettered markers are business events. Marker E, the September 2025 return-window cut, spikes volume briefly and steps sentiment up permanently.字母标记为业务事件。E 点，2025 年 9 月退货期缩短，让量短暂冲高，让负面情绪永久上一个台阶。
Monthly contacts (thousands)月度联络量（千）
Negative-sentiment share (%)负面情绪占比（%）
Contact volume by customer tenure按客户入会时长的联络量
Q3 · Who consumes capacity: new customers or the base?Q3 · 产能被谁消耗：新客还是存量？
Beyond the Standard · Customer-Base Audit C3 adaptation标准之外 · Customer-Base Audit C3 改编
The layer-cake, quarterly. New-customer contacts track acquisition; the widening tenured layers are the structural load. Growth quarters that look like service crises are often just the C3 doing what it does.季度层叠图。新客联络跟随获客节奏；不断变宽的老客层才是结构性负荷。看似服务危机的增长季，往往只是 C3 在做它该做的事。
Service demand by value tier按价值层级的服务需求
Q3 · Which tiers generate which demand, per hundred customers?Q3 · 每百客户口径下，各层级产生什么需求？
Contacts per 100 customers per year. Normalizing removes population size, so the VIC column reads as service intensity, the design input for routing and clienteling scope.每百客户年联络次数。归一化剔除人群规模后，VIC 列读出的是服务强度，即路由与专属服务范围的设计输入。
contacts per 100 customers, low to high每百客户联络次数，由低到高
Phrase shift after the policy change政策变更后的措辞漂移
Q1 · What language did the September policy put in customers' mouths?Q1 · 九月的政策把什么话放进了客户嘴里？
Log-odds of phrase occurrence, six months after the return-window cut vs six months before. The rising side is the policy speaking through customers.措辞出现的对数几率，退货期缩短后六个月对比前六个月。上升的一侧，就是政策借客户之口在说话。
falling · more common before下降 · 变更前更常见rising · more common after上升 · 变更后更常见
Topic share across earned, owned, and paid signal classes. Slopes are the finding: counterfeit anxiety runs 18/2/1, loud where we do not ask, near-silent where we do.话题在公开、自有、付费三类信号中的份额。斜率就是发现：正品焦虑 18/2/1，在我们不问的地方响亮，在我们问的地方近乎沉默。
Interval service level时段服务水平
Q8 · Which breaks first at the interval: volume, AHT, shrinkage, or adherence?Q8 · 时段层面先崩的是量、AHT、损耗还是遵时？
Daily averages hide intraday structure. Diverging color around the 80/20 target makes the Monday 9:30 surge and the Friday evening dip read as shapes: a scheduling fix first, an IVR-deflection fix second.日均值掩盖日内结构。围绕 80/20 目标的发散配色让周一 9:30 的冲击和周五晚间的低谷读成形状：前者先修排班，后者再修 IVR 分流。
under 60低于 6060–7070–8080–8888 plus88 以上Target: 80% answered in 20s目标：20 秒内应答 80%
The power of one一个人的力量
Q7 · What does one agent buy at the margin?Q7 · 边际上多一个坐席买到什么？
Erlang C on the Mandarin voice queue slice, AHT 210s, target 80% in 20s. Queues are nonlinear: move the agent slider one body at the margin and watch service level fall off a cliff. The strip shows adjacent staffing levels, and the cards act out the first immutable law: service level up, occupancy down. Per COPC, service level is measured on transactions offered to the queue, which is what the model computes.普通话语音队列切片上的 Erlang C，AHT 210 秒，目标 20 秒内 80%。队列是非线性的：把坐席滑块在边际上挪一个人，看服务水平跌落悬崖。色条显示相邻人力档位；卡片演的正是第一条不变定律：服务水平升，占用率降。按 COPC，服务水平以进入队列的 offered 量计，模型算的就是它。
Service level (20s)服务水平（20 秒）
Average speed of answer平均应答速度
Occupancy占用率
Service level at adjacent staffing相邻人力下的服务水平
Shrinkage waterfall损耗瀑布
Q8 · Where do scheduled hours actually go?Q8 · 排班工时到底去了哪里？
Shrinkage decomposition from scheduled hours to handling. Cleveland treats the rostered staff factor as the schedule overlay computed from these decrements, and COPC requires shrinkage as a forecast input to the capacity plan. Every decrement has a named owner: HR owns leave, ops owns adherence, WFM owns the waiting time.从排班工时到实际处理的损耗分解。Cleveland 把 rostered staff factor 视作由这些减项算出的排班放大系数；COPC 要求损耗作为产能计划的预测输入。每个减项都有名有主：HR 管请假，运营管遵时，WFM 管等待时间。
Forecast bias, signed预测偏差，带符号
Q6 · How wrong is the forecast, in which direction, at which lag?Q6 · 预测错多少、往哪个方向错、在哪个提前期错？
Actual minus forecast at capacity lag, against a ±5% band. Direction assigns the pain: under-forecast and customers pay in service level, over-forecast and the CFO pays in idle hours. November and June are the commercial calendar arriving unannounced, the same finding as the trigger view from the other side.产能提前期口径的实际减预测，对照 ±5% 容差带。方向决定谁疼：低估则客户在服务水平上买单，高估则 CFO 在闲置工时上买单。11 月与 6 月是商业日历不打招呼地到场，与触发视图互为镜像。
Retrial inflation, the invisible queue priced重拨膨胀，看不见的队列定价
Q5 · During incidents, how much offered volume is the same customer retrying?Q5 · 事故期间，offered 量里有多少是同一客户在重拨？
Cleveland: invisible queue, retrial; Erlang C assumes lost calls vanish · TP customer-level linkageCleveland：看不见的队列与重拨；Erlang C 假设流失来电消失 · TP 客户级关联
A 72-hour payment outage. Offered volume is inflated by the same customers redialing; the shaded wedge is self-inflicted volume. Staff and write the postmortem on unique demand, the lower line.一次 72 小时支付故障。offered 量被同一批客户的重拨推高；阴影楔形就是自造的量。按下方那条唯一客户线排人力、写复盘。
Drivers on the left, outcomes on the right, width is share of total volume. Refund and returns sends the largest share back within 30 days. Failure demand is a defined term in the COPC glossary, so this view speaks the auditor's language, not ours.左为驱动，右为结果，宽度为占总量比例。退货退款把最大份额在 30 天内送了回来。失败需求是 COPC 术语表定义的词条，这张图说的是审计师的语言，不是我们的行话。
Time to recontact再联络时间
Q14 · After we resolve, do they come back, and how fast?Q14 · 解决之后，他们回不回来，多快回来？
Cumulative share contacting again, by days since first contact. A steep first 72 hours means resolution failed on the spot; a climbing tail means the process breaks repeatedly. Endpoints reconcile with the Sankey, and the 30-day customer-level window is the next-issue-avoidance lens from The Effortless Experience rather than same-issue FCR.首联后再联络的累计比例。头 72 小时陡峭说明当场没解决；尾部持续爬升说明流程反复失效。端点与桑基图对账；30 天客户级窗口是 The Effortless Experience 的"下一问题规避"视角，而非同题 FCR。
Next-issue transition matrix下一问题转移矩阵
Q14 · When they come back, what do they come back about?Q14 · 回来的时候，他们为了什么回来？
The Effortless Experience: next issue avoidance · derived from the linked recordsThe Effortless Experience：下一问题规避 · 由关联记录实时推导
Rows are the first contact's driver, columns the repeat contact's driver, cells the share of that row's repeats. The diagonal is unresolved same-issue; the off-diagonal cells are the next issues an agent could have preempted on the first call. Computed live from the 12,000-record set.行为首联驱动，列为重复联络驱动，单元格为该行重复量占比。对角线是同题未解决；对角线之外是首联本可预先化解的下一问题。由 12,000 条记录实时计算。
share of the row's repeat contacts占该行重复联络的比例
Channel-switch matrix渠道切换矩阵
Q11 · Which channel did they try before this one, and where does self-service leak?Q11 · 来这里之前他们试过哪个渠道，自助在哪里漏？
Rows are the starting channel, columns where the journey resolved. The outlined cells price the self-service estate in phone calls. Journeys, not sessions, which requires the stitching only the linked dataset provides.行为起始渠道，列为旅程最终解决处。描边单元格用电话量为自助资产定价。是旅程而非会话，这只有关联数据集拼得出来。
share of journeys旅程占比self-service leak自助泄漏
Deflection honesty拦截诚实度
Q12 · When the bot says contained, does the customer agree a week later?Q12 · 机器人说拦截了，一周后客户同意吗？
Containment measured at session end flatters the bot. Re-measuring with customer-level linkage over seven days produces the honest figure, the same linkage moat as the Sankey, extended to the AI channel.会话结束时点的拦截率美化了机器人。用客户级关联在七天窗口重测才是诚实数字，与桑基图同一条关联护城河，延伸到 AI 渠道。
In-session escalations sit outside both containment figures. Bounce-back matched on customer ID and driver within 7 days.会话内转人工不计入两个拦截口径。回流按客户 ID 与驱动在 7 天内匹配。
Effort to repurchase, behavior over claims费力度到复购，行为优先于表态
Q13 · Does effort predict repurchase in the base, beyond what surveys claim?Q13 · 费力度能否预测存量客户的复购，而不止于问卷说法？
The Oracle case linked effort to stated satisfaction and intent. This version keeps the question and swaps the outcome for observed repurchase from the linked base: same gradient, behavioral answer, a revenue number instead of a survey number.Oracle 案例把费力度关联到表态满意与留存意愿。此版保留问题、把结果换成关联客户底盘上的实测复购：同一梯度，行为答案，用收入数字替代问卷数字。
Issue resolved问题已解决Issue not resolved问题未解决
Effort from a single post-contact question, 1 to 5, bucketed. Repurchase observed in the linked customer base.费力度来自联络后单题（1 至 5）分档；复购在关联客户底盘中观测。
In-call emotion arcs通话内情绪弧线
Q13 · Where inside the interaction does effort spike?Q13 · 费力度在交互内部的哪个位置飙升？
Median sentiment trajectory through the interaction, by driver, minute by minute. The red dot marks the low point: identity checks for payment drivers, the policy reveal for refunds. Where the dip sits tells you what to fix inside the call.按驱动逐分钟的中位情绪轨迹。红点标最低点：支付类在身份核验处，退款类在政策告知处。低谷的位置告诉你该修通话里的哪一段。
QA calibration质检校准
Q15 · Is QA calibrated to customers or to itself?Q15 · 质检校准的是客户，还是它自己？
Every point is an agent: internal QA score against customer CSAT. The bottom-right quadrant is QA theater, high marks customers do not feel; the top-left is the transcript set to mine for training content. COPC calibration sessions require correlating monitor scores with customer experience measures; this scatter is that requirement drawn.每个点是一名坐席：内部质检分对客户 CSAT。右下象限是质检表演，客户感受不到的高分；左上是值得挖掘话术的转写集。COPC 校准会要求把监听评分与客户体验度量做相关校验，这张散点就是那条要求画出来。
Agents坐席High QA, low CSAT高质检低 CSATLow QA, high CSAT低质检高 CSAT
Critical error accuracy, three components关键错误准确率，三分量
Q16 · Are customer, business, and compliance accuracies each at target?Q16 · 客户、业务、合规三条准确率是否各自达标？
The Standard requires the three accuracies monitored as distinct components; averaging them hides a compliance breach behind a customer-accuracy surplus. Dashed lines are the Appendix 1 benchmark targets; rust points are misses.标准要求三条准确率分列监控；求平均会让客户准确率的盈余盖住一次合规失守。虚线为附录 1 基准目标，锈红点为未达月份。
Rows are what the customer says, columns are what broke, cells are monthly contacts. The two outlined cells are the quarter's fix list. Hover a cell for the named fix owner.行是客户说什么，列是什么坏了，格是每月联络量。两个描边格就是本季度的修复清单。悬停可见指名的修复责任人。
contacts per month, low to high每月联络量，由低到高
Repeat rate by agent tenure按坐席司龄的重复率
Q17 · At what tenure do agents stop generating repeats, and does attrition harvest them first?Q17 · 坐席到几个月不再制造重复联络，流失是否在此之前就把人收割了？
COPC 3.6, Exhibit 3: attrition · Glossary: attrition costs include the learning curveCOPC 3.6、附表 3：流失 · 术语表：流失成本含学习曲线
Handled-weighted repeat-contact rate by tenure band, from the agent roster. The curve is the learning-curve cost COPC prices into attrition, drawn: every agent lost before month 12 is repeat contacts you paid to create.按司龄段、处理量加权的重复联络率，取自坐席花名册。这条曲线就是 COPC 计入流失成本的学习曲线：每个不满 12 个月就流失的坐席，都是你花钱制造出来的重复联络。
Agent performance grid坐席绩效表
Q18 · Whose calls do we mine, whose do we coach?Q18 · 挖谁的录音，辅导谁？
COPC 3.4 Monitoring and coaching · 3.5 Staff performance · click a header to sortCOPC 3.4 监听与辅导 · 3.5 员工绩效 · 点击表头排序
The supervisor classic, sortable. Sort by repeat rate to find the mining set; sort by tenure to see the learning curve as a roster, not a chart.班组长的经典视图，可排序。按重复率排，找到值得挖掘的话术集；按司龄排，把学习曲线读成一张花名册。
Cost-to-serve Pareto服务成本 Pareto
Q19 · What does each driver cost, fully loaded?Q19 · 每个驱动全口径成本是多少？
COPC CX R8.0 · 4.2 Overall cost performance · Exhibit 1: Cost per XCOPC CX R8.0 · 4.2 总成本绩效 · 附表 1：Cost per X
Drivers ranked by cost share (volume × handle time), with the cumulative line. Authenticity checks rank 11th by volume and 5th by cost; volume dashboards would never surface it.按成本份额（量 × 处理时长）排序，附累计线。正品鉴别按量排第 11、按成本排第 5；只看量的看板永远不会把它浮出来。
Cost per contact trend单联络成本趋势
Q19 · Is unit cost moving because of mix or because of efficiency?Q19 · 单位成本在动，是结构还是效率？
COPC CX R8.0 · Exhibit 1: Cost per X, measured monthlyCOPC CX R8.0 · 附表 1：Cost per X，按月度量
Blended cost per contact, derived from the same driver mix and handle times as every other view. When this line moves, the Pareto above says whether the mix shifted or a driver got heavier.混合单联络成本，与其余视图同一套驱动结构与处理时长推得。这条线一动，上方 Pareto 会告诉你是结构变了，还是某个驱动变重了。
Service recovery, with its counterfactual服务补救，带反事实
Q21 · Where is the holdout for every intervention?Q21 · 每个干预的对照组在哪里？
Repurchase after a complaint: outreach cohort vs randomized holdout. The shaded gap is the only honest number in the deck; without the holdout, attribution is fiction.投诉后复购：外呼干预组对随机对照组。阴影缺口是整套材料里唯一诚实的数字；没有对照组，归因就是小说。
The Fortune ribbon运势彩带
Q22 · Do interactions move customers between states, or merely correlate with them?Q22 · 交互是推动客户状态迁移，还是只与状态相关？
Beyond the Standard · River of Time: HMM state transitions vs RFM标准之外 · River of Time：HMM 状态迁移对 RFM
Latent-state mix of the contacted base (HMM) with the share RFM flags, overlaid. The state mix shifts at month 14; RFM catches up at month 18. Four months is the intervention window everyone else misses.被联络客群的隐状态构成（HMM），叠加 RFM 标记比例。状态结构在第 14 个月开始迁移，RFM 到第 18 个月才追上。四个月，就是所有人错过的干预窗口。
Engaged活跃Cooling降温At risk流失风险Flagged by RFMRFM 标记
Closed-loop verification funnel闭环验证漏斗
Q21 · Of what we detect, what share do we prove fixed?Q21 · 识别到的问题里，有多大比例被证明已修复？
From detected issue to verified fix, with median cycle time per stage. Most VoC programs stop at stage three and call it closed loop. The last two bars are the product.从识别到验证修复，每级标注中位周期。多数 VoC 项目停在第三级就自称闭环。最后两条，才是产品。
Probe index探针索引
The discovery instrument: 23 questions, each wired to its evidence探查工具本身：23 个问题，每个接到它的证据
Domains: demand origins · capacity and forecast · journeys and effort · quality and people · money · proof域：需求来源 · 产能与预测 · 旅程与费力度 · 质量与人员 · 成本 · 验证
Every question either forces a decision or exposes a measurement the operation does not have. Buttons jump to the view that answers it. Questions without a view yet are marked as backlog with the data they need.每个问题要么逼出一个决策，要么暴露一项对方没有的度量。按钮跳转到回答它的视图；暂无视图的标注为待建及所需数据。
Data explorer, the records behind every chart数据浏览器，每张图背后的记录
Open any contact: metadata, handle-time split, full transcript, download打开任意联络：元数据、处理时长切分、完整转写、可下载
Honesty note: no audio in this demo; each record carries a deterministic diarized transcript standing in for the recording and its ASR output诚实声明：演示不含音频；每条记录携带确定性生成的分角色转写，替代录音及其 ASR 输出
Filter the 12,000 linked records, click a row to open it. Voice records show timestamps and hold segments; chat shows the message thread. Export the full set as JSON or CSV, or download any transcript as text. In production this panel links the call recording itself.筛选 12,000 条关联记录，点击行打开。语音记录含时间戳与保持段；在线渠道按消息串呈现。全集可导出 JSON 或 CSV，任意转写可下载文本。生产环境中，此面板直接链接录音本身。