数据可视化项目提案起草

会员增长与留存仪表盘叙事化设计——提案简介

项目目标

• 构建一套面向高层与运营团队的叙事化仪表盘，清晰呈现会员增长与留存的关键驱动，支持从“发生了什么”到“为何发生”再到“如何行动”的决策闭环。
• 统一指标口径，提供可追溯的指标解释与数据血缘，降低跨部门沟通成本。
• 通过可视化与交互引导，快速定位异常、识别高价值与高风险细分人群，并量化策略影响。

受众与使用场景

• 受众：管理层、增长与留存运营、市场渠道、产品经理、客户成功/客服。
• 频率：管理层周/月度复盘；运营与渠道日常监控与试验评估。
• 场景：增长分解、渠道投放优化、入职体验与转化优化、流失预防、续费/挽回策略评估。

关键业务问题

• 增长：净增长由哪些因素驱动（新增、复活、流失）？不同渠道/方案/人群的贡献与效率如何？
• 留存：各注册/付费 cohort 的留存曲线、回访行为与生命周期转移特征如何？
• 风险：哪些细分人群流失风险上升？异常发生在何时、何渠道、何版本？
• 价值：留存与生命周期价值（LTV）如何联动？单位投入的留存收益回报如何？

核心指标与口径（待与业务确认）

• 活跃会员数（MAU/WAU/DAU）：去重 member_id，按自然日/周/月。
• 新增会员：首次激活/首次付费的唯一用户数，明确试用转正口径。
• 流失会员：计入在观察窗口内到期且未续期，含宽限期规则。
• 复活会员：历史流失后再次激活/付费。
• 净增长：新增 + 复活 − 流失。
• 留存率：基于首激活或首付费 cohort 的 D7/W4/M1/M3/M6 留存。
• 渠道转化：渠道→注册→试用→付费漏斗转化率。
• 生命周期价值（可选）：分层 LTV 与回本周期（需交易与成本数据）。 注：时间粒度、去重规则、跨端身份合并、Grace Period、退款/冻结处理需在指标字典中固化。

叙事框架与信息架构

• 概览（What）：北极星指标与净增长分解，异常与波动归因提示。
• 驱动（Why）：渠道与产品维度的新增和转化、cohort 留存与回访路径、生命周期状态迁移。
• 行动（How）：高风险细分识别、策略影响预估、A/B 试验读数与对照。
• 证据（Trust）：指标定义、数据新鲜度、数据质量状态与版本变更记录。

可视化方案概述

• KPI 总览卡：活跃会员、净增长、流失率、复活率；提供环比/同比与置信区间或控制线。
• 增长分解瀑布图：净增长 = 新增 + 复活 − 流失，支持按渠道/地区/方案拆解。
• 渠道漏斗与效率：分渠道的注册→试用→付费漏斗（标准化率），投入产出对比（可选）。
• Cohort 留存热力图：按注册/付费周或月构建，支持按渠道/首单方案筛选与对比。
• 生存/流失风险曲线：Kaplan-Meier 生存曲线与危险率小图，洞察流失高发时点。
• 生命周期状态迁移：活跃→休眠→流失→复活的月度 Sankey/转移矩阵。
• 细分对比小倍图：按地区/渠道/方案的留存曲线并排；轴域一致、样本量提示。
• 季节性与趋势分离：滚动均线与同比视图，区分促销/节假日效应。
• 异常检测提示：基于移动控制图或阈值偏差的自动标注与注释锚点。
• 策略影响模拟（可选）：下拉选择人群与干预（如首月流失-1pp），实时估算影响规模与收入增量。

交互与个性化

• 全局筛选：时间、渠道、地区、计划类型、设备、用户新老（新客/复活/存量）。
• 维度下钻：从总览到渠道→活动→素材或从留存到用户路径明细（权限控制）。
• 归一化与对齐：支持 cohort 规模归一化、统一起点对齐、右删截尾标注。
• 书签/故事线：预设关键故事视图（月度经营复盘、渠道复盘、留存专项）。

数据与技术架构（参考）

• 数据模型
  • 事实：fact_membership_daily、fact_subscription、fact_events（激活/续费/取消）。
  • 维度：dim_member、dim_channel、dim_plan、dim_calendar、dim_geo、dim_device。
• 粒度与快照：日粒度快照，关键 cohort 与迁移矩阵预计算物化视图。
• 身份解析：跨端 ID 映射与去重策略；迟到数据与回溯更新策略。
• 性能：聚合表与缓存策略，大表列存与分区，<2s 交互响应目标。
• 安全：行列级权限控制，PII 最小化与脱敏。

质量保障与治理

• 指标字典与血缘：定义、计算逻辑、依赖表与版本管理。
• 质量监控：时序完整性、重复/缺失、滞后监控与告警；样本量下限提示。
• 变更管理：口径调整灰度发布与用户变更日志。

可达性与设计规范

• 色彩：色盲友好调色板，定性色用于状态/渠道，一致编码。
• 版式：移动优先的响应式布局；高对比度与字号规范；可下载长图/CSV。
• 标注：关键信息注释、异常与事件锚点、一致的单位与小数位。

交付物与里程碑（示例，时间以工作周计）

• 第1-2周：需求澄清与指标对齐；低保真信息架构与线框图。
• 第3-4周：数据建模与样例数据接入；高保真原型（Figma）与交互流程。
• 第5-6周：仪表盘开发与性能优化；质量校验与对齐评审。
• 第7周：试运行与用户培训；反馈修订与文档交付（指标字典、使用手册）。
• 第8周：上线与观测；迭代计划制定（策略模拟与实验读数模块可二期交付）。

成功衡量

• 指标理解时间下降（如关键问题从>15分钟降至<5分钟）。
• 异常发现与定位效率提升（告警到定位路径<10点击）。
• 决策产出：明确的留存提升实验与复盘节奏（季度≥2个有效策略落地）。
• 仪表盘使用率与粘性（周活用户数、核心视图停留时长、书签使用量）。

风险与依赖

• 口径不一致或渠道归因偏差影响结论；需建立统一归因与回溯规则。
• 数据滞后与迟到事件；需明确 T+1/T+0 新鲜度与回填窗口。
• 身份合并错误引入留存偏差；需稳定的 ID 图谱与评估样本。
• 小样本 cohort 波动；需最小样本门槛与置信区间提示。

本提案简介旨在对目标、方法与交付路径形成一致预期。详细技术规格、数据字典与原型稿将在需求澄清后提交。

Public Mobility Data Visualization POC — Proposal Introduction

Purpose This proof-of-concept (POC) will demonstrate an interactive, decision-focused visualization layer for public mobility data. It will unify operational and planning perspectives by surfacing reliable, real-time and historical insights on ridership, service reliability, and network performance, enabling faster diagnostics and evidence-based service adjustments.

Objectives

• Integrate and visualize key mobility datasets to validate technical feasibility and user value.
• Deliver a modular, scalable visualization layer with map-centric exploration and time-based analytics.
• Provide actionable KPIs for operations, planning, and policy evaluation.
• Establish data governance, accessibility, and privacy practices suitable for production scale-up.

Primary Users and Core Questions

• Operations managers: Where and when do headway gaps, bunching, or on-time performance (OTP) failures occur?
• Service planners: How do ridership, load factor, and transfer wait times vary by corridor and time-of-day?
• Policy analysts: What are equity and accessibility impacts across demographic and spatial segments?
• Customer comms: How to summarize disruptions and reliability trends for stakeholders?

Data Scope (POC)

• GTFS Static (routes, trips, stops, schedules) and GTFS Realtime (TripUpdates, VehiclePositions, Alerts).
• AVL/AVM (vehicle locations, timestamps), APC (automatic passenger counts) or fare/smartcard aggregates.
• Supplemental: weather, major events, roadworks/incidents (as available).
• Optional: shared micro-mobility or bike-share station status for first/last-mile context.

Key Metrics and Definitions

• Ridership: boardings/alightings; Aggregation by stop/route/time.
• Load factor: passengers ÷ vehicle capacity.
• Reliability: OTP (% within schedule tolerance), headway adherence, excess waiting time (EWT), travel time reliability (e.g., 90th/50th percentile ratio).
• Disruptions: count, duration, affected trips/stops, impact radius.
• Transfers: average wait, transfer success within threshold.
• Equity: KPI distribution by area classification (e.g., low-income or low-access zones).
• Sustainability (optional): estimated CO2e per passenger-km (methodology disclosed if included).

Proposed Visualizations (POC scope)

• Network Performance Map: deck.gl layers over base map; route color encodes reliability or load factor; stop-level circles with tooltips; time slider for playback.
• Headway Adherence Heatmap: route x time-of-day grid; color shows deviation from scheduled headway; drill to trip-level charts.
• Ridership and Load Factor Timeseries: small multiples by route/segment; peak identification and trend comparison (week-over-week).
• OD Matrix or Flow Map (aggregated): zone-to-zone ridership as heatmap; optional H3 hex aggregation.
• Disruption Impact View: timeline of alerts, affected routes/stops highlighted on map; pre/post KPI comparison.
• Transfer Experience: histogram of transfer wait times; spatial view of transfer hotspots.

Interaction Model

• Global controls: date range, time-of-day, day type (weekday/weekend), route/line filters, direction, headsign.
• Brushing and linking: selection on map filters all panels; hovering reveals contextual metrics.
• Drill-through: stop → segment → trip; automatic context retention.
• States: saved views and shareable links for reproducibility.

Architecture and Technology (POC)

• Ingestion: GTFS Static (batch), GTFS-RT (stream or frequent poll), APC/AVL (batch or stream).
• Storage/Modeling: DuckDB or Postgres for POC; dbt for transformations (trips, stop_times, events, KPIs).
• Indexing: H3 or spatial indexes for map tiles; pre-aggregations for time buckets.
• Visualization: deck.gl + Mapbox (map), Vega-Lite or Observable Plot (charts), or Superset for rapid assembly.
• Orchestration: lightweight scheduler (e.g., Airflow/Prefect optional for POC).
• Deployment: containerized app; CDN for static assets; optional feature flags for experiments.

Data Model (POC)

• Fact tables: events_vehicle (positions), events_trip (arrival/departure), fact_ridership (by stop/time), fact_disruptions.
• Dimensions: dim_route, dim_trip, dim_stop, dim_calendar, dim_zone (H3 or admin boundaries).
• KPI tables: precomputed reliability, headway adherence, EWT, load factors by route/stop/time grain (5–15 min).

Accessibility and UX Standards

• Color palettes color-vision safe (Okabe–Ito or ColorBrewer qualitative/diverging).
• WCAG 2.1 AA: keyboard navigation, focus order, alt text for charts, sufficient contrast.
• Dual encoding (color + shape/size) for critical states; tooltips with plain-language labels.
• Map clutter control: progressive disclosure and density-based aggregation.

Privacy, Security, and Governance

• No raw PII in the POC UI. For fare/smartcard data, use aggregation thresholds (e.g., k ≥ 50 per cell/time) and hashing with salted rotation where needed.
• Time and spatial binning to prevent re-identification.
• Role-based access; audit logs for data exports.
• Data retention policy documented for streams.

Performance and Reliability

• Target <2 s median response for typical filter changes; <500 ms for cross-filter updates with cached aggregates.
• Pre-aggregation by time buckets (e.g., 15 min) and tiles; vector tiles or MVT for high-performance maps.
• Graceful degradation: fallback to static aggregates if RT feed is unavailable.

POC Plan and Timeline (6–8 weeks)

• Week 1: Requirements and data audit; KPI definitions; success criteria finalized.
• Week 2: Data ingestion/modeling; schema and dbt pipelines; initial QA.
• Week 3: KPI computation; map and chart prototypes; interaction design.
• Week 4: Integration; cross-filtering; accessibility pass; performance tuning.
• Week 5: User testing with target personas; iterate on usability and latency.
• Week 6: Documentation; handover; demo; backlog for scale-up.

Deliverables

• Interactive POC dashboard with map and KPI panels.
• Data model and transformation scripts; reproducible environment.
• KPI dictionary and visualization style guide.
• Ops handbook (data refresh, monitoring) and privacy assessment.
• Demo dataset and scripted scenarios for stakeholder review.

Success Criteria

• User task completion: identify top 5 reliability bottlenecks and quantify EWT within 10 minutes.
• Performance: median interaction latency <2 s; map renders <1 s on standard hardware.
• Data quality: <2% orphaned stops/trips; clock skew resolved within 5 s tolerance.
• Adoption: positive usability scores from at least three target user roles.

Risks and Mitigations

• Data gaps or inconsistent IDs: implement reconciliation rules; fallback joins on shape and time proximity.
• Clock drift across sources: server-side normalization; tolerance windows.
• Sparse APC coverage: blend with modeled estimates; flag confidence levels.
• RT feed instability: cache last-known-good; display data freshness indicators.

This POC will validate end-to-end feasibility and user value, establish a scalable visualization blueprint, and provide a clear pathway to production with defined governance, accessibility, and performance standards.