识别数据潜在异常值

结论摘要

• 异常指标与优先级（按影响与幅度综合排序）
  1. 热销SKU库存周转天数：21→35天（+14天，约+67%）—严重异常，高优先级
  2. 缺货率：1.1%→3.8%（+2.7pct，约+245%相对增幅）—严重异常，高优先级
  3. 7日留存：-3pct —中高优先级（核心健康指标显著下滑）
  4. 注册量：-5% —中优先级（漏斗顶部收缩）
  5. 首购转化率：-0.6pct —中低优先级（幅度较小，需确认是否超出常规波动）

关键诊断与假设

• 供应链信号冲突且指向结构性问题：热销SKU周转天数显著延长与缺货率飙升并存，通常意味着库存错配（库存积压在错误区域/非热销品或在仓而不在货架/前端），或补货/分配节奏与需求峰值错位。需优先验证口径一致性与SKU/渠道层面的分布。
• 用户指标受连锁影响：缺货与履约体验恶化会直接损伤首购与留存，当前留存下滑可能部分由供给问题驱动。
• 增长侧可能存在投放或季节性扰动：注册下降与首购小幅下滑需排查渠道/页面/促销变化；但优先级低于供应链异常。

行动项（按优先级执行） A. 供应链应急与结构优化（高优先级）

• 立即（48小时内）：
  • 对Top 50热销SKU做SKU×仓×渠道的库存可用性与缺货热力图；定位“有库存但不可售/不在正确节点”的错配。
  • 启动紧急调拨与补货；目标将全站缺货率拉回≤2%。
• 一周内：
  • 需求预测短期校正：采用最近4周滑动窗口并引入促销/节假日信号，对热销SKU上调安全库存因子20–30%。
  • 分配优化：按渠道/地区销量弹性重分配库存；设置跨仓转运触发阈值与SLA。
  • 运营联动：在供给受限的SKU上暂缓促销与投放，避免放大缺货损失；对页面启用“替代推荐/到货提醒”。
• 监控与告警：
  • 建立SKU日级EWMA控制图；告警规则：缺货率>2%连续2天或周转天数WoW>+20%。

B. 留存修复（中高优先级）

• 分段诊断：按渠道、设备、首购SKU、履约体验（到货时效/是否遇到缺货）拆解7日留存下滑的贡献。
• 干预与补偿：对受缺货影响的首购用户推送补偿优惠与替代推荐；优化到货通知与召回策略，目标次周留存回升≥1pct。
• 产品实验：对新用户上线轻量化引导与个性化Feed，A/B测试提升7日留存≥1pct。

C. 漏斗修复（中/中低优先级）

• 注册量：复盘W35→W36的投放与落地页更改；将预算从高成本/低转化渠道回撤至高转化渠道；目标次周注册环比+3%。
• 首购转化：排查支付失败率、缺货拦截逻辑、价格/优惠有效性；为缺货SKU启用替代推荐与延迟发货选项，目标提升≥0.5pct。

潜在异常值识别与验证（方法与步骤）

• 口径核查（必须先做）：确认“热销SKU名单”、库存周转天数计算方法（是否为DIO=平均库存/日均销出）、缺货判定规则在W35与W36未变更；若有变更需回溯重算。
• 统计检测框架：
  • 比例类（缺货率、首购转化、留存）：使用过去12周的均值与标准差做两比例z检验与控制图；阈值建议|Z|>2标记异常。
  • 规模类（注册量）：做时间序列季节性分解（如STL），对残差应用±2σ控制限判断异常。
  • 周转天数（SKU级）：用IQR法标记本周变动>+50%且落在上界外的SKU为异常值；用Bootstrap估计均值变动的置信区间。
• 关联验证：计算周级缺货率与首购/留存的相关系数（Pearson或Spearman）；若相关系数>0.5且显著，优先投入供应链修复。
• 异常点清单产出：列出SKU/渠道/地区层面的Top异常（周转天数激增与缺货率>5%的组合），作为优先调拨名单。

资源分配建议

• 70%资源投入供应链（数据与运营双线）：异常监控搭建、预测与分配优化、紧急补货与调拨执行。
• 20%投入留存修复（用户补偿、召回与产品引导实验）。
• 10%投入增长漏斗（渠道与落地页优化、首购路径修复）。

目标与里程碑

• 次周目标：缺货率≤2%；热销SKU周转天数<28天；注册止跌、首购转化回到W35水平；7日留存回升≥1pct。
• 两周内：热销SKU周转天数≤24天；完成异常监控仪表盘并上线自动告警。

说明

• 以上优先级基于当前缺乏历史方差与样本量的前提，采用效应量与业务影响的综合判断。建议尽快补充过去12周的基线与方差，以统计显著性做最终异常确认与资源分配微调。

Executive summary

• Paid ads efficiency deteriorated: impressions +25% with CTR down from 2.2% to 1.5% (-32% relative). Despite spend +18%, click volume likely fell ~15%, implying an effective CPC up ~39%. This is the primary anomalous channel.
• Search traffic declined 12% (abnormal negative trend), likely depressing discovery and contributing to lower AOV via mix shifts.
• Livestream traffic rose 40% (positive anomaly). Orders +3% suggest livestream likely offset declines elsewhere, but the 9% drop in AOV implies heavier discounting or a shift to lower-priced SKUs. Net revenue is likely down ~6% (1.03 × 0.91 ≈ 0.94).
• Two SKUs show high refund rates: 5.4% and 6.1% (product-level anomalies). These are potential outliers warranting immediate root-cause analysis.

Channel anomalies and diagnosis

1. In-platform ads (abnormal)

• Pattern: Impressions +25% with CTR down to 1.5% causes estimated clicks to fall ~14.8% versus prior (1.25 × 1.5% vs 2.2%). With spend +18%, effective CPC rises ~39%.
• Likely causes: lower-quality/new placements, weakened creative relevance, broader targeting diluting intent, frequency fatigue, or algorithmic changes pushing reach over relevancy.

2. Search (abnormal negative)

• Pattern: -12% traffic.
• Likely causes: ranking loss on key terms, listing relevance degradation (titles, attributes), price competitiveness changes, inventory/availability issues, or budget/visibility shifts from search to other channels.

3. Livestream (positive but affecting mix)

• Pattern: +40% traffic, orders +3% overall, AOV -9%.
• Likely causes: promotion-heavy livestreams driving lower-priced SKUs, aggressive coupons, or bundles that reduce per-order value.

Product anomalies and potential outliers

• SKUs with refund rates 5.4% and 6.1% are outliers relative to typical portfolio expectations. Without portfolio distribution, treat ≥5% as a high-risk threshold.
• Potential causes: quality defects (batch-specific), sizing/fit issues, misleading photos/descriptions, fulfillment damage, or host claims not aligning with product performance.

Recommended actions (prioritized) A) Paid ads optimization

• Immediate budget reallocation: Pause/trim placements and audiences with CTR below historical norm; divert spend to high-intent keywords and proven placements. Set guardrails (e.g., minimum CTR or ROAS thresholds).
• Creative and targeting refresh: Test new hooks, product angles, and formats; tighten audience definitions; cap frequency to reduce fatigue. Run rapid A/B tests with clear success criteria (CTR uplift ≥15%, CVR non-degrading).
• Placement audit: Compare CTR and CVR by placement/time/device; remove underperforming inventory.
• Bidding strategy: If CPC-based, lower bids on broad/low-CTR segments; if CPM-based, optimize for clicks/conversions objectives and exclude low-engagement inventory.

B) Search recovery

• Ranking and relevance: Audit top 50 keywords for share-of-voice changes; optimize titles, attributes, and images for clickability and relevance.
• Price and promo alignment: Ensure price competitiveness on core search queries; introduce targeted coupons rather than sitewide discounts.
• Inventory health: Fix stockouts, slow shipping flags, or negative reviews impacting rank.
• Bid/visibility: Restore budget to high-converting search terms; consider branded vs. non-branded balance to stabilize discovery.

C) Livestream performance with AOV protection

• Offer design: Shift from blanket discounts to value-add bundles, tiered upsells, and thresholds (e.g., free shipping > a target basket).
• Host and slot analysis: Measure CVR, AOV, and refund rates by host/session; prioritize sessions that keep AOV and returns in line.
• Cross-sell mechanics: Promote accessories/variants that lift AOV without increasing returns.

D) High-refund SKUs remediation

• Root-cause analysis: Pull refund reasons by code, cohort, and batch; inspect QA and fulfillment damage rates; compare size/color variants.
• Listing fixes: Clarify specs, sizing charts, and use cases; update images/video; adjust claims used in ads/livestream.
• Controlled testing: Temporarily reduce exposure; reintroduce with corrected listing and stricter targeting. Monitor refund rate each week; set a stop-loss threshold (e.g., ≥3.5% sustained).

Statistical checks to confirm anomalies and significance

• CTR change: Run a two-proportion test (pre vs. post CTR) using exact impression counts to confirm statistical significance; monitor by placement and audience.
• Channel traffic shifts: Use a weekly control chart to flag points beyond 2 SD from the prior 8–12 week mean; separate seasonality and event effects.
• Refund rate outliers: Compute portfolio median and IQR; flag SKUs with refund rates > median + 1.5×IQR as outliers. If counts are low, use Fisher’s exact test or binomial CI to assess whether ≥5% is above expected.
• Mix impact on AOV: Decompose AOV change into price effect vs. mix effect using a Laspeyres/Paasche-style index; identify SKUs disproportionately driving the 9% decline.

KPIs and monitoring plan

• Ads: CTR, CPC/eCPC, CVR, ROAS by placement/audience; weekly anomaly alerts for CTR drops >0.3pp or ROAS declines >15%.
• Search: Share-of-voice on top keywords, click share, CVR, rank position trends.
• Livestream: Session-level CVR, AOV, refund rate within 14/30 days, margin per order.
• Products: Refund rate by SKU/variant, top refund reasons, defect rate, net revenue impact.

Expected outcomes

• Restore CTR by 15–25% via creative/targeting and placement pruning, reducing eCPC and stabilizing ROAS.
• Recover search traffic by addressing ranking and listing relevance.
• Maintain order growth from livestreams while lifting AOV through structured offers.
• Reduce high-refund SKU rates toward portfolio norms, protecting net revenue.

Next steps (data needed)

• Weekly impressions, clicks, spend, orders, revenue by channel/placement.
• Search keyword rank and click share.
• Livestream session metrics.
• SKU-level order counts, refund counts and reasons, batch/inventory info.

These actions target the identified abnormal channels (paid ads and search) and high-refund SKUs, while preserving gains from livestream traffic and correcting the AOV decline.