返回 Papers
AI 底层逻辑 / 经典论文

AI Customer Vulnerability:可访问与包容性架构

Inclusive AI 不是“更有同理心的 chatbot”。在金融零售里,它是一套 customer-outcome control system:把客户在某段旅程中的支持需求、可访问性需求和高伤害风险识别出来,用最小侵入的方式提供帮助,并证明客户没有因为处境困难、沟通障碍或数字无障碍需求而受到更差服务。

210ai-foundations/papers/130-ai-customer-vulnerability-accessibility-inclusive-ai-architecture.md

AI Customer Vulnerability / Accessibility / Inclusive AI Architecture 解读

配对阅读:本篇的操作手册版(模板/RACI/门禁/runbook)是 docs/AI_CUSTOMER_VULNERABILITY_ACCESSIBILITY_INCLUSIVE_AI_PLAYBOOK.md。第一遍读本篇建立原理与架构判断;第二遍做案例时再用 playbook 查表落地,两者不需要重复精读。

重要说明: 本文只讨论金融零售 inclusive AI 的产品与架构设计,不构成法律、监管、ADA/WCAG 合规认证、消费者保护、医疗/心理判断、能力评估、信贷/保险/投资建议、模型验证或客户通知结论。真实项目必须由 Legal、Compliance、Accessibility、Customer Experience、Conduct Risk、Privacy、Model Risk、Operational Risk、Frontline Operations、Complaint Operations、Fraud/Scam Risk、Product、Data Governance、Information Security 等共同确认。


Source Anchors

SourceLink用途
WCAG 2.2https://www.w3.org/TR/WCAG22/用 Perceivable / Operable / Understandable / Robust 及 success criteria 建立数字渠道和 AI interface 的 accessibility baseline
DOJ ADA web guidancehttps://www.ada.gov/resources/web-guidance/用 ADA web accessibility guidance 训练 web/mobile/AI self-service 渠道的可访问性治理语言
CFPB compliance circulars landing pagehttps://www.consumerfinance.gov/compliance/circulars/用 consumer financial protection、unfair/deceptive/abusive conduct、servicing、complaints 和市场行为风险的监管语言组织 conduct-risk thinking;具体 circular 适用性须由合规/法律判断
FTC advertising and marketing guidancehttps://www.ftc.gov/business-guidance/advertising-marketing用 truthful marketing、substantiation、dark patterns、endorsement/claim control 的思想约束 AI-generated advice、sales script、hardship offer 和 vulnerability-sensitive messaging
NIST AI RMFhttps://www.nist.gov/itl/ai-risk-management-framework用 Govern / Map / Measure / Manage 组织 vulnerable-customer AI risk、harm taxonomy、control effectiveness、monitoring 和 continuous improvement
ISO/IEC 42001 overviewhttps://www.iso.org/standard/42001用 AI management system、policy、roles、operation、performance evaluation、audit 和 improvement 建立 inclusive AI operating model

核心导读

Inclusive AI 不是“更有同理心的 chatbot”。在金融零售里,它是一套 customer-outcome control system:把客户在某段旅程中的支持需求、可访问性需求和高伤害风险识别出来,用最小侵入的方式提供帮助,并证明客户没有因为处境困难、沟通障碍或数字无障碍需求而受到更差服务。

AI 改变的是服务处理链的早期判断。传统流程往往等客户明确投诉、失败多次或进入人工队列后才补救;AI 可以在交互摩擦、客户陈述、账户事件、语言偏好、欺诈/诈骗信号和员工观察中发现 support need,并动态调整说明方式、渠道、handoff、safe pause 或 specialist review。但这种能力也会引入新的伤害:把情境信号变成人格标签,把保护变成剥夺自主权,把困难信息用于销售或定价,把弱 sentiment 推断当作事实。

因此,本篇的学习重点是“情境支持”而不是“识别脆弱客户”。证据链要能重放客户说了什么、系统观察到什么、AI 建议了什么、最终客户看见/听见什么、人工如何决定、投诉如何闭环。治理边界要明确禁止诊断式推断、永久标签化、无目的敏感信息扩散和基于 vulnerability signal 的商业利用;高伤害场景需要可解释升级、人工复核、可访问替代渠道和可撤回的客户偏好管理。

问题定义

客户脆弱性不是一个稳定、永久的人格标签,而是一种情境风险:失业、丧亲、诈骗、语言障碍、无障碍需求、认知负荷、还款困难、投诉压力、家庭胁迫或服务失败,都可能使同一个客户在某个旅程中更容易受到伤害。

AI 系统的风险不只是“没有识别出 vulnerable situation”。更常见的是:

风险表现架构含义
过度推断从年龄、语速、停顿、设备、语言或行为推断 disability、capacity、mental health必须使用 situation signal,而不是 person label
过度干预因模型认为客户脆弱而阻断交易、隐藏选项、降低自主权干预强度要和可证明 harm risk 对齐
低质量帮助用同一套 bot flow 处理 hardship、bereavement、scam、complaint、accessibility需要 specialist handoff 和 journey-specific controls
证据缺口无法证明客户看到什么、AI 建议什么、员工如何处理final-channel capture 和 evidence ledger 必须进入设计
尊严损害客户被贴标签、重复陈述创伤、敏感信息扩散controlled vocabulary、role-based access、note hygiene

成熟问题不是“模型能否判断这个人 vulnerable”,而是:

What situation signals show elevated support or harm risk?
What is the least intrusive helpful response?
What choice remains with the customer?
What must be escalated to a trained human?
What data is necessary and purpose-bound?
What evidence proves fair and dignified treatment?

核心原理/方法

核心方法是 situation-aware support。字段和流程应表达支持需要,而不是客户缺陷:

support_need_type
vulnerability_context
elevated_harm_risk
requested_accommodation
customer_preference
handoff_reason
agent_assist_recommendation

避免的字段和话术:

is_vulnerable = true
mentally_unfit
elderly_risky
low_capability
problem_customer

信号分层:

Signal classExamples使用边界
Explicit customer statement“我失业了”“我看不懂”“有人让我转钱”强 support trigger,但只记录完成目的所需事实
Customer-selected preferencelarge print、screen reader、language、relay call用于等效服务,不推断 diagnosis
Interaction frictionabandon、重复错误、长时间停留、failed upload只用于 soft support offer 或 UX defect
Account/event contextmissed payment、overdraft、death notification、complaint、unusual transfer用于 routing and support,不直接不利处理
Frontline observationbranch/call-center notes、coercion concern、scam script需要 controlled vocabulary、review and QA

关键原则:

  1. accessibility 是 release gate,不是 UI checklist。
  2. plain language 不能删掉金融含义、费用、期限、权利和限制。
  3. vulnerability signal 只能用于保护和支持,不能用于高压销售、价格提取或服务降级。
  4. emotion/sentiment 是弱信号,不是诊断、投诉结论或不利决策依据。
  5. final message、handoff packet、human decision 和 complaint outcome 必须可重放。

系统/架构模型

参考架构:

customer interaction
  -> accessibility and preference layer
  -> consent / purpose / data minimization gate
  -> situation signal detection
  -> risk tiering and confidence governance
  -> inclusive UX adaptation
  -> agent-assist recommendation with guardrails
  -> human handoff / specialist escalation
  -> complaint / remediation / fraud / hardship linkage
  -> evidence ledger and QA sampling
  -> model risk, conduct risk and accessibility governance

核心组件:

Component职责
Accessibility baselineAI chat、voice、forms、upload、notifications、documents、handoff 满足可访问性测试
Preference and accommodation service保存客户明确选择的 channel、language、format、assistive preference
Signal detection layer识别 support need,但标记 source、confidence、sensitivity、allowed use
Risk tiering engine区分 informational、support、high-harm、urgent escalation
Inclusive interaction orchestratorplain language、step-by-step、channel switch、human option
Agent-assist guardrail给员工提示事实、uncertainty、prohibited actions 和 escalation basis
Specialist handoffhardship、bereavement、fraud/scam、accessibility、complaint 专业队列
Evidence ledger记录 signal、prompt、model、source、recommendation、human decision、final content
Governance loopaccessibility audit、conduct QA、model validation、complaint RCA、CAPA

关键机制与取舍

Vulnerability-sensitive architecture 的难点在于干预强度和客户自主权的平衡。

Pattern使用时机好设计坏设计
Soft support offer低置信度 friction 或客户表达困惑提供分步说明、专员、渠道切换告诉客户“你似乎无法理解”
Plain-language toggle复杂费用、拒绝、hardship、fraud、投诉说明simple / detailed 含义一致简化版隐藏关键限制
Channel switch客户请求电话、branch、relay、paper、secure message保留上下文与 case id让客户从 bot 起点重来
Safe pausescam/coercion 高风险、不可逆动作解释风险、提供复核、专员无解释冻结或羞辱客户
Specialist handoffbereavement、hardship、accessibility complaint、distresshandoff packet 减少重复陈述多部门要求客户重复创伤
Customer-controlled disclosure客户愿意保存偏好/授权明确用途、访问、保留、修改扩散到所有 CRM notes
Appeal/review pathAI 或员工建议导致 hold、限制、拒绝人工复核和投诉入口“系统判断无法继续”

Handoff tiering:

TierTriggerAI actionHuman action
0 Accessibility preference客户选择格式、语言、辅助路径应用 preference,确认是否保存普通服务团队
1 Support need困惑、流程摩擦、plain-language 请求简化解释、分步选项、渠道切换员工继续服务
2 Vulnerability contexthardship、bereavement、language barrier、distress推荐专员,限制 sales pressuretrained specialist
3 High harm riskscam/coercion、elder risk concern、重大 financial losssafe pause、fraud/hardship/complaint escalationspecialist + supervisor
4 Urgent sensitiveself-harm language、abuse/coercion immediate risk、legal-sensitive complaintapproved emergency/escalation script,停止自动化policy-defined crisis/legal path

证据与控制

Evidence ledger 最少要捕获:

case_id
customer-stated issue
support_need_type
requested accommodation or preference
signal source and confidence
data sensitivity and retention rule
AI run id / model / prompt / source
recommendation and prohibited actions shown to employee
human decision and reason
final-channel content
handoff queue and SLA
complaint/remediation link
QA result and CAPA id

控制矩阵:

Control objectiveControl activityEvidence
确保 AI channel 可访问WCAG/accessibility release gate 覆盖 chat、forms、voice、documents、handofftest report、defect closure、assistive tech transcript
避免客户标签化schema 区分 situation signal、support need、preferencedata dictionary、field access rule、QA note audit
最小化敏感数据purpose-based collection、role access、retentionprivacy review、retention config、access logs
保留自主权解释、选择、复核、complaint routefinal message、override path、review decision
降低认知负荷approved plain-language patterns、review before submissioncontent review、readability evidence、user test
升级高伤害场景scam、hardship、bereavement、accessibility complaint、distress tieringhandoff packet、specialist outcome
控制 agent-assist禁止 diagnosis、unsupported promise、coercive scriptprompt policy、eval result、employee confirmation
监控公平性friction、completion、reversal、complaints by permitted segments/proxiesmonitoring dashboard、model risk review
投诉闭环complaint record 链接 AI run、channel、content、support need、RCAcomplaint file、CAPA

指标应平衡 protection、autonomy、access、conduct 和 evidence:assistive-tech completion、repeat-story rate、handoff quality、false pause、human review reversal、complaint uphold、prohibited diagnosis rate、hallucinated policy rate、final-channel capture、CAPA aging。

金融零售/AI产品场景

  1. Collections hardship chatbot:客户说失业和看不懂方案,AI 触发 hardship path、plain-language options、specialist handoff,并阻止高压话术。
  2. Scam transfer warning:客户在电话指导下发起大额转账,系统 safe pause,解释风险,升级 fraud specialist,同时保存客户最终响应。
  3. Bereavement servicing:客户提交死亡证明后,document reuse 和 sensitive note control 避免反复要求讲述关系和上传同一文件。
  4. Elder-risk branch assist:员工不能写“老人糊涂”;系统要求可观察事实、交易异常、客户陈述和 supervisor review。
  5. Credit adverse action explanation:AI 只能用 approved reason 和 plain-language pattern 起草,不能删掉权利、期限和限制。
  6. Accessibility complaint:screen reader 无法完成 fraud dispute,系统创建 accessibility incident、alternative channel、remediation 和 product defect backlog。
  7. Wealth/insurance call:grieving 或 distressed customer 触发 sales suppression、suitability review 和 conduct QA。

反模式

反模式风险更好的控制
permanent vulnerable flagstigma、bias、privacy risksituation-based support_need with retention/access
age-only elder-risk rule不准确且可能不公平behavior/context + human review
accessibility 只做自动化扫描漏掉 keyboard、screen reader、cognitive journeymanual assistive tech QA
LLM 随意简化 regulated content遗漏费用、期限、权利、限制approved content patterns and review
scam protection 变 paternalism客户失去控制且无法解释safe pause + review/appeal
agent-assist diagnosis员工采纳 unsupported conclusionno-diagnosis prompt、QA、note hygiene
vulnerability signals 用于销售exploitation and conduct risksuppression rules
complaint 不链接 AI traceRCA 和 remediation 失效complaint schema includes AI run/content/channel
automation rate 作为主指标把困难客户困在 self-servicebalanced outcome metrics

最终心智模型

Inclusive AI 的核心不是识别“脆弱客户”,而是识别高伤害情境并提供最小侵入的支持。成熟架构要能证明:客户没有被标签化,帮助没有变成强制,AI 渠道可访问,高风险情境被有尊严地升级,隐私和自主权被保留,投诉能回到证据、模型、内容、员工和流程根因,并转化为可执行改进。


SOTA 状态标注 (2026-07-01)

本篇属于第二、三遍深读池(参考架构/深读笔记),未列入 12 周主线必读。时效基线为写作时点;引用前请按 CLAUDE.md 全局时效性硬规则复查最新进展。模块级 SOTA 对照见 docs/AI_SYSTEMATIC_LEARNING_ROADMAP_2026.md 各周「2026 SOTA 对照」行与文末「SOTA 检查」。