Contract Intelligence: Workflow, Use Cases & Checklist合同智能：实施流程、应用场景与检查清单

Contract intelligence connects agreements to daily business decisions合同智能把协议与日常业务决策连接起来

Contract intelligence is the operating capability that turns agreements and related records into structured, searchable, traceable information for business action. It connects the authoritative document with parties, entities, products, value, term, renewals, notices, obligations, rights, risks, amendments, owners, evidence, and events. The aim is not to replace the contract; it is to make the contract population usable without losing the source.

合同智能是一项运营能力，它把协议和相关记录转换为用于业务行动的结构化、可检索、可追溯信息，并将权威文档与主体、实体、产品、金额、期限、续签、通知、义务、权利、风险、修订、负责人、证据与事件连接。目标不是替代合同，而是在不丢失来源的情况下让合同集合可以被业务使用。

A useful service answers practical questions: Which agreements renew next quarter? Which supplier obligations support a critical service? Who owns a customer notice? Which contracts contain a control-change restriction? What evidence proves performance? Which fields are verified and which remain uncertain? Every answer should lead back to the relevant document, version, text, reviewer, and update history.

有用的服务能够回答实际问题：下季度哪些协议续签？哪些供应商义务支持关键服务？谁负责客户通知？哪些合同包含控制权变更限制？什么证据可以证明履约？哪些字段已经验证，哪些仍不确定？每个答案都应回到相关文档、版本、原文、审核人和更新历史。

Distinguish intelligence from automation, review, and analytics区分合同智能、自动化、审查与分析

Legal document automation creates documents from templates, data, and rules. Automated contract review analyzes text for issues during review. Contract analytics aggregates data to measure patterns, performance, and trends. Contract intelligence focuses on the connected, queryable contract record used to answer questions and drive actions throughout the agreement life cycle. One system may support several capabilities, but the operating models remain different.

法律文档自动化依据模板、数据与规则创建文档；自动化合同审查在审核过程中分析文字与问题；合同分析汇总数据以衡量模式、绩效和趋势；合同智能则聚焦贯穿协议生命周期、可供查询并推动行动的关联合同记录。一个系统可能支持多种能力，但其运营模型仍然不同。

Define what the service will and will not do. It may identify renewal dates, route obligation reminders, support portfolio searches, or prepare a first-pass risk list. It should not silently determine legal conclusions, approve payments, contact counterparties, alter contractual rights, or treat extracted data as more authoritative than executed text. Publish these boundaries for users and downstream integrations.

定义服务会做什么和不会做什么。它可以识别续签日期、发送义务提醒、支持组合搜索或准备第一轮风险清单，但不能静默作出法律结论、批准付款、联系相对方、改变合同权利，或让提取数据优先于已签文字。应向用户和下游集成公布这些边界。

Choose the business problem before the platform先选择业务问题，再选择平台

A strong business case starts with observed failure: missed renewals, late notices, slow due diligence, uncertain obligations, duplicate spend, inconsistent supplier records, delayed revenue, poor handoffs, incomplete evidence, or repeated manual searches. Identify the affected population, users, frequency, handling time, delay, error, financial exposure, service impact, and current control.

良好的商业论证从可观察失败开始：错过续签、通知迟延、尽调缓慢、义务不清、重复支出、供应商记录不一致、收入延迟、交接不良、证据不完整或重复人工搜索。识别受影响合同集合、用户、频率、处理时间、延误、错误、财务风险、服务影响与现有控制。

Prioritize use cases where the answer can be defined, source documents exist, ownership is available, actions are repeatable, and value can be measured. Avoid promising universal search across every contract before resolving document completeness, language, scans, amendments, entity mapping, access, and validation. A narrow service that users trust can expand; a broad service that produces uncertain answers loses adoption quickly.

优先选择答案可定义、源文档存在、负责人明确、行动可重复且价值可衡量的场景。在解决文档完整性、语言、扫描件、修订、实体映射、访问与验证前，不要承诺覆盖所有合同的通用搜索。用户信任的窄服务可以扩展；产生不确定答案的宽服务则会迅速失去采用。

Establish the authoritative contract population建立权威合同集合

Bring together signed agreements, amendments, renewals, schedules, order forms, statements of work, side letters, notices, guarantees, approvals, and incorporated terms. Reconcile contract repositories with procurement, sales, finance, entity, property, HR, vendor, customer, and records systems. Record missing, unsigned, duplicated, superseded, corrupted, inaccessible, or uncertain documents rather than hiding gaps.

汇集已签协议、修订、续签、附表、订单、工作说明、附函、通知、保证、审批和纳入条款。把合同库与采购、销售、财务、实体、不动产、人力、供应商、客户和档案系统核对。记录缺失、未签、重复、被替代、损坏、不可访问或不确定的文档，而不是隐藏缺口。

Assign a stable agreement identifier and connect every related document. Identify the executed master, effective amendments, current schedules, applicable orders, and authoritative version. Preserve original files, hashes or equivalent integrity evidence, source location, ingestion date, access classification, retention category, and chain of custody where required. Intelligence is unreliable when the document hierarchy is wrong.

分配稳定的协议标识并连接所有相关文档。识别已签主协议、有效修订、当前附表、适用订单与权威版本。保留原始文件、哈希或同等完整性证据、来源位置、导入日期、访问分类、保留类别，以及需要时的保管链。文档层级错误时，合同智能不会可靠。

Design a data model around decisions and actions围绕决策与行动设计数据模型

Model agreements, documents, parties, legal entities, business units, products, locations, monetary values, dates, obligations, rights, clauses, risks, notices, events, amendments, owners, approvals, evidence, and tasks as connected objects. Define stable identifiers and relationships. A renewal belongs to an agreement; a notice requirement may relate to an event; an amendment changes specified provisions; one obligation may depend on several documents.

把协议、文档、主体、法律实体、业务单元、产品、地点、金额、日期、义务、权利、条款、风险、通知、事件、修订、负责人、审批、证据与任务建模为关联对象，并定义稳定标识与关系。续签属于某份协议；通知要求可能与事件相关；修订改变指定条款；一项义务可能依赖多份文档。

For every field, define meaning, type, format, allowed values, source, extraction method, reviewer, confidence, effective period, sensitivity, owner, update trigger, and downstream use. Separate observed text, normalized value, human conclusion, and business status. For example, a contract date may be extracted from text, normalized to a calendar value, validated by a reviewer, and then used to create an approved reminder.

为每个字段定义含义、类型、格式、允许值、来源、提取方法、审核人、置信、生效期间、敏感性、负责人、更新触发与下游用途。区分观察到的文字、标准化值、人工结论与业务状态。例如，合同日期可以从文字提取、标准化为日历值、由审核人验证，再用于建立批准的提醒。

Combine machine extraction with risk-based human validation把机器提取与基于风险的人工验证结合

Use optical recognition, parsing, rules, and language models to identify candidate fields and relevant text, but preserve the source and confidence. Test scans, tables, handwritten changes, long schedules, multiple languages, defined terms, cross-references, amendments, and competing dates. Require “not found” or “uncertain” when evidence is insufficient.

使用光学识别、解析、规则与语言模型识别候选字段和相关文字，但应保留来源与置信。测试扫描件、表格、手写修改、长附表、多语言、定义术语、交叉引用、修订与冲突日期。证据不足时应返回“未找到”或“不确定”。

Set validation levels by use. Search and discovery may tolerate a clearly labeled candidate. A renewal notice, payment, regulatory obligation, termination decision, or transaction dependency needs stronger human confirmation. Sample low-risk fields, fully review high-impact fields, and route anomalies. Record false positives, false negatives, corrections, and root causes to improve the service.

根据用途设置验证级别。搜索与发现可以容忍明确标记的候选结果；续签通知、付款、监管义务、终止决定或交易依赖则需要更强人工确认。抽查低风险字段，全面复核高影响字段，并路由异常。记录误报、漏报、更正与根因，以改进服务。

Connect clauses, entities, obligations, events, and amendments连接条款、实体、义务、事件与修订

Search becomes more useful when records are connected. A control-change question should find the affected entity, parent relationships, agreement, definition, assignment language, related termination right, notice mechanism, amendment, counterparty, business owner, and transaction event. A security-incident question should connect notification deadlines, affected data, responsible team, customer contacts, insurance, and evidence.

记录相互连接时，搜索才更有用。控制权变更问题应找到受影响实体、母公司关系、协议、定义、让与文字、相关终止权、通知机制、修订、相对方、业务负责人和交易事件。安全事件问题则应连接通知期限、受影响数据、责任团队、客户联系人、保险与证据。

Use source-backed search and filters rather than presenting generated summaries alone. Users should be able to move from a portfolio result to the agreement, field, source excerpt, confidence, reviewer, and change history. Apply permissions at the object and document level so a useful relationship graph does not expose restricted employment, investigation, pricing, personal, or privileged information.

使用有来源支持的搜索和筛选，而不是只展示生成摘要。用户应能从组合结果进入协议、字段、来源摘录、置信、审核人与变更历史。对对象和文档应用权限，使有用的关系图不会暴露受限的雇佣、调查、定价、个人或特权信息。

Turn verified obligations into controlled work把经过验证的义务转换为受控工作

An obligation record should identify the responsible party, beneficiary, required action, conditions, start, due date, recurrence, evidence, dependency, exception, approval, notice, remedy, source, owner, backup owner, status, and closure rule. Separate a contractual deadline from an internal target and calendar both. Do not create tasks from uncertain extraction without visible review status.

义务记录应识别责任方、受益方、所需动作、条件、开始、到期、重复、证据、依赖、例外、审批、通知、救济、来源、负责人、后备负责人、状态与关闭规则。区分合同期限与内部目标，并把两者加入日历。不要依据不确定提取建立没有明显审核状态的任务。

Design reminders, escalation, reassignment, delegation, completion evidence, review, and reopening. Integrate with work systems only when identifiers and permissions are reliable. When an amendment, renewal, termination, entity change, product change, incident, or new law affects the record, update the contract view and downstream tasks through an approved change process.

设计提醒、升级、重新分配、授权、完成证据、复核与重新打开。只有标识和权限可靠时才与工作系统集成。当修订、续签、终止、实体变化、产品变化、事件或新法律影响记录时，通过批准的变更流程更新合同视图与下游任务。

Define a service for each business use case为每个业务场景定义服务

Each service needs a defined request, permitted users, data fields, search logic, validation level, response format, service target, escalation, owner, and feedback loop. Avoid a single unlabeled search box that makes every result look equally authoritative.

每项服务都需要明确申请、允许用户、数据字段、搜索逻辑、验证级别、响应格式、服务目标、升级、负责人和反馈循环。避免使用一个没有标签的搜索框，让所有结果看起来具有同等权威性。

Build contract intelligence in controlled stages分阶段建立受控合同智能

1. Select one business service.选择一项业务服务。 Define users, contract population, questions, decisions, actions, baseline, value, and risk boundaries.定义用户、合同集合、问题、决策、行动、基线、价值与风险边界。
2. Reconcile documents and hierarchy.核对文档与层级。 Collect authoritative agreements and related records, identify gaps, assign IDs, versions, access, and retention.收集权威协议与相关记录，识别缺口，分配标识、版本、访问与保留。
3. Design the minimum data model.设计最小数据模型。 Define objects, fields, relationships, sources, validation, owners, confidence, and update triggers.定义对象、字段、关系、来源、验证、负责人、置信与更新触发。
4. Extract, validate, and test.提取、验证并测试。 Use representative documents, edge cases, amendments, languages, permissions, corrections, and end-to-end actions.使用代表性文档、边界案例、修订、语言、权限、更正与端到端行动。
5. Launch, operate, and expand.上线、运营并扩展。 Monitor adoption, quality, actions, exceptions, access, changes, and value before adding more fields or populations.在增加字段或合同集合前，监控采用、质量、行动、例外、访问、变更与价值。

Govern the intelligence record as sensitive business data把智能记录作为敏感业务数据治理

Classify documents, fields, embeddings, indexes, prompts, outputs, logs, and exports. Apply least privilege, role and matter access, segregation of duties, strong authentication, encryption, monitoring, controlled sharing, retention, legal hold, deletion, backup, recovery, and incident response. Confirm where data is processed, which vendors and subprocessors are involved, and how the organization exits with documents, metadata, relationships, audit history, and corrections intact.

对文档、字段、嵌入、索引、提示、输出、日志与导出分类。采用最小权限、角色和事项访问、职责分离、强身份验证、加密、监控、受控共享、保留、法律保全、删除、备份、恢复与事件响应。确认数据在哪里处理、涉及哪些供应商与分处理者，以及组织如何在保留文档、元数据、关系、审计历史与更正的情况下退出。

Create ownership for the document population, taxonomy, fields, extraction models, validation rules, business services, integrations, and permissions. Every change should include purpose, impact, approval, test evidence, release, rollback, and communication. Review access and data quality regularly. A verified field can become stale when a contract is amended or a business event occurs.

为文档集合、分类法、字段、提取模型、验证规则、业务服务、集成与权限建立负责人。每项变更都应包含目的、影响、批准、测试证据、发布、回滚与沟通。定期复核访问与数据质量。合同修订或业务事件发生后，已验证字段也可能过时。

Measure answer quality, action quality, and business value衡量答案质量、行动质量与业务价值

Track document coverage, hierarchy completeness, field population, citation coverage, validation status, accuracy by field, false positives, false negatives, stale records, correction time, unresolved items, and permission incidents. Measure retrieval usefulness, not just search volume: did the answer cite the right source, identify uncertainty, and support the intended decision?

跟踪文档覆盖、层级完整性、字段填充、引用覆盖、验证状态、按字段准确率、误报、漏报、过期记录、更正时间、未解决事项与权限事件。衡量检索有用性，而不只是搜索量：答案是否引用正确来源、识别不确定性并支持预期决策？

Pair quality with operational outcomes such as renewal decisions completed on time, notices sent correctly, obligations closed with evidence, due-diligence response time, manual search time, avoided duplicate work, revenue or savings supported, and incidents prevented or contained. Include platform, integration, migration, validation, legal review, operations, training, support, and governance costs when evaluating return.

把质量与运营结果结合，包括按时完成续签决定、正确发送通知、带证据关闭义务、尽调响应时间、人工搜索时间、避免重复工作、支持的收入或节省，以及预防或控制的事件。评估回报时纳入平台、集成、迁移、验证、法律审核、运营、培训、支持与治理成本。

Evaluate solutions with a representative contract population使用代表性合同集合评估解决方案

Test signed originals, amendments, scans, tables, multilingual agreements, multiple entities, unusual dates, overlapping terms, sensitive documents, and missing records. Run real questions for renewal, notice, payment, data, audit, control change, termination, and obligation evidence. Require movement from result to source and test corrections, permissions, exports, integration failures, and recovery.

测试已签原件、修订、扫描件、表格、多语言协议、多实体、异常日期、重叠期限、敏感文档与缺失记录。针对续签、通知、付款、数据、审计、控制权变化、终止与义务证据运行真实问题。要求从结果回到来源，并测试更正、权限、导出、集成失败与恢复。

Assess configuration, data model flexibility, source citations, validation workflow, search filters, relationship mapping, obligation operations, language support, audit, access, integrations, portability, vendor controls, implementation ownership, service levels, and total cost. A polished answer to one sample contract is not evidence that the service can govern a changing portfolio.

评估配置、数据模型灵活性、来源引用、验证工作流、搜索筛选、关系映射、义务运营、语言支持、审计、访问、集成、可移植性、供应商控制、实施责任、服务水平与总成本。对一份示例合同给出精美答案，不能证明服务能够治理不断变化的合同组合。

Connect intelligence to review and document workflows把合同智能连接到审查与文档流程

Use a complete contract review checklist to define the clauses, dependencies, and evidence that matter during intake and validation. When approved data and clauses are used to create new documents, apply the controlled template, rule, approval, testing, and handoff process described in the legal document automation workflow.

使用完整的合同审查清单定义导入与验证时重要的条款、依赖与证据。当经过批准的数据和条款用于创建新文档时，应采用法律文档自动化流程中的受控模板、规则、审批、测试与交接。

Keep the authoritative executed agreement at the center. Generated documents, extracted fields, summaries, analytics, tasks, and integrations are governed views or downstream actions. They should remain connected to the source and should never silently overwrite it.

始终以权威已签协议为中心。生成文档、提取字段、摘要、分析、任务与集成是受治理视图或下游行动，应保持与来源连接，并且不能静默覆盖来源。

Avoid turning uncertain extraction into confident action避免把不确定提取变成确定行动

Common failures include buying a platform before selecting a service, importing incomplete contracts, ignoring amendments, defining hundreds of unused fields, hiding confidence, treating summaries as sources, granting broad access, creating reminders from unverified dates, sending duplicate tasks, failing to update after amendments, and measuring only documents ingested or searches run.

常见失败包括：选择服务前购买平台；导入不完整合同；忽略修订；定义数百个未使用字段；隐藏置信；把摘要当作来源；授予宽泛访问；依据未验证日期建立提醒；发送重复任务；修订后不更新；以及只衡量导入文档或搜索次数。

Escalate missing authoritative documents, conflicting dates, unclear parties, failed optical recognition, unsupported languages, extracted legal conclusions, sensitive information exposed through search, obligations without owners, material actions without approval, untraceable corrections, and vendor claims that cannot be reproduced on representative data. Convenience cannot substitute for reliable evidence.

权威文档缺失、日期冲突、主体不清、光学识别失败、不支持语言、提取出法律结论、敏感信息通过搜索暴露、义务没有负责人、重大行动没有审批、更正无法追溯，以及供应商主张无法在代表性数据上复现，都应升级处理。便利不能替代可靠证据。

Use AI to build a traceable first-pass contract record使用 AI 建立可追溯的第一轮合同记录

AI can classify documents, detect candidate fields, extract relevant text, compare amendments, identify possible obligations, create preliminary relationships, and answer source-backed questions. Provide the complete agreement hierarchy, data dictionary, requested business question, output format, validation rules, and allowed source population. Require citations, confidence, and “not found” for missing evidence.

AI 可以分类文档、识别候选字段、提取相关文字、比较修订、识别潜在义务、建立初步关系并回答有来源支持的问题。应提供完整协议层级、数据字典、业务问题、输出格式、验证规则与允许来源集合，并要求引用、置信以及证据缺失时返回“未找到”。

AI cannot verify business facts, determine applicable law, approve a legal conclusion, contact a counterparty, authorize payment, change rights, or replace qualified review. Human owners must confirm the complete contract, current law, facts, hierarchy, extracted data, obligations, permissions, decisions, and actions. Content and tools support information organization and pre-signature risk triage; they do not constitute legal advice.

AI 不能验证业务事实、决定适用法律、批准法律结论、联系相对方、授权付款、改变权利或替代具备资质的审核。人工负责人必须确认完整合同、现行法律、事实、层级、提取数据、义务、权限、决定与行动。内容与工具仅支持信息整理和签署前风险初筛，不构成法律意见。

Contract intelligence FAQ合同智能常见问题

Sources and Method来源与方法

• NIST — Artificial Intelligence Risk Management Framework美国国家标准与技术研究院 — 人工智能风险管理框架
• UK Information Commissioner's Office — Artificial intelligence and data protection guidance英国信息专员办公室 — 人工智能与数据保护指南
• US National Archives — Records management policy and guidance美国国家档案馆 — 档案管理政策与指南
• US Federal Acquisition Regulation — Subpart 4.8, Government Contract Files美国《联邦采购条例》— 第 4.8 分部分，政府合同档案