C

Concord

Concord's AI extracts contract terms, answers questions about agreements and proposes wording changes in its Copilot. Horizon adds custom extraction agents that populate properties across both existing agreements and future uploads within a chosen scope. Those properties can feed reports and workflow conditions, making the models part of ongoing contract operations. The platform also retains drafting, approval routing, electronic signatures, version control and renewal reminders that function independently of AI.

Its value therefore extends beyond the models, while the extraction and conversational tools form a central capability rather than an isolated writing feature. Access to Horizon matters when evaluating the agent functionality.

Explanatory notes accompany AI extracted properties so a reviewer can understand how a value was identified. The agent builder also lets users test extraction on sample agreements before activation. These are useful review aids, but Concord describes no linked support for every generated legal assertion and no reproducible accuracy benchmark. The service terms expressly acknowledge hallucinations and require customers to evaluate outputs, including human review.

A separate ethics statement recognizes incorrect, biased and misleading responses. Explanations of extracted values consequently support inspection without establishing that suggested clauses or answers have been validated against authoritative legal sources.

Customers build extraction agents by describing a desired property, reviewing the definition and testing it on up to ten sample agreements. They can limit deployment by folder, dates, creator, amount or other properties. Once activated, an agent processes future agreements in scope, with a separate action for existing agreements, and respects existing access permissions. Results include explanations for review. The terms require human evaluation of output, while later releases allow properties to drive workflow events.

The published account does not fully define which consequential downstream actions require approval or how uncertain extraction returns to a person, leaving those controls partly specified.

Customers and individual users are identified across several industries. PAAY describes importing files through Google Drive and extracting their data; Vecna Robotics describes avoiding missed dates and transposed numbers; and LeoVegas describes reviewing governing law and renewal information across its portfolio. Navarino also describes transferring completed agreements to SharePoint through Zapier. These accounts establish practical use beyond a list of logos.

Broad claims about hours saved and reduced contract delays do not include a common baseline, sample definition or independent measurement method. They should not be read as a measured accuracy result for the newer custom extraction agents or every Copilot task.

Customer data is confidential information under Concord's commercial agreement, with access restricted to personnel who need it and obligations before compelled disclosure. The agreement gives Concord broad rights to maintain and improve the service, but expressly prohibits using customer data or the customer's specific output to train its AI models. Administrators control roles, contract access and sharing permissions.

The extraction agent explanation also states that inaccessible agreements are not processed or revealed to the user. These controls address access and model use directly, although no distinct attorney client privilege undertaking or professional work product framework is provided for law firm matters.

Concord requires customers to evaluate the accuracy of AI output with human review and acknowledges that it may contain hallucinations. Its ethics principles treat AI as assistance to people and warn against entering sensitive information unrelated to the contract being handled. The product is nevertheless promoted to business teams that manage agreements without routinely involving legal counsel, and Copilot can explain risks and suggest changes.

The commercial agreement does not expressly disclaim the provision of legal advice or map those tasks to a lawyer's professional duties. Approval workflows provide organizational controls without a published jurisdiction specific framework for legal supervision.

The AI principles cover privacy, human accountability, bias mitigation and transparency. Product and engineering teams are to receive education on fairness, with training scenarios intended to represent different industries and legal frameworks. The company also promises indications when an automated system is being used and acknowledges potentially biased responses. These commitments identify concerns and broad practices, but not a named accountable risk owner, release criteria, published bias test or results across contract types and languages.

Customer feedback is part of the stated improvement process. The available account remains a principles statement rather than a verifiable program of model evaluation and oversight.

Concord specifies encryption in storage and transit, restricted employee access, customer roles, multifactor authentication, daily encrypted backups and external penetration testing twice a year. Its model provider is described as bound to zero data retention, and named subprocessors have listed processing locations. The privacy policy caps debug data at six months, allows indefinite anonymized analytics and offers a data copy on request within thirty days of termination.

Commercial terms require return or destruction of confidential information when the subscription ends. Precise backup removal and prompt schedules remain incomplete, and the incident policy does not state a customer notification deadline.

The commercial agreement provides an ordinary intellectual property indemnity for the system and caps aggregate liability at fees paid in the preceding twelve months. It also warrants material conformity with service documentation. However, the AI disclaimer expressly excludes liability for damages resulting from or relating to output and warns that output may be inaccurate or hallucinated. The customer indemnity separately includes claims arising from use of generated content.

Those provisions leave no affirmative protection for the legal correctness of an AI clause or contract analysis. The general service indemnity should therefore not be treated as a specific remedy for an AI error.

The Salesforce guide explains installation, account connection, Opportunity layout changes, template selection and fields sent from account, opportunity, contact and product objects. Agreements can be generated from Salesforce and populated with its data, with synchronization in both directions. Administrators also set the maximum sharing rights available for templates and uploaded files. This is a concrete implementation path rather than a connector logo.

Other integrations include storage services, calendars, a Word plugin, Zapier, APIs and MCP access through Claude or ChatGPT. Those connections have different plan requirements, with the Salesforce workflow available on Business and Enterprise and the open API listed on Enterprise.

Storage choices in the United States and Europe come with Concord's hosted service. Its subprocessor list places AWS hosting in the EU for European customers and North America for North American customers. The same list places OpenAI extraction in North America and Google model tooling and OCR in North America and the EU. Choosing a storage region consequently does not keep every AI processing step inside it. No private deployment or complete tenancy description is provided.

The privacy policy also permits international transfers under stated safeguards, so regional processing requirements need to be matched to the selected features.

Through its Conveyor portal, Concord offers a SOC 2 Type II report and states participation in the Cloud Security Alliance STAR Level One program. The latter is a self assessment rather than an independent audit. It also states certification under the Data Privacy Framework and its British and Swiss extensions. Separate references to AWS certifications describe infrastructure assurance and should not all be attributed to Concord itself.

The public security explanation does not identify the SOC auditor, examination period or full service scope. Encryption, penetration testing and incident procedures provide additional operational detail without establishing the quality of AI contract interpretation.

OpenAI is named as a provider for AI based extraction, Google for model tooling and OCR, and LangSmith for model tooling. The subprocessor list supplies processing locations and transfer mechanisms for those services, alongside AWS hosting and other operational suppliers. Concord also says its AI provider has agreed to zero data retention. The disclosure does not identify every model version or explain which provider handles each Copilot and agent operation.

There is no public commitment to advance notice before a model version or supplier changes. Buyers can therefore identify important suppliers while still lacking the complete model configuration for a particular workflow.

Essentials costs $499 a month with annual billing and includes five users, AI Copilot, extraction and unlimited agreements, signatures, viewers and guests. Business costs $899 and Enterprise $1,299 on the same billing basis, each also including five users. Extra users cost $49, $69 or $89 monthly respectively. The plan comparison identifies workflow, integration and administration differences, with volume discounts offered on Enterprise.

Commercial terms cap renewal price increases at five percent and require sixty days of notice for cancellation or downgrade. A separate implementation fee is not clearly stated, so the published subscription amounts may not cover every possible deployment charge.

Legal, procurement, finance, sales, HR, operations and IT teams handling their organization's agreements are all addressed by Concord. Essentials targets a small business or one department; Business supports multiple teams; Enterprise adds subsidiary management and other controls for larger organizations. Named customers span robotics, gaming, construction, payments and education. Templates, approvals and guest collaboration let legal teams set boundaries while colleagues conduct routine contract work.

The claim to support any contract is not accompanied by a tested matrix of languages, jurisdictions or specialist legal issues. Its main audience is corporate contracting, without a detailed published workflow for law firms delivering advice across separate client matters.