PwC and AWS unveil an AI contract-review system that cuts review time by up to 90%

PwC and AWS have unveiled AIDA, a contract analysis system that turns long, unstructured contracts into structured data and natural-language answers. The platform runs on AWS and uses Amazon Bedrock to find key terms, deadlines, and obligations, so nobody has to dig through dozens of pages by hand.

How AIDA works

AIDA targets legal, compliance, and procurement teams, where contract volume grows faster than the capacity to read them. Instead of conventional keyword search, the system combines OCR, custom extraction rules, and LLMs in Amazon Bedrock. Teams get more than matched phrases. They get structured fields, each tied to a specific passage in the contract. That takes some of the load off teams that usually compile this data by hand for approvals, record-keeping, and reporting.

For the demo, PwC loaded samples from CUAD, an open legal dataset. AIDA then builds a semantic representation of the documents, indexes it, and uses Retrieval-Augmented Generation to answer questions grounded in the contract text. This matters in legal work. When the question involves deadlines, penalties, or the parties' obligations, users can check an answer against a link to the source paragraph instead of taking the model's word for it.

Three core modes

AIDA is not a single chat-with-your-PDF interface. Its core idea is a set of modes for different stages of contract analysis. AWS highlights three baseline scenarios that save the most time: bulk template-based field extraction, targeted questions about a single document, and cross-document search across a whole set of agreements. As a result, the service works both for the initial tagging of a large contract portfolio and for quick checks during negotiations or audits.

• Template-based extraction: a team defines fields such as termination period, renewal terms, or usage rights once, then applies the same rules to thousands of contracts.
• Single-document chat: users can ask about a specific date, obligation, or restriction and get an answer anchored to the relevant spot in the text.
• Project-wide global chat: AIDA compares several contracts at once, finds common clauses and differences in obligations, and consolidates the findings into a single answer.
• Metadata filtering: results can be narrowed by contract type, date, business unit, or jurisdiction to keep different classes of documents separate.

According to PwC, AIDA cut manual contract review time by up to 90% in client deployments. AWS also cites a major film and TV studio, where the system similarly sped up licensing rights research. It identifies broadcast, streaming, theatrical, and derivative rights, helping the studio quickly understand what it can show, repackage, or take to new markets. This matters most in media and entertainment, where a rights mistake usually costs more than the automation itself.

Security and system integrations

AIDA is architected as an enterprise service, not an experimental prototype. AWS WAF, a load balancer, and NGINX sit at the front end. Amazon Cognito handles access and connects to Microsoft Entra ID or Okta, and permissions can be configured at both the application and project levels. Data is encrypted in transit and at rest: documents and OCR output live in Amazon S3, while structured results go to Amazon RDS.

To handle large volumes, the system runs on Amazon ECS on AWS Fargate with an Amazon SQS queue, so uploading hundreds or thousands of documents doesn't block the interface. Vector search runs on Amazon Bedrock Knowledge Bases and Amazon OpenSearch Serverless, while Guardrails in Amazon Bedrock handle content filtering, personal data protection, and prompt security. AIDA can add a human-in-the-loop review step before sending results to downstream systems, and the data then flows into CLM, ERP, CRM, and data stores via Lambda, EventBridge, and SQS.

What it means

PwC and AWS are showing how GenAI is moving out of demos and into narrow, business-critical processes where errors are expensive. If the system really does extract fields reliably, answer with citations, and plug into CLM or ERP, legal AI starts paying for itself through concrete economics rather than promises: shorter review cycles, less manual work, and faster identification of risks, deadlines, and obligations in contracts.