PwC and AWS Demonstrate AI System for Contract Analysis with 90% Verification Reduction

PwC and AWS presented AIDA — a system for contract analysis that transforms long unstructured contracts into structured data and natural language answers. The platform is built on AWS and uses Amazon Bedrock to find key terms, deadlines, and obligations without manual searching through dozens of pages.

How AIDA Works

AIDA is designed for lawyers, compliance teams, and procurement — areas where the volume of contracts grows faster than the resources to read them. Instead of traditional keyword search, the system combines OCR, custom extraction rules, and LLM models in Amazon Bedrock. The team receives not just found phrases, but structured fields linked to specific contract fragments. This removes some of the manual burden from teams that typically collect such data by hand for reconciliation, accounting, and reporting.

In a demo, PwC uploaded samples from the open legal dataset CUAD into the system. AIDA then builds a semantic representation of the documents, indexes it, and uses Retrieval-Augmented Generation to answer questions based on the contract text. This is important for legal scenarios: the answer can be verified by a reference to the original paragraph, rather than trusting the model at its word when it comes to deadlines, penalties, or party obligations.

Three Key Modes

The main idea behind AIDA is not a single interface for chatting with a PDF, but several modes for different stages of contract analysis. AWS highlights three basic scenarios that deliver the greatest time savings:

• Template extraction — the team sets fields once, such as termination date, renewal conditions, or usage rights, and then applies the same rules to thousands of contracts.
• Chat on a single document — you can ask the system about a specific date, obligation, or restriction and get an answer tied to the needed location in the text.
• Project-wide chat — AIDA compares multiple contracts at once, searches for common clauses, differences in obligations, and consolidates findings into a single answer.
• Metadata filtering — results can be narrowed by contract type, date, business unit, or jurisdiction to avoid mixing different document classes.

According to PwC, in client deployments AIDA reduced manual contract review time by up to 90%. AWS separately cites an example of a major film and television studio where the system similarly accelerated license rights research: it identifies broadcast, streaming, theatrical, and derivative rights to quickly understand what can be shown, repackaged, or launched in new markets. For media and entertainment this is especially important, because the cost of a rights error is usually higher than the cost of automation itself.

Security and System Integrations

AIDA's architecture is built as an enterprise service, not an experimental prototype. It features AWS WAF, a load balancer, and NGINX on the input, access is managed through Amazon Cognito with integration to Microsoft Entra ID or Okta, and permissions can be configured at both the application and project level. Data is encrypted in transit and at rest: documents and OCR extractions are stored in Amazon S3, structured results in Amazon RDS.

To handle large volumes, the system uses Amazon ECS on AWS Fargate and Amazon SQS queues, so loading hundreds and thousands of documents doesn't block the interface. Vector search is built on Amazon Bedrock Knowledge Bases and Amazon OpenSearch Serverless, and Guardrails in Amazon Bedrock handle content filtering, personal data protection, and prompt security. Before sending results to downstream systems, AIDA can include human-in-the-loop verification, and the data itself is passed to CLM, ERP, CRM, and repositories through Lambda, EventBridge, and SQS.

What This Means

PwC and AWS demonstrate how GenAI moves from demonstrations to narrow, expensive-to-business processes. If the system truly and consistently extracts fields, answers with citations, and integrates into CLM or ERP, then legal AI starts to pay for itself not with promises, but with concrete economics: shorter review cycles, less manual labor, faster risk identification, and clarity on deadlines and obligations in contracts.