When Apex Financial approached us, their legal team was processing an average of 340 contracts per month. Each contract required manual review against a 47-point compliance checklist, cross-referencing with current regulatory guidance, and flagging obligations for the relationship management team.

The average review time per contract was 3.2 hours. With a team of 12 lawyers and paralegals, contract review was consuming over 40% of total legal department capacity.

The Business Problem

The problem wasn’t just cost. It was throughput and risk.

When deal flow increased, the legal team became the bottleneck. Contracts queued for 10–14 days before review began. In competitive situations, this delay cost the firm deals. In one documented case, a competitor closed a transaction while Apex’s legal team was still reviewing the initial term sheet.

The risk dimension was equally concerning. Human reviewers, working under time pressure on repetitive checklists, miss things. An internal audit had identified missed obligations in 12% of reviewed contracts — a rate that was acceptable as an industry benchmark, but not to Apex’s legal leadership.

What We Built

The solution was an AI agent system designed around three core components:

Document ingestion and normalisation. Contracts arrived in various formats — PDFs, Word documents, scans of physical originals. The ingestion layer converted all formats to a normalised text representation, preserving document structure and handling multi-document packages (main agreement plus schedules, exhibits, and amendments).

LLM-powered compliance review. The core analysis layer used a RAG system built over Apex’s regulatory knowledge base, updated weekly from official sources. For each contract, the system worked through the 47-point checklist, retrieving relevant regulatory guidance for each point, generating a structured analysis with confidence scores, and flagging items requiring human review.

Obligation extraction and routing. The system extracted time-sensitive obligations (payment schedules, renewal notice periods, reporting deadlines) and automatically populated the relationship management team’s tracker. For obligations flagged with high confidence, this happened without human review in the loop.

Results After 90 Days

The headline numbers from the 90-day review post-deployment:

  • Review time per contract: reduced from 3.2 hours to 28 minutes average (including human review of flagged items)
  • Error rate (missed obligations): reduced from 12% to 4% — a 67% improvement
  • Queue time: from 10–14 days to same-day for standard contracts
  • Legal team capacity freed: 62% of contract review time, redirected to higher-complexity advisory work

At Apex’s billing rates, the capacity freed by the system represented $2.1M in annualised value — partly through cost avoidance (headcount that would have been added to handle deal flow growth) and partly through recovered deal velocity.

What Made It Work

Several factors made this deployment successful that are worth noting for teams evaluating similar applications:

The compliance checklist was already well-defined. Apex had spent years refining their 47-point framework. We didn’t have to build the decision logic from scratch — we had to automate the application of it. Projects that require the AI to define the review criteria, not just apply them, are significantly harder.

The human review layer was designed thoughtfully. We didn’t try to remove human lawyers from the process — we redesigned what they spent their time on. The system surfaced the highest-risk flags for human review and handled the routine with confidence. Lawyers trusted the system because they understood where its boundaries were.

We spent the first three weeks on data quality. Before any model training, we audited five years of historical contracts and compliance decisions. We found that 23% of the training data had annotation inconsistencies that would have degraded model performance. Fixing this before training was essential.

If you’re evaluating AI for legal or compliance document review, the key questions are: how well-defined is your compliance framework? How consistent is your historical review data? And how much human review do you need to maintain for regulatory and liability reasons? We’re happy to work through these with you.

Request a free consultation with our legal AI team.