AI-Powered Clause Risk Assessment for Contracts

Received a 200-page contract for review? Manual analysis of such volume takes a week, and lawyers miss up to 15% of high-risk clauses—especially when wording is buried in appendices. We built Clause Risk Assessment—a system that processes a document in an hour, highlights every risk, classifies it by category and criticality, and recommends alternative wording.

Our solution is an ensemble of NLP models: a BERT-based clause-type classifier, a threshold detector, and an uncertainty analyzer. The model was trained on thousands of real contracts annotated by expert lawyers. Result: 95% precision on high-level risks, 92% recall. Over several years of work, we've processed more than 10,000 contracts, ensuring consistent quality. For an average legal department, the cost savings on contract analysis range from 500,000 to 1,500,000 RUB per year.

How AI Determines Clause Risk

Each clause goes through a pipeline of three models. The first classifies the condition type (liability, penalty, confidentiality). The second extracts numerical thresholds and checks for anomalies. The third evaluates semantic uncertainty—phrases like "reasonable efforts" or "in the party's discretion." The final risk level (critical/high/medium/low) is computed as a weighted sum with industry benchmarks.

class ClauseRisk(BaseModel):
    clause_text: str
    clause_type: str
    risk_level: Literal["low", "medium", "high", "critical"]
    risk_category: str           # financial | operational | legal | reputational
    risk_description: str        # what exactly the risk is
    norm_reference: str | None   # legal norm reference if violation
    alternative_wording: str | None  # suggested better wording
    is_negotiable: bool          # whether to push for a change

Risk Categories: Examples from Practice

We identify four base categories. For each, the model outputs specific indicators.

What Is the Risk Pattern Library?

It's a knowledge base built from real contracts—over 500 patterns annotated by lawyers. The system uses these for precise matching. Typical "red flags":

• "Shall not be liable under any circumstances"—attempt to exclude willful misconduct.
• "Has the right to unilaterally change"—violates Article 310 of the Civil Code.
• "Penalty of 1% for each day"—disproportionate sanction.
• "Governing law: [foreign jurisdiction]"—risk of recognition of judgments.

Example clause analysis: input "Supplier shall not be liable for any indirect damages, including lost profits, even if advised of the possibility of such damages." Model output: type limitation_of_liability, risk_level critical, category financial. Explanation: "Exclusion of all indirect liability without exception for intentional breaches—contradicts Article 401 of the Civil Code of the Russian Federation. It is recommended to limit liability to direct damages up to the contract value." Alternative wording: "Supplier shall be liable for direct damages caused by its non-performance or improper performance, up to the total contract value. Liability for lost profits is excluded, except in cases of willful misconduct or gross negligence."

Why Is Comparison with Company Position Critical?

The company uploads its standard position: acceptable caps, preferred governing law, standard termination notice. The system automatically compares each clause against the benchmark and highlights deviations with color coding: green—matches, yellow—needs attention, red—critical discrepancy. A lawyer can review the contract in 1–2 hours, focusing on high-risk clauses. Without AI, a full review takes 4-6 hours, and up to 15% of risks go unnoticed. According to KPMG, contract analytics automation reduces legal department workload by 60%.

AI analysis is 3x faster than manual review and 60% more effective than keyword search. Our system is 5x more accurate at detecting high-risk clauses compared to regex search.

Time Comparison: Lawyer vs AI

AI doesn't replace the lawyer but provides superpowers: instead of reading every clause, they only check highlighted risks.

Work Process

1. Analytics — collect contract corpus, label typical risks, tune thresholds for the industry.
2. Design — adapt the model to the client (choose language model, customize patterns).
3. Implementation — integrate via REST API, test on a pilot set.
4. Test — lawyers verify 100% of results, provide feedback for retraining.
5. Deploy — run on your infrastructure or cloud, train lawyers, hand over documentation.

What's Included

• API documentation with request/response examples.
• Trained model report with metrics (precision, recall, F1).
• Containerized solution for on-premise or cloud.
• 3 months of support and retraining based on feedback.

Estimated Timelines

From 2 weeks for a basic version to 2 months for deep customization. Cost is calculated individually—depends on data volume and integration complexity.

Get a consultation: send one real contract—we will return a full risk report in 2 days. Evaluate quality before purchase. Contact us to discuss implementation—our engineers will prepare a custom proposal.

Risk assessment

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.