DataRight

Inspiration

Companies increasingly build revenue-generating products using third-party datasets, AI models, APIs, and software. However, the rights governing these components are scattered across licences, customer agreements, order forms, and amendments.

During due diligence, finding a relevant clause is only the beginning. Teams must also determine whether the document is signed, whether it covers the correct legal entity, whether a later amendment changed the original restriction, and which products and revenues depend on that right.

We built DataRight to answer one practical question:

Does the company actually have the contractual rights required to generate its revenue?

What it does

DataRight transforms fragmented contracts into a traceable, time-aware contract memory.

The platform provides five connected workflows:

• Document Library: import contracts, licences, order forms, and amendments. Text is extracted page by page so every finding remains connected to its source.
• Rights Register: organise permissions, explicit restrictions, and issues requiring legal review.
• Amendment Detection: identify when a newer document may modify an earlier clause while requiring human validation before applying the change.
• Contract Memory: ask questions in natural language and receive a conclusion with the applicable conditions, document, page, and exact supporting passage.
• Economic Dependency Mapping: connect contractual rights to technology components, products, and annual recurring revenue.

DataRight also knows when it cannot answer. If a document is unsigned, covers another legal entity, or does not contain enough evidence, the system returns “insufficient evidence” instead of inventing a conclusion.

Demonstration scenario

Our synthetic due diligence data room contains:

• 8 contractual documents;
• 41 extracted rights;
• 5 technology components;
• 3 revenue-generating products;
• €700,000 in annual recurring revenue.

The original Atlas data licence prohibits commercial use. DataRight detects that amendment D05 may authorise limited commercial use for ClauseLens Enterprise from 1 July 2026.

The amendment is not applied automatically. A user must first review the original clause, the proposed replacement, the effective date, and the system’s confidence score.

After human validation, €300,000 of ARR moves from “needs review” to a documented commercial permission subject to conditions.

The remaining exposure is clearly separated:

• €240,000 of ARR remains associated with an explicit contractual restriction affecting DealRoom Search.
• €160,000 of ARR remains supported by insufficient evidence because one agreement covers the wrong legal entity and another document is an unsigned draft.

These amounts represent revenue connected to contractual issues requiring review. They are not presented as certain losses, provisions, or valuation adjustments.

How we built it

We built DataRight as a web application combining document processing, structured contract data, retrieval-augmented generation, and financial dependency mapping.

Each document is recorded with its parties, legal entity, effective date, signature status, and relationship to other documents. Extracted rights retain their source document, page, clause, classification, and applicable conditions.

Contract Memory retrieves relevant passages and generates a structured answer containing:

1. A conclusion;
2. The applicable conditions;
3. The relevant date and entity;
4. The document, page, and clause;
5. The exact supporting passage;
6. The evidence status.

The financial model connects components to products and recurring revenue. Each product is counted only once in its most restrictive applicable category, preventing the same revenue from being counted multiple times.

Challenges we faced

The main challenge was preserving legal context across document versions. A later amendment may replace one clause while leaving the rest of the original agreement in force.

We therefore had to model:

• Effective dates;
• Parent-document relationships;
• Clause-level amendment scope;
• Signature status;
• Legal entities;
• Product-specific permissions;
• Continuing restrictions.

Another challenge was controlling AI uncertainty. Semantic similarity alone is not enough to establish a contractual right. We designed abstention as a core feature so the system can clearly explain what evidence is missing.

Finally, we had to translate legal findings into useful financial information without presenting contractual exposure as a certain financial loss.

Accomplishments we are proud of

• Building a working end-to-end contract due diligence prototype;
• Keeping every answer connected to an exact source;
• Combining legal rights with product and revenue dependencies;
• Detecting amendment relationships while preserving human control;
• Handling dates, legal entities, unsigned documents, and missing evidence;
• Creating independent test questions designed to expose hallucinations and scope errors.

What we learned

We learned that trustworthy legal AI requires more than retrieving similar text. Reliable answers depend on document hierarchy, dates, entities, signatures, product scope, and the ability to abstain.

We also learned that contractual analysis becomes much more actionable when it is connected to economic dependencies. Legal, finance, and product teams can then prioritise remediation using the same evidence.

What’s next

Our next steps are to:

• Support scanned agreements with OCR and table extraction;
• Add role-based approval workflows and audit logs;
• Compare clauses across multiple amendments;
• Integrate with virtual data rooms and contract management platforms;
• Benchmark extraction and abstention accuracy on a larger reviewed dataset;
• Generate remediation tasks directly from unresolved contractual issues.

DataRight turns contracts into decisions that remain connected to their evidence, their effective date, and the revenue they protect.

Built With

• artificial-intelligence
• contract-analysis
• document-processing
• due-diligence
• explainable-ai
• financial-analysis
• legal-tech
• natural-language-processing
• pdf
• retrieval-augmented-generation
• web-application