Back to Deal Haven

AI Policy

This policy explains how Deal Haven develops, uses, and oversees artificial intelligence (AI) on the platform. It applies to all AI features you see today and any we add in future.

What this policy covers

We use AI in three areas:

  1. Document extraction — Reading uploaded PDFs and Excel files and suggesting financial figures for your deal models. You must review and approve suggestions before they apply—nothing is forced on you.
  2. AI Deal Assistant — A chat in each deal where you can ask questions about finances, risks, and structure, and attach documents for help analysing them.
  3. Offer letter generation — Drafting indicative offer letters from your deal data. You always review and edit before saving or sending.

These features use AI language technology from OpenAI. All requests go through our secure systems—your browser does not connect straight to OpenAI; we handle that for you.

Parts of Deal Haven that do not use AI (for example core spreadsheet calculations in your models) are not covered by this policy—they use normal maths and rules, not machine learning.

For a friendly, step-by-step description of how each AI feature works in practice, see How We Use AI.

Our principles

Fairness

We watch how well AI performs across different document types, companies, and currencies so that one group of users is not left behind. Our AI works with business and financial documents, not personal decisions about individuals—but we take fairness seriously as we grow.

Transparency

You will always know when you are using AI. For extraction, you see confidence hints and sources in the document so you can judge the suggestions yourself. In chat, we clearly say the assistant is AI and that answers may be wrong.

We do not present AI output as if a human had already verified it—unless you (or your process) have actually reviewed and approved it.

Accountability

Each AI feature has an owner inside our company who is responsible for how it behaves and how we comply with rules.

We keep records of AI use so we can understand what happened, which version of the technology was used, and how outputs were used. For the Deal Assistant we keep a detailed history of interactions for security, quality, and compliance—monitored through our internal processes.

Human oversight

Document extraction: No AI figures are applied to your models without your review. Important fields—such as revenue, total assets, and EBITDA (earnings before interest, tax, depreciation, and amortisation)—need a clear decision from you even when the AI looks confident. You can reject any value or the whole import.

AI Deal Assistant: The chat is for help and thinking, not for taking actions for you. It cannot change your deal or send anything on your behalf. Every reply reminds you that it may be wrong and is not financial advice. You choose what to share and how to use the answers.

Safety and reliability

If extraction quality is too poor to trust, we stop and tell you rather than filling your model with bad numbers. Files that clearly aren’t financial statements are rejected early where we can detect that.

To protect the service and your costs, we limit how fast you can send chat messages, how large prompts and replies can be, and what the AI is allowed to do inside our product (only approved actions we’ve built). Organisations can also set optional monthly limits on AI chat spend. We delete files you uploaded to chat but never sent so they don’t sit around forever. Assistant replies are saved on our servers—not written from your browser in a way that could be tampered with.

Privacy and data protection

We use your data only for what you asked for—extracting numbers or answering your questions about a deal. Data sent to OpenAI is covered by their terms for business API use, which do not allow them to use your content to train their public models.

  • Extraction: your uploaded file is sent to OpenAI for processing.
  • Deal Assistant: your messages, the deal information needed to answer, summaries and lookups we run for you inside the product, and attachments you provide.
  • Offer letters: the selected scenario’s numbers from your model, your name, and contact details needed for the letter.

We do not use your data to train or fine-tune our own AI. Sensitive content is not written to routine application logs. You can delete chat conversations; we keep internal records of AI use for compliance and quality monitoring—no more than what was already part of the conversation.

Environmental responsibility

We choose models that balance accuracy with efficiency. We use smaller, faster models where we can and only move to heavier processing when needed. We monitor usage to avoid unnecessary waste.

Compliance

We align our AI governance with ISO/IEC 42001, an international standard for managing AI responsibly. We also follow EU and UK rules as they develop.

We document how our AI features are classified under applicable law. For example, under the EU AI Act, what we offer today is limited-risk—it is built to inform people and meet transparency expectations, not the stricter rules for high-risk systems. We review that classification when we add or materially change AI features.

Today’s features are designed to support human decisions—especially in financial analysis—not to replace human judgment or run your deal on autopilot.

Reporting concerns

If something seems wrong, unfair, or unexpected, we want to hear from you. You don’t need to be technical or be sure it’s a bug.

You can report:

  • From the import review screen — “Report an issue”
  • From the AI Deal Assistant — “Report a concern”
  • From offer letter review — “Report a concern”
  • From Settings — AI Feedback
  • By email: team@dealhaven.app

We aim to acknowledge reports within 48 hours.

What to report

Examples:

  • Numbers that are often wrong for a certain type of document
  • A single value that looks nothing like the source document
  • A chat answer that could mislead someone who trusted it
  • Analysis that contradicts the source files
  • Worries about how your data is handled
  • Anything that feels biased or unfair

Policy updates

We review this policy at least once a year and whenever we add or materially change AI features, or when regulations require it.

Last reviewed: 4 April 2026