AI disclosure

In short

Draaftly uses AI to read and to draft. It does not use AI to calculate, and it does not publish anything without a person approving it. Draaftly never uses your data to train AI models. Our AI providers, Anthropic and Google Cloud, do not use customer data from their commercial services to train their models by default.

1. Which models we use

Drafting and analysis in the Draaftly platform are done by Claude models from Anthropic, accessed through Google Cloud. Both appear on our sub-processors list. If we change provider or add one, that list changes first and customers are told.

2. What the AI does

• Reads the documents you upload and the public sources gathered for a project.
• Extracts claims, figures and requirements, and records them with their source, date and page.
• Drafts narrative: the text of a proposal, a report section, an executive summary.
• Suggests structure, gaps and questions worth asking.

3. What the AI does not do

• It does not do the arithmetic. Market sizing, cash flows, P&L, NPV, IRR, payback, DSCR and sensitivities are produced by our own calculation engine from inputs recorded in the assumptions register. The AI writes about those results; it does not generate them.
• It does not invent evidence. Sources, respondents, quotes, client names, projects and CVs come from material you provide or from sources that are recorded. Where evidence is missing, the gap is logged as a stated assumption rather than filled in.
• It does not decide anything. No automated decision with legal or similarly significant effect is made about anyone.
• It does not take instructions from your documents. Uploads are treated as data to be analysed. Text inside a document that purports to instruct the system is treated as content, not as a command.

4. Your data and the models

Draaftly never uses your data to train AI models. Our AI providers, Anthropic and Google Cloud, do not use customer data from their commercial services to train their models by default. Material sent to a model for a task is sent to produce your output, and for nothing else. Our providers’ commercial terms, not their consumer terms, apply to this processing. We do not share your material between customers, and the retrieval that finds material during a project is separated per organisation.

5. Human review

Every project moves through Draft, Review, Approved and Published. Nothing is exported as final until someone on your side approves it. Where our own team delivers a study as a service, a qualified person reviews and signs off the work before it reaches you.

6. Limits you should know about

• AI models can be wrong, and can be confidently wrong. Check anything that matters against the source.
• Sources can be out of date, or can say something slightly different from the summary of them. The source snapshot is kept so you can read the original.
• Output quality depends on what goes in. Thin evidence produces a study that depends on assumptions, which is why we label them.
• Draaftly does not promise that any lender, investor, authority or client will accept a document it helped produce.

7. Your obligations

Review output before relying on it or sending it on, keep the sources and assumptions attached where a reader needs them, and follow any disclosure rules that apply to you — your client’s, your regulator’s or, for students, your institution’s. The acceptable use policy has the detail.

8. Hisaab

Hisaab does not use AI models to process sales data. Billing, kitchen tickets, stock and reporting are ordinary software.

9. Changes

If we change how AI is used in a way that affects customers, we will update this page and tell them. Questions go tohello@draaftly.com, and the engineering view of the same subject is on ourtrust page.