How we use AI

1. What this page covers

hibretOne CIC (company number 14228154, registered office Studio 9, 50-54 St Paul's Square, Birmingham B3 1QS) builds and runs Seghu, ScaleupKickstarter and 1001Trades. We are registered with the Information Commissioner's Office under number ZB679491.

This page explains how we use artificial intelligence across those products, in plain language. It sits alongside our privacy notice, which covers personal data generally, and our terms of service. We keep a fuller internal AI policy covering our own governance and staff conduct. If you are a funder, a partner or a customer who needs to see it, email info@hibretone.org and we will send it.

Only Seghu is live today, so everything specific below describes Seghu. ScaleupKickstarter and 1001Trades come under this policy before any AI feature of theirs goes live.

2. Where we use AI, and where we do not

We use AI in a small number of places, and all of them help you write or understand something. You start each one deliberately.

We use AI to:

• Answer your questions. The chat assistant, the conversational structure assessment, and the review interview for organisations that already exist.
• Draft and improve text you will edit. Improving a community interest statement you have written, writing worked examples into empty fields so you can see what good looks like, and drafting a first version of a public benefit statement.
• Summarise. Plain English summaries of your funding packs, and short explanations of why a structure scored the way it did.
• Read a document you upload. If you upload a business plan, we send its text to a model to pull out details such as your organisation name, registered address and the people involved, so you do not have to retype them.

We do not use AI to:

• Choose your legal structure. That is a scoresheet. Each of the eight structure types starts on a fixed number of points and your answers move those points up or down by amounts set in advance.
• Score your funding readiness. The same kind of scoresheet, over five funding types.
• Match you to funders. A three stage calculator run privately on your own answers. It removes funders you structurally cannot approach, scores the rest against fixed weightings, then applies spread rules. Even the sentences explaining why a funder matched are assembled by our own code from your scores, not written by a model.
• Generate your incorporation and governance documents. Those are a mail merge. Your saved answers are poured into fixed templates that a person wrote, on our own server, with no AI involved at the moment of generation.
• Check a company name. That is a direct query to the official Companies House register.

The same inputs always produce the same scores and the same documents. That is deliberate, and it is why we can explain any result to you line by line.

3. AI does not decide anything about you

No score, no recommendation and no document is produced by AI. They are calculated by fixed rules from the answers you give us, and they carry no legal or similarly significant effect. They exist to inform you. Every decision, including whether to incorporate, which structure to use and which funder to approach, is yours, and you are the one who files anything with a registrar.

One honest qualification. If you answer the structure questions by chat rather than by filling in the form, an AI model translates what you write into the answers the scoresheet then scores. It also fills in answers you implied but did not state outright, so that the assessment can run. The arithmetic afterwards is fixed and predictable, but its inputs came through that translation, so read your results and tell us if anything looks wrong. You can always redo the assessment using the form instead.

4. You are told when you are using AI

Nothing is sent to an AI model unless you ask for it. There is no background AI reading your work. Every AI feature is something you start, by pressing a button or typing into a chat.

Most AI features say so where you use them, on the button or in the panel. A few do not label it clearly enough yet and we are fixing that. If you are ever unsure whether something used AI, ask us and we will tell you.

5. Where your text goes, and how long it is kept

When you use an AI feature, your text leaves our servers. Here is exactly where it goes.

All AI traffic passes through one gateway, OpenRouter, which forwards it to a specific model host that we choose in advance for each job. Those hosts are Google Cloud Vertex AI, which runs our main model, and CoreWeave, which runs a smaller model for short pieces of text. We do not let requests wander to whichever provider is cheapest.

The gateway itself does not keep what you send or what comes back. It records only the technical details of the request, such as how many words were processed, which model answered and what it cost, and it does not use anything you send to train models.

Every request we send also carries an instruction that your text must not be collected for training by the model host. On our two normal routes we require zero retention as well, which means the model host does not keep your text after answering.

If both of those routes are unavailable, for example when they are busy, we fall back to a third host, Mistral AI. That route is not zero retention: prompts sent to it are kept for up to 30 days. We are telling you this because it applies to every AI feature, including a business plan upload, which can carry people's names and addresses.

Your side of an AI conversation is stored against your own project so you can return to it, and it is deleted when you delete the project or your account. Deleting your copy does not recall anything already sent to a model host.

6. Keeping users apart, including competitors

Our platforms are used by organisations that may be chasing the same grant, the same contract or the same customers. So this matters.

Nothing you write is used to train or improve any model, for us or for anyone else. We do not fine tune models on customer content, and we have no mechanism to do so. There is no shared search index and no store of everyone's documents that a model can draw on.

No other user's content is ever included in a request we make on your behalf. Each request is built fresh from your own material. Every read of your data is scoped to your account by our application.

Our matching and scoring work the same way for everybody. Nothing in the calculation changes according to what you pay, who referred you, when you signed up, or any relationship you have with us. Your results are computed privately from your own answers, and no other user sees them or is affected by them.

7. What AI here cannot do, and what we do not claim

AI writing tools are confident and sometimes wrong. Ours are no different.

• AI drafted text can be inaccurate, out of date or unsuitable for your organisation, and it can sound authoritative while being wrong. Read it before you use it.
• We are not a law firm, we are not accountants and we are not regulated financial advisers. Nothing the platform produces, with or without AI, is legal, tax or financial advice.
• We do not promise that using Seghu will get you funded, registered or approved.
• We do not use AI to work out things about you that have nothing to do with your answers, and we do not use it to infer health, ethnicity or any other special category of personal data.

8. What human oversight actually means here

You are the human in the loop, and we would rather say that plainly than pretend otherwise.

We do not read AI output before it reaches you. On a chat that would be impossible, since the reply appears as it is written. What we do instead is make sure nothing AI touches is ever final without you:

• AI drafted text lands in a field you can edit, replace or delete.
• Nothing is filed, submitted or signed on your behalf. You download your documents and file them yourself.
• Every document we generate carries its review status on the page, so you can see whether a professional has checked that template.
• Our document templates are reviewed and frozen before release. AI helps us write and maintain them during development. It does not run when you press generate.

If an AI feature gives you something wrong, tell us at info@hibretone.org. We want to know.

9. Your rights, and how to complain

You can ask us what personal data we hold about you, have it corrected or deleted, and receive a copy. You can download your data yourself from Settings, Privacy and Data. Our privacy notice sets out the full list of rights and how to use them.

You can ask us how an AI assisted result was produced and what it was based on, and we will explain it in terms you can check. Because our scoring is rule based, we can usually show you the actual reasoning rather than describe it.

If you are unhappy with anything here, including something an AI feature did, email info@hibretone.org with "data protection complaint" in the subject line. We will acknowledge it within 30 days, investigate and tell you the outcome. You can also complain to the Information Commissioner's Office at ico.org.uk, whether or not you come to us first.

10. How we govern this

A named person inside hibretOne is accountable for AI across the Group. They keep a register of every AI feature we run, what it does, which model host serves it and what data it uses, and they sign off a data protection impact assessment before a feature that carries real risk goes live. Every AI supplier is checked before we use it, and rechecked when it changes.

We are a small organisation and we would rather tell you what we actually do than describe a programme we cannot run. Where something on this page is work in progress, we have said so.

11. Changes to this page

We will update this page as the products change. Material changes will be announced by email or a notice on the platform. This page summarises our internal AI policy, which is reviewed at least once a year.