AI
The 21RISK AI Assistant
What the AI assistant is What the AI assistant is
The 21RISK AI assistant is a chat-based helper built into the platform. It can answer questions about your data, help you find and complete work, and automate repetitive tasks — all inside the same access model that already governs the rest of 21RISK.
This article is for administrators and end users who want to understand what the assistant does and how access and cost controls work. The platform-level security model of our AI implementation (how requests are processed, what data leaves the platform, and the trust boundaries involved) is documented separately for security reviewers in our compliance documentation.
Important: The assistant always acts with your permissions. It can only see and change what you could see and change yourself. Granting someone AI access never widens their underlying permissions.
What it can help with What it can help with
- Answering questions about sites, boards, audits, action items and insurance data you already have access to.
- Finding the right place to do something and walking you through it.
- Running supported actions on your behalf (for example creating or updating records) using your own permissions.
- Searching the 21RISK documentation and summarising it for you.
- For tenant owners: answering questions about the PDF, Word and Excel files uploaded to 21RISK, such as insurance policies and engineering reports, with the file and page it answers from, and finding uploaded photos by what they show.
Asking about your documents Asking about your documents
21RISK reads the text of every PDF, Word and Excel file uploaded to it: insurance policies, engineering reports, attachments and other documents. Tenant owners can ask the assistant what those files say, for example Which of our policies exclude flood? or Summarise the exclusions in the 2026 master policy. The assistant searches the files, reads the pages it needs, and names the file and page it answers from, with a link that opens the page so you can check it.
- Owners only, for now. Who may see a file depends on what it is attached to, such as a site, an item or the Insurance board's settings. Until those rules apply to the search, only tenant owners, who can see every file, can ask about them.
- Words and meaning. The search finds pages by their words, so exact terms such as amounts and clause numbers find their page. In organizations that have turned on AI, it also finds pages by meaning, in any language: a question about the flood deductible finds a Danish page about selvrisiko ved oversvømmelse. For that, the text of each uploaded page is sent once to Voyage AI, through our AI provider gateway, to be turned into an embedding (see Sub-processors).
- Where a file is. Each file the assistant finds or reads comes with what holds it: the risk improvement, finding, claim or Compliance 2.0 question it is attached to, or the insurance policy, engineering report or insurance value it belongs to, with the site and board. You can ask about the files of one kind of place or one site, for example What do the attachments of our risk improvements in Esbjerg say about sprinklers?
- Photos. In organizations that have turned on AI, the assistant also finds uploaded photos and other images by what they show and the text in them, for example Find photos of corroded sprinkler valves or Which photo shows the type plate of the boiler? Each image is read once by a model, which describes it in a sentence or two and copies the text visible in it, such as signs and type plates; the assistant links to the image so you can look at it yourself. The image is sent to Anthropic through our AI provider gateway, with zero data retention (see Sub-processors).
- Scans. Scanned pages, and other pages whose text couldn't be read in full, may not be found yet. The search tells the assistant when the files it searched include any.
Who can use it Who can use it
AI chat is controlled at two levels:
| Level | Control | Where |
|---|---|---|
| Organization | AI must be enabled for your organization | Settings → AI → Settings |
| User | The user must be granted AI chat access | Settings → AI → Access |
A user only sees the Chat entry when both are true: AI is enabled for the organization, and they have been granted access. Access is granted to individual users or to user groups (everyone in the group gets access).
To grant access, an IT admin navigates to Settings → AI → Access, picks the User groups or Individual users tab, selects who to add, and clicks Add.
Monthly spend limits Monthly spend limits
AI usage has a cost. Everything AI does in 21RISK counts toward your organization's AI spend for the calendar month: chat, agents, reading engineering reports into suggested risk improvements, and going to pages and filtering tables with AI. Enterprise organizations get €50 of AI free each month. Spend above that, up to the organization limit, is billed. A month's charge is fixed when the month ends, so changing the limit afterwards does not change it. It is added as one line to the first licence invoice drafted after that month ends: with monthly billing, usually the one after next.
21RISK provides spend controls in EUR:
- Organization limit — a monthly cap on total AI spend for the whole organization, free allowance included (Settings → AI → Settings). New organizations start at €50. For enterprise organizations that equals the free allowance, so nothing is billed until an IT admin raises the limit; other organizations pay for all their AI spend. When reached, new AI requests are blocked until the next calendar month or until the limit is raised.
- Per-grant limit (optional) — when you grant access to a user or group, you can set an optional monthly cap on that person's own chat spend, and change it later at any time. When a person reaches it, they can't send new chat messages until the next calendar month or until the cap is raised.
The Usage tab shows this month's spend, how much of the free allowance is used, and how much will be billed.
When a person is covered by several grants at once, their effective monthly cap is resolved like this:
- An individual grant with a cap wins for that person.
- An individual grant without a cap falls back to the group level.
- The group level is the most generous (highest) cap among the user's granted groups — and unlimited if any of those groups has no cap.
- If no cap applies anywhere, the user is only bounded by the organization limit.
Tip: Use group grants to manage access at scale, and add an individual grant only when one person needs a different limit than their groups provide.
Choosing models Choosing models
IT admins can choose which AI models are available to the organization and set a default model (Settings → AI → Settings). Prices for each model are shown in EUR. Users can switch models per conversation among the allowed models.
Privacy and your data Privacy and your data
Conversations are stored in 21RISK so you can return to them. Administrators can review usage on the Usage tab. The details of how requests are routed, what is sent to model providers, and how the data is handled are covered in our compliance documentation for security reviewers.