Trust CentreResponsible AI statement
How WiseBlock uses AI, and where it stops.
WiseBlock uses AI to read and to draft: a photo of a fault, a bank statement, an email, a question from your team. This statement lists every feature that uses it, what it sees, who provides it, and who makes the decision afterwards.
- DecidesPeople. AI drafts, sorts and suggests
- SeesOnly what the feature needs
- TrainingNot used to train Anthropic's or OpenAI's models
- ControlEach agent switched on per organisation
01Where AI is used
Every feature that uses AI.
If a feature isn't on this list, it doesn't send information to an AI model.
| Feature | What it does | Provider | What it sees | Who decides |
|---|---|---|---|---|
| Platform assistant | Answers your team's questions about the portfolio from live data. | Anthropic (Claude) | Buildings, units, balances, checks and tasks. Residents appear as a unit and building, not a name, unless the question names them. No email addresses, phone numbers or payment references. | Your team reads the answer. It can't change anything. |
| Email Triage agent | Sorts incoming building email by type and urgency, and flags what needs a person. | Anthropic (Claude) | The emails it sorts. | Your team acts on them. |
| Bank Specialist agent | Reads an uploaded bank statement into lines. | Anthropic (Claude) | The statement. | Your team matches and reconciles. |
| Fire risk assessment drafting | Drafts assessment text from what's recorded about a building. | Anthropic (Claude) | The building's details and recorded findings. | The assessor edits and signs off. |
| Maintenance Surveyor agent | Drafts a repair's title, description, category, urgency and trades from its photos. | OpenAI | The photos and what was written with them. | Your team edits the draft; nothing is saved unchecked. |
| Invoice reading | Reads a scanned contractor invoice so its lines can be checked. | Mistral AI | The invoice. | Checks run, then a person approves. |
| AI receptionist | Answers residents' questions in the portal by chat (Jordan) and voice (Emma), and can text them a payment link. | VAPI | The resident's conversation, and what's needed to send the link. | The resident. Anything that needs a decision goes to your team. |
| Report layout | Lays out generated documents, such as a fire risk report. | Gamma | The report's content. | Your team reviews the document. |
Not AI: the Contractor Matcher scores firms from your own quotes and work orders, with its reasons shown; and the Money Collector's reminder calls read out a set message. Residents can turn reminder calls off in their portal preferences.
02Information
What's sent, and what isn't.
A feature sends what it was asked to work on, at the moment it's used, and nothing else.
-
Minimised first
The platform assistant is told about residents as "Unit 44 · the building", linked to their page, not by name or contact details. Names are only included when your question names that person.
-
Only on demand
Photos, statements, invoices and emails are sent when the feature that reads them runs, not in bulk or in the background.
-
Logged by category, not content
Each AI request is logged against its agent with what it cost. The assistant's log records what kind of information was sent, not the question's words, which often name people.
-
Retention and training
Anthropic's and OpenAI's business terms don't use what we send through their APIs to train their models, and allow them to keep it for a limited period (up to 30 days) to detect misuse before deleting it. Other providers process information under their business terms with us.
03Human review
No significant decision about a resident is made by AI.
What a resident owes comes from the ledger, not from a model. Which contractor gets a job, whether an invoice is paid and whether a repair is approved are decided by people, and spending over a building's limit needs a named approver.
AI output arrives as a draft or a suggestion in the same screens your team already uses, and nothing it drafts is saved until someone has checked it.
Residents can always reach a person: by messaging their managing agent through the portal or contacting the office directly. The receptionist doesn't stand in the way.
04Your controls
Switched on, set and switched off by you.
On per organisation
Each AI agent is activated in Settings › AI agents by an administrator. Until then it does nothing.
Set how it works
Where an agent offers it, choose what it drafts, how much it reads, and standing instructions it follows.
See what it did
Every request is logged against the agent, with what it used and what it cost.
Off again
Deactivate an agent at any time; the assistant only appears while its agent is on.
Questions
AI at WiseBlock, answered.
Is our data used to train AI models?
Not by Anthropic or OpenAI, whose business terms don't use API data for training. Other providers process information under their business terms with us; ask us for the detail on any of them.
Can we use WiseBlock without the AI features?
Yes. The AI agents are switched on per organisation, and everything they help with can be done by hand.
Does the AI ever contact residents by itself?
The AI receptionist answers residents who contact it in the portal. It doesn't start conversations. Automated reminder calls read a set message, and residents can turn them off.
Where is AI processing done?
The AI providers listed above mostly process in the United States, Mistral AI in France. See Hosting and data locations in the Trust Centre.
Who do we ask about AI and our data?
Write to info@wiseblock.co.uk with "Privacy" in the subject.
Related
Next step
See it run on your blocks.
A 30-minute demo on a portfolio like yours: fire safety, repairs, residents and service charges, with the agents switched on.