AI block management software
The AI reads. People decide.
Servanda uses AI for the reading that buries a managing agent or a board: certificates, handover packs, sales packs and questions about the building. A person confirms, sends, pays and decides, every time.
What the AI does
Five kinds of reading.
Each one is a job a person did by hand, and each one ends with a person checking the result.
- It reads every certificate you upload.Fire risk assessments, EICRs, lift examinations, gas, legionella, asbestos: the date, the next due date, the assessor, the outcome and every finding with its priority. The reading is shown beside the document and a person confirms it before anything is recorded. If the same report was logged before, it says so.
- It reads the outgoing agent's handover pack.Drop every file in at once. Each is read, given a proposal (file it against a check, put it in the library, hold it for review, or skip it) and the reason in words. For a board it also offers up the flats and their shares, what each flat owed, the budget and the contractors. Filing stays a person's act.
- It reads the last sales pack.Scan a building's last LPE1 or FME1 once and its answers become the building's answer bank, so the next pack starts mostly answered. Compliance dates still come from the live register.
- It researches a new building.Companies House and the web, to fill the blanks a handover left: the company, its directors, the building's details. It only fills empty fields, and runs once per take-on.
- It answers questions from the building's own records.A board asks in plain words and gets an answer drawn from its own register, books and documents; an agent's team asks the same about a sale or a pack. Every answer says it is automatic, and a person is one press away.
What it never does
The lines it does not cross.
Built into the software, not written in a policy. A managing agent answers for what goes out in its name, so nothing goes out that a person did not send.
- Writes to a resident
- Never. An update reaches a leaseholder only when a person presses Send.
- Files a document
- Never on its own. It proposes; a person files, in their name.
- Pays or moves money
- Never. Servanda records who authorised a bill, and a person pays it from the building's own bank.
- Decides
- Never. Votes, awards, budgets and notices are decided and sent by people.
- Poses as a person
- Never as a person, and never as qualified. Automatic answers say they are automatic.
- Answers distress
- Never. A message that reads as distress is held for a person the same day, with Samaritans and 999 shown, and no model sees it.
What is not AI
Rules, not guesses.
Where the law sets a date or a figure, Servanda uses a written rule that gives the same answer every time and is tested before it ships.
- The register: what is in date, due soon, overdue or has no record
- Every clock: Section 20 periods, sales pack working days, complaint deadlines, the 18 month rule
- Contractor reminders, which stop after three unanswered emails
- Service charge demands checked against the law, and the prescribed summary of rights
- Bank statement lines matched to flats by tenant code, flat and name
- Budgets split to the penny by the lease's shares
For RICS regulated firms
The standard, built in.
The RICS professional standard on responsible use of artificial intelligence in surveying practice applies from 9 March 2026.
- The register of AI systems.Every place the software uses a model, what for, since when and when it is next reviewed, drawn live from your own records.
- The dip sample.Ten readings a quarter, chosen fairly, for a person to review against the document. Recording the review is one press, in their name.
- The due diligence answers.The standard's questions about the system, its data, its limits and its risks, answered in writing, including what we measured and what is not published.
- What is not AI.Listed too, so your register does not overstate what a model does in your office.
- What is AI block management software?
- Block management software that uses a language model for the reading a managing agent or a board would otherwise do by hand: certificates, handover packs, sales packs and questions about the building. In Servanda the model reads and proposes, and a person confirms, sends and decides.
- Which AI does Servanda use?
- Anthropic's Claude models, through Anthropic's commercial service. Anthropic is named as a sub-processor in the data processing agreement, and inputs sent to it are not used to train models. The database and files stay in London.
- How accurate is it?
- Accurate enough to save the typing and not accurate enough to trust blind, which is why a person confirms every reading. On 567 real compliance documents in September 2026, seven of sixty proposed labels were wrong when checked by hand, for example a lift repair sheet read as a LOLER examination. Readings below high confidence are held for review.
- Can a RICS regulated firm use it?
- Yes, and it helps you meet the standard. The RICS professional standard on responsible use of AI in surveying practice applies from 9 March 2026. Servanda keeps the register of AI systems it asks for, draws a quarterly dip sample of ten readings for a person to review and records the review, and gives written answers to the standard's due diligence questions, all on one settings page.
- Does the AI chase contractors or send reminders?
- No. Reminders, deadlines and the register are written rules, tested and predictable. The model is used only to read and to answer when a person asks.
- Is it for managing agents or for boards?
- Both. Agents run their whole portfolio on it in their own brand; boards that run their own block or estate use it directly with qualified support included. The AI does the same reading for both.