AI for facilities managers: 6 practical use cases that save hours every week

Facilities managers are absorbing more responsibility every year and getting less back-office support to handle it. AI is the realistic answer to that pressure when applied to specific tasks with the right tool and a clean source of data. Here are six practical use cases that have moved from novelty to standard practice in 2026, with the prompts to copy.

Time saved per task: the headline numbers

1. Weekly utilisation summary, written for you

Tool: Claude with Projects

Upload your desk booking export and check-in data to a Claude Project so the context persists week to week. Then prompt:

You are a workplace analyst. This CSV contains desk booking and check-in data for our office this week. Write a 200-word executive summary for our COO covering: peak day, average utilisation, top three zones by demand, and one recommended action.

A summary that used to take half a day arrives in three minutes. Read it, sense-check the numbers against the data, send it.

2. Maintenance request triage

Tool: Claude or ChatGPT with a system prompt

Set up a system prompt once and reuse it daily:

You are a facilities operations assistant. Classify each incoming maintenance request as: Critical (safety/security), Urgent (operational impact within 24 hours), Routine (schedule within the week), or Low (schedule at next cycle). Explain the classification in one sentence.

Paste the morning’s requests in bulk. Get an instant priority list. The judgement call on genuinely ambiguous tickets still sits with you, but the volume work is done.

3. Lease renewal business case

Tool: Claude Projects with your 90-day occupancy data

Upload the data and prompt:

Using this occupancy data, write a one-page business case for reducing our office footprint by 20% at our next lease renewal. Include: average utilisation rate, peak versus trough comparison, estimated annual cost saving at £[X] per sq ft, and the three supporting data points a CFO would find most compelling.

The document that used to need a consultant takes 15 minutes. Verify every number against your source data before sending.

4. Vendor comparison brief

Tool: ChatGPT or Claude with web search enabled, or Perplexity

Compare [Vendor A] and [Vendor B] for meeting room management software. I need: pricing model, SSO and Azure AD integration, auto-release capability, and G2 review summary. Format as a decision table.

Three hours of RFP research compressed into one. Cross-check pricing and feature claims against each vendor’s own site before sending it up the chain.

5. Contractor induction documents

Tool: Claude

Write a contractor site induction checklist for a [city] CBD office building. Include: health and safety obligations, visitor badge requirements, emergency evacuation procedure, data handling rules, and sign-off confirmation. Tone: clear and non-legalistic.

A solid first draft arrives in 90 seconds. Hand it to your H&S lead for final review.

6. Live operational data inside your AI via MCP

Tool: Claude with an MCP connector to your workplace platform

The frontier use case in 2026. If your workplace management platform exposes an API, a Model Context Protocol connector lets Claude query live occupancy directly. Instead of exporting CSVs every week, you ask:

What is current desk utilisation on Floor 3 right now? Show me which zones are above 80% and which are below 30%.

You get an answer without touching a dashboard. The MCP specification covers how this works. This converts AI from a writing tool into an operational intelligence layer. Setup is a one-day job for IT and the capability is permanent.

How to weigh the risks

Not every risk is equal. Map them by impact and likelihood so you know which to address first.

Risks every FM needs to flag

• Data leakage. Never upload personally identifiable employee data (names + locations + times) to a public AI tool. Use anonymised or aggregated exports only.
• Hallucinated citations. AI invents statistics to fill your business case if you let it. Every number in a document going to leadership must be verified against the source data.
• Over-reliance on AI classifications. AI misses the local context an experienced FM would catch. Use it to triage volume, not to replace the judgement call on ambiguous requests.
• MCP security. Any MCP connection to live operational data is an integration point your IT team needs to evaluate for access controls and data exposure scope. Treat it like any other third-party API integration.