Disclosure
Pacific Software Ventures (PSV) builds AI systems for commercial real estate firms in Microsoft 365, Claude and ChatGPT, and runs the CRE AI Institute. So yes, we have a stake in how you set Copilot up. The jobs, checks and settings below work in any setup, ours included, and every result links to its full case study.
What is Copilot for commercial real estate?
Copilot for commercial real estate means using licensed Microsoft 365 Copilot inside the apps your firm already runs on. In Outlook, it triages broker emails, pulls the deal terms into a table for your tracker and drafts the reply. In Excel, it audits formulas, ties the rent roll to the T-12 and hunts hardcodes, while your model's formulas do the math and no input goes in without a source. In Teams, it turns calls into recaps and action lists. Copilot Studio builds agents for repeat jobs like deal screening. It works when SharePoint permissions are clean, deal rooms are locked down and a person signs off on every number and every message that leaves the firm.
What does Microsoft 365 Copilot do well in a CRE firm?
Copilot is strongest where your work already lives: the inbox, the workbook and the meeting. It works inside the app, and it reads only what your login can already open in Microsoft 365. That makes it good at reading, checking and first drafts. It also makes loose permissions the main risk. Know which tier you're on. Free Microsoft 365 Copilot Chat works from the web, files you upload and agents billed as you go. The licensed Microsoft 365 Copilot works inside Outlook, Excel and Teams, grounded in your mail, files and meetings through Microsoft Graph. Everything below assumes the license. The judgment stays with whoever's name is on the memo. And the value comes from running the same job the same way, every time.
| App | CRE jobs it does well | Keep a person on |
|---|---|---|
| Outlook | Summarize a broker thread, pull price guidance and the call for offers date, draft a reply that sounds like you. | Sending anything to a broker, lender or LP, and any number quoted in it. |
| Excel | Explain an inherited model, audit formulas, tie the rent roll to the T-12, flag hardcodes. | Every input, every NOI adjustment and the output the memo quotes. |
| Teams | Recap an IC or broker call with decisions and owners. | Deciding what the call actually decided, and anything that goes in the IC record. |
| Word and PowerPoint | First draft of an IC memo section or quarterly letter from approved files. | The recommendation and every figure an investor sees. |
| Copilot Chat and Copilot Studio | Free Copilot Chat works from the web and files you upload. Asking across your mail, files and meetings takes a Microsoft 365 Copilot license. Copilot Studio builds agents for repeat jobs. | The buy box, templates and approved documents the agent reads, and who can use it. |
How should a CRE firm use Copilot in Excel for underwriting models?
Use Copilot in Excel to check the model, not to fill it from memory. Inputs come from a source document, formulas do the math, and a reviewer makes sure the two tie. Think of it as the associate who re-ties your model the night before IC: fast, picky, and never the one setting assumptions. It does its best work on a clean range or an Excel table, in a workbook saved to OneDrive or SharePoint.
- Audit the formulas. Ask it for every formula that breaks the pattern of its row: a cash flow line pointing at the wrong year, a sum that stops one row short. Then you check each flag yourself.
- Tie the rent roll to the T-12. Put both in the workbook and ask for the checks in the table below, each one back as pass or break with the numbers side by side. A break is a question for the broker. Don't smooth it over. AI rent roll analysis
- Hunt hardcodes and plugs. Ask it to highlight every typed number inside a formula row. Inherited models have way more plugs than anyone admits.
- Set up sensitivities, let the model run them. Copilot can build the grid, say exit cap rate against rent growth. The model's formulas fill in every value.
- Never let it invent an input. Ask for a market rent that isn't in your files and you might get a very plausible number with no source behind it. Every input should point to an OM page, a T-12 line or a lease. A blank, flagged cell beats a guess.
| Check | What it compares | What a break usually means |
|---|---|---|
| Occupancy | Physical occupancy on the rent roll against the T-12's economic occupancy (collected rent over GPR) and the model's vacancy and credit loss | A stale rent roll, an expired lease still counted as occupied, or bad debt the model leaves out |
| Gross potential rent | Current GPR on the rent roll against the latest T-12 month annualized and the model's year one | Loss-to-lease (in-place rent below market), concessions burning off, the trailing total lagging the current roll, or vacant units left out |
| Concessions | Concessions on the rent roll against the T-12 concession line | Concessions netted into rent on one side only |
| Other income and expenses | T-12 line items against the model's categories | One-time income annualized, or a cost mapped twice |
| Lease dates | Lease end and option dates against the model's rollover schedule | A lease ending earlier than modeled, or an unmodeled option |
| Hardcodes | Typed numbers inside rows that should be formulas | A plug someone typed to make a total match |
A worked example (illustrative)
An analyst opens the model for a placeholder deal, Example Court, with the broker's rent roll and T-12 on their own tabs, and types:
On the Rent Roll and T12 tabs, run these checks and return a table with check, rent roll value, T-12 value, model value and pass or break: physical against economic occupancy, current gross potential rent against the latest T-12 month annualized, concessions, other income by line. Then list every typed number inside the Cash Flow tab's formula rows. Do not change any cell.
What comes back is a list of breaks, with the cells behind each one. The analyst traces every break to the source, fixes the model or asks the broker, and only then reads the returns. Checks across tabs work best in Copilot's agent mode in Excel. On standard Copilot, run one tab pair at a time.
How does Copilot help with broker and LP correspondence in Outlook and Teams?
Most of an acquisitions inbox is broker email. Most of an IR inbox is LP questions. Copilot in Outlook and Teams handles both well, as long as it drafts and a person sends.
- Broker email intake. Ask Copilot to pull the fields in the table below from each new broker email and hand them back as a row you paste into your deal tracker. Want the row written for you? A Copilot Studio agent or Power Automate flow can add it for a person to review. Anything the email doesn't say stays blank and flagged.
- Replies in your tone. Copilot drafts the pass, the rent roll request or the tour confirmation from the thread, using a short note on how your firm writes. A person reads and sends every message.
- LP answers from approved documents. An LP asks about distributions or occupancy. Copilot finds the answer in the last quarterly report and drafts a reply that quotes it and cites the document. It only quotes what IR has already sent, never a draft model. AI investor reporting
- Teams recaps for IC and broker calls. Turn transcription on and Copilot in Teams writes the recap: decisions, owners, open questions. The deal lead fixes it, names included, before it goes in the file.
| Field | Where it usually is | If it's missing |
|---|---|---|
| Property, address, asset type, size | Subject line, teaser or OM summary | Leave blank and ask in the reply |
| Asking price or guidance | Email body or call for offers notice | Mark as unpriced, not zero |
| Marketed NOI or cap rate | OM financial summary | Leave blank. Your model computes its own |
| Call for offers and tour dates | Email body or calendar invite | Ask the broker |
| Broker and data room status | Signature block and later threads | Flag as NDA pending |
Can you build CRE agents in Copilot Studio?
Yes. A Copilot Studio agent is three things: written instructions, the knowledge it's allowed to read (a SharePoint folder, say) and the tools it can call. Published agents run in Teams and Microsoft 365 Copilot. That takes a Copilot license or Copilot Studio capacity your admin sets up.
- A deal screening agent. Its instructions are your written buy box: markets, asset types, size, hurdles and kill criteria. It reads each teaser and comes back with pursue, pass or needs information, plus the rule behind the call. Keep the buy box in a SharePoint document your team actually edits. AI for CRE acquisitions
- An LP question agent. It drafts answers from one fund's approved documents and cites the source. IR sends.
- A lease question agent. It answers questions like when a renewal option has to be exercised, straight from the executed leases, quoting the clause. AI lease abstraction
- The CREAII connector from the CRE AI Institute. The Institute's CRE Skills (deal screening, underwriting, variance commentary and more) are served through CREAII, a standard MCP connector. It's built to be added in Copilot Studio as a tool, and each member signs in as themselves. The same connector works in Claude and ChatGPT. The CRE AI Institute
- Keep sign-in per person. Don't run tools that read deal files on one shared login. Per-person sign-in keeps each user to what they can already reach, and the log shows who asked what.
How a deal screening agent is set up in Copilot Studio
Instructions:
You screen broker teasers for our acquisitions team against Buy Box.docx.
Pass any deal that meets a kill criterion in the buy box, such as a ground lease or one tenant paying over half the rent.
If the buy box needs a field the teaser does not give, answer needs information and name the field. Never estimate it.
Answer in this format: decision (pursue, pass or needs information), the buy box rule behind it, the teaser line.
Knowledge: One SharePoint folder, Acquisitions/Buy Box, holding the buy box document. No live models and no deal rooms.
Test before publishing: Run it on five past teasers where you already know the outcome. Publish when it gets all five right, or when every miss traces back to a buy box line you then rewrite.
Owner: The head of acquisitions owns the buy box document and reruns the five tests whenever it changes.
Copilot vs ChatGPT vs Claude for commercial real estate
Compare by job, not by brand. Microsoft's Copilots run more than one model family, OpenAI's and Anthropic's included, so the real question is where each assistant sits and what it can reach. Copilot lives inside Outlook, Excel and Teams. ChatGPT and Claude get to your files through connectors.
- If the firm lives in Microsoft 365. Start with Copilot for the inbox, the meeting and the open workbook. It's already where the work is, and it inherits your permissions.
- For long documents and method work. Run the same OM, rent roll and T-12 through each assistant and check every figure against the source. Our benchmark page shows how the models score on CRE reading and math. CRE model benchmarks
- ChatGPT in a CRE firm. Same jobs, in ChatGPT Business and Enterprise. ChatGPT for commercial real estate
- Claude in a CRE firm. Projects, Skills and connectors for a fixed method on every deal. Claude for CRE operators
- Claude and ChatGPT side by side. Our head-to-head on underwriting and OM reading. Claude vs ChatGPT for CRE
| CRE job | Microsoft 365 Copilot | ChatGPT | Claude |
|---|---|---|---|
| Broker email triage and replies | Built into Outlook, reads the thread in place | Through a mail connector your admin approves | Through a Microsoft 365 or mail connector your admin approves |
| Formula checks and tie-outs | Works inside the open workbook in Excel | Analyzes an uploaded copy of the workbook | Excel add-in, or an uploaded copy of the workbook |
| Reading a long OM, PSA or lease set | Reads files in OneDrive and SharePoint | Reads uploaded files and connected drives | Reads uploaded files and connected drives |
| Meeting recaps | Built into Teams with transcription on | Record mode on Business and Enterprise, or a transcript you provide | From a transcript you provide |
| Reusable agents and methods | Copilot Studio agents, published to Teams | Custom GPTs and projects in a workspace | Projects and Skills |
| CRE AI Institute Skills via CREAII | Built to be added through Copilot Studio | As a connector | As a connector |
Is Copilot safe for deal documents? The tenant settings to check first
On a Microsoft 365 work account, Microsoft says prompts, responses and the data Copilot reaches through Microsoft Graph aren't used to train its foundation models. Copilot also follows your existing permissions and sensitivity labels. So the usual risk isn't training. It's that Copilot makes loose permissions easy to find. Have IT check these first.
- Confirm everyone is on a work account. The consumer Copilot app is a different product with different terms. A personal login on deal files sits outside your controls. Is AI safe for deal documents
- Fix oversharing before rollout. Find the sites shared with the whole company and the old deal rooms nobody closed. Copilot will surface a file to anyone who can technically open it. SharePoint access reports and Restricted SharePoint Search help while you clean up.
- One locked site per live deal. Each live deal gets its own site or channel with named members, and you close it when the deal ends. NDA material stays there.
- Label the sensitive files. Apply Microsoft Purview sensitivity labels to NDA material, LP data and fund financials. Copilot respects them and carries the label onto anything it creates.
- Decide on web search. With web grounding on, Copilot can send a search query to Bing, under different terms. Keep deal names out of any prompt that needs the web.
- Set retention and keep the audit log. Set retention for Copilot interactions in Purview to match your records policy, and keep audit logging on.
- Read the NDA and the data licenses. Whether a processor like Microsoft counts under an NDA is a question for your counsel. Keep data vendors' exports out unless you have their written permission. Some licenses bar putting their data into any outside AI tool.
Examples: AI inside a firm's own tenant and model
Two published PSV clients, each result in the case study's own words. One runs inside its own Microsoft 365 tenant. The other runs in its own underwriting model.
- Accomplish Collective, inside its own Microsoft 365 tenant. A real estate sponsor put four AI employees on deal sourcing, underwriting, asset management and investor reporting, inside its own Microsoft 365 tenant. Read the case study
- Westlake Realty Group, AI for asset management decisions. Westlake Realty Group cut per-deal underwriting from 6-8 hours of senior time to a 20-minute review, in its own model with every number traced to source. Read the case study
How to roll Copilot out in a CRE firm
- Pick three jobs, by role. Say broker intake for acquisitions, tie-outs for underwriting and LP drafts for IR, each with a named owner.
- Clean permissions first. Do the oversharing review before a single license goes out.
- Write the method down. The buy box, tracker columns, checks and reply tone go in SharePoint documents Copilot can read. Agents read those approved documents, not live models, and every document has a named owner who keeps it current.
- Train by job, not by feature. Show an analyst how to tie out a model. Skip the tour of buttons. The CRE AI Institute
- Check the work for a month. Track one number per job: tie-out breaks Copilot caught against breaks found in review, broker emails logged in the tracker the same day, and LP drafts sent with no figure changed. When something's off, fix the method, not just the output. AI implementation for CRE
Have it built, train the team, or set it up yourself?
- Have it built inside your Microsoft 365 tenant (ours). We set up the agents, written methods and connectors around your buy box, model and templates, inside your own tenant. Then we stay on as your AI team as the mandate and the portfolio change. How PSV works with CRE firms
- Train the team by job (ours). The CRE AI Institute teaches underwriting, screening and reporting on practice files, in Copilot, ChatGPT or Claude. The CRE AI Institute
- Set it up yourself in Copilot Studio. Doable for one agent, if someone owns its instructions, folders and permissions and keeps them current.
Common questions
- What is Microsoft Copilot for commercial real estate?
- It's Microsoft 365 Copilot used inside the apps a CRE firm runs on: Outlook for broker and LP email, Excel for model checks, Teams for call recaps and Copilot Studio for agents that run repeat jobs. It only reads what each user can already open. Free Copilot Chat covers the web and uploaded files. Working inside those apps and across your mail and files takes a Microsoft 365 Copilot license.
- Can Copilot in Excel underwrite a commercial real estate deal?
- It can check the underwriting. It shouldn't do it. Use it to audit formulas, tie the rent roll to the T-12 and flag hardcodes. Inputs come from source documents, your formulas do the math and your team sets every assumption.
- Is Copilot or ChatGPT better for commercial real estate work?
- Depends on the job. Copilot sits inside Outlook, Excel and Teams with your Microsoft 365 permissions, so it's the natural place to start for email, meetings and the open workbook. For long documents, run the same files through each one and check every figure.
- Does Microsoft train its models on our deal documents in Copilot?
- On a Microsoft 365 work account, Microsoft says no: prompts, responses and data reached through Microsoft Graph aren't used to train its foundation models. The consumer Copilot app is a different story, with different terms.
- What CRE agents can we build in Copilot Studio?
- The common ones screen teasers against your written buy box, answer LP questions from approved fund documents and answer lease questions from executed leases. Each reads only the folder it needs, with per-person sign-in.
- Does the CRE AI Institute work with Microsoft Copilot?
- Yes. The Institute teaches by job in Copilot, ChatGPT and Claude. Its Skills are served through the CREAII connector, a standard MCP connector built to be added in Copilot Studio as a tool, with each member signing in as themselves.
- What should a CRE firm fix before turning on Copilot?
- Permissions. Close old deal rooms, pull company-wide sharing off sensitive folders, lock each live deal's site, label NDA and LP material, and decide whether web search stays on.