Disclosure
Pacific Software Ventures (PSV) implements AI for commercial real estate firms and runs the CRE AI Institute, so we have a stake in how you do this. The steps, tables and questions below work with any partner, or with none, and every result links to its full case study.
How do you implement AI in a commercial real estate firm?
AI implementation in commercial real estate is not buying licenses. It's six steps. Find the workflows that eat the most senior hours. Write your firm context once (buy box, underwriting standards, templates) so every tool reads the same rules. Put one workflow into production in your own model or template. Add a review gate where a named person signs every output. Train each role on its own work. Agents and connectors come last. The firms that get results ship one workflow end to end before they touch the second.
Step 1: Which workflows cost your firm the most hours?
Start with an inventory, not a tool. List every recurring piece of work a senior person touches: who does it, how often, what goes in, what comes out. The column that matters is senior hours per run. That's the time AI gives back first, and it's what your partners will measure.
- List the work, not the software. Write it the way your desk says it: underwrite a deal, build a BOV, answer an inbound lead.
- Score each workflow on four questions. How many senior hours does one run take? How often does it run? Are the inputs documents your firm already has? Is there a finished output a person can check? Score high on all four and you've found your first candidate.
- Pick one, and write down what done looks like. Name the output (the filled model, the BOV in your template, the reply in the CRM), the person who signs it and the check they run. If you can't describe done in two sentences, it isn't ready yet. Get a free AI assessment
- Check how ready the team is. Who already uses an AI assistant, and for what? Where do your documents live? The answers tell you whether training or setup goes first. Take the AI readiness quiz
| Workflow | Who does it today | How often | Inputs | Output | Senior hours per run |
|---|---|---|---|---|---|
| Underwriting a new deal | Associate, reviewed by a principal | Every deal that clears the screen | OM, rent roll, T-12, leases | Your model, filled | Your figure |
| Broker opinion of value | Broker and analyst | Every listing pitch | Comparables, property data, notes | BOV in your template | Your figure |
| Inbound lead response | Broker or coordinator | Every inquiry, any hour | Listing inquiries, CRM records | Qualified reply and next step | Your figure |
| Monthly asset review | Asset manager | Monthly, per asset | Property financials, budget, rent roll | Variance commentary | Your figure |
Step 2: Write your firm context once
Firm context is the short stack of documents that says how your firm works. Every AI workflow and every agent reads it before it touches a deal. Skip it, and the same rent roll comes back with a different vacancy assumption from every person who prompts. Write it once, in documents your team can edit, and give one person the job of keeping it current.
- Keep it short and specific. Write rules a new associate could follow. "Underwrite vacancy at the greater of in-place or submarket" is a rule. "Be conservative" isn't.
- Separate firm rules from deal facts. The buy box and standards apply to every deal. Deal facts (this rent roll, this OM) come in with each piece of work and stay out of the firm layer.
- Version it like a policy. When IC changes a standard, update the document that week and date it. An agent reading last quarter's rules will underwrite to last quarter's box. How a Company Brain holds firm context
| Document | What goes in it | What reads it |
|---|---|---|
| Buy box | Markets, property types, deal size range, hold strategy, return targets as you state them, kill criteria | Screening, underwriting, sourcing |
| Underwriting standards | Vacancy and credit loss rules, management fee floor, reserves, capex treatment, how you adjust a seller's T-12 | Rent roll and T-12 work, model filling, memos |
| Templates | Your model, memo, BOV, letter and report formats, with the sections in your order | Every drafting workflow |
| Glossary | Your terms, like how you define in-place NOI or a stabilized asset | Every workflow |
| Where records live | Folders, naming, and the source of truth for rent rolls, comps and contacts | Agents and connectors |
| Approval matrix | Who signs which output, and what needs a second reviewer | Review gates |
Step 3: Put one workflow into production
Production means the workflow runs on live deals, in your own model or template, and a named person signs the output every time. So a pilot on sample files in a sandbox doesn't count. Here's the order we run it in.
- Build it in your own model or template. The AI fills the input cells of the spreadsheet your team already trusts and drafts into the template your partners already read. Your formulas do the math, so the reviewer checks inputs and judgment, not arithmetic.
- Trace every number to its source. Every figure the AI enters carries the document and page (or cell) it came from. If the documents don't support a field, it stays blank and gets flagged. Nobody guesses.
- Test it on deals you already underwrote. Run deals you already underwrote through it, including the ones you passed on, and compare with your team's work. On every difference, decide who was right.
- Run it beside the old process, then cut over. The team does the first live deals both ways. Once the reviewer trusts what comes back, the AI version becomes the standard.
- Measure senior hours, not usage. Count senior time per run, before and after. Logins and prompt counts prove nothing.
What a first workflow in production looks like
Each of these firms started with one workflow, in its own model or template, with a person signing the output. Westlake then took the same approach into AI for asset management decisions. Its published figure is the underwriting result.
- Westlake Realty Group, underwriting in its own model. 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
- Saunders Real Estate, broker opinions of value. Saunders Real Estate cut senior time per broker opinion of value from 4-8 hours to about 30 minutes, drafted in its own template off its own comparables. Read the case study
- Stride Commercial, an inbound lead desk. Stride Commercial's AI inbound desk cut lead response from four hours to under five minutes, with 24/7 coverage and about a day a week of senior time back. Read the case study
Step 4: Put a review gate on every output
At a CRE firm, governance mostly comes down to four questions. Who signs? Can the reviewer trace the numbers? Where do deal documents go? Which data is allowed in at all? Answer them in a short written AI use policy before the first live deal.
- A named person signs every output. The AI drafts, fills and flags. It never approves a deal, sends a BOV or releases an investor letter. A person does.
- Deal documents stay in your firm's business accounts. Run the work in your Microsoft 365 or Google tenant, on a business Claude, Copilot or ChatGPT plan whose terms exclude training on your data, or in your own cloud. Keep confidential deal files out of personal accounts. Full stop. Is AI safe for deal documents
- Keep data vendors' exports out unless you have their written permission. Read your data vendors' terms first. At least one major vendor's current terms bar its data, exports included, from outside AI tools without written permission. Work from your own files and the documents the seller or broker sent.
- Log every run. Record what was read, what was produced, what a person changed and who signed. The log answers where a number came from.
| Gate | What the reviewer checks | Who signs |
|---|---|---|
| Inputs | The right documents went in, the latest versions, nothing from a source your terms bar | The person who runs the workflow |
| Extraction | A sample of figures traced back to the page or cell they came from | Analyst or associate |
| Judgment | Assumptions follow the firm's standards, and every departure is explained | Principal or asset manager |
| Release | Nothing leaves the firm (to a lender, an investor, a client) without a named approver | The person your approval matrix names |
Step 5: Train each role on its own work
Train by job, on the workflow you just put into production, in whichever assistant your firm runs: Claude, Copilot or ChatGPT. The skills carry across assistants. Only the job changes.
- Analysts and associates. Reading a rent roll and T-12 with the AI, checking traced figures, flagging what the documents don't support, and filling the model without retyping. AI rent roll analysis
- Acquisitions. Screening OMs against the buy box, asking the documents the right follow-up questions, and drafting the first look for IC. AI offering memorandum analysis
- Asset managers. Drafting monthly variance commentary against budget, flagging line items past the firm's threshold, and answering lease questions (renewal options, expense caps, co-tenancy) with the page cited. AI for CRE asset management
- Brokers. BOVs in the firm's template, inbound replies that sound like your team, and prospect research that ends in a list a broker approves before anything goes out. AI for CRE brokers
- Partners and leadership. What to ask for, how to read a traced output and what the policy allows. AI training for CRE teams
- Where to train. Train the team in-house on your own workflows, or send people through the CRE AI Institute. It teaches the same work by role, for Claude, Copilot and ChatGPT. The CRE AI Institute
Step 6: Scale to agents and connectors
Once one workflow runs in production and the team trusts it, the second one gets cheaper. The firm context is written, the review gates exist, and people know how to check output.
- Add the next workflow from your inventory. Take the next highest scorer from the Step 1 list. It reads the same firm context, so most of the setup's already done.
- Connect the systems the work lives in. Deal folder or data room first, then the CRM, then the property management system's exports, ideally under each person's own access. A connector inherits whatever permissions it's granted, so check first whether it runs as the user or as a shared service account.
- Turn repeat work into agents. An agent runs a workflow when something happens: a new OM lands, a month closes, an inquiry comes in. It does the reading and drafting, then puts the result in front of the person who signs. AI agents for CRE
- Keep someone on it. Standards, templates and models change. Someone has to update the firm context, re-test workflows and add the next one. At our clients, PSV stays on as the embedded AI team that does this.
Why CRE AI rollouts stall
We hear the same four stories on calls with firms whose first try went nowhere. It's a lot like a lease-up with no leasing plan. Plenty of activity, no signed tenants.
- Licenses before a plan. Seats for everyone, no workflow named, and a usage report nobody wants to read.
- No firm context. Everyone prompts differently, the outputs disagree, and the partners blame the AI for rules nobody wrote down.
- A pilot with no finish line. Nobody defined done, so the test never ends.
- No owner. AI is everyone's side project, which means it's nobody's. The standards go stale.
Build, buy or partner?
- Partner with an implementation team (ours). We run all six steps with you, from the assessment to agents and connectors inside your firm's accounts. Then we stay on as your embedded AI team while your standards and portfolio change. What you own is set out in the engagement terms. How PSV implements AI
- Build it in-house. Cheap to start if someone has the time. Write the firm context and set up the assistant you already license to cite its sources. Your team owns maintenance, testing and every update. Claude for CRE operators
- Buy point software. Each product covers one step, in the vendor's screens and workflow. Your team stitches the steps together and keeps each product's rules in line with your own standards. Best AI tools for CRE
Your CRE AI strategy on one page
A CRE AI strategy doesn't need a deck. It needs nine lines a partner can read in two minutes and hold the team to. The right-hand column shows how an example buy-side firm might fill it in, so don't read it as a client's plan.
| Line | What to write | Illustration |
|---|---|---|
| Goal | The business result, in one sentence | Underwrite every deal that fits the box the week it arrives |
| First workflow | The one workflow from your inventory, and what done looks like | New deal underwriting, filled in our model, traced to source |
| Owner | One named person accountable for the rollout | Head of acquisitions |
| Firm context | Which documents, who keeps them current | Buy box, standards, model and memo templates, owned by the VP |
| Review gate | Who signs each output | Principal signs every underwriting before IC |
| Data rules | Where documents live, what may not go in | Our business accounts only, and no vendor exports without written permission |
| Training | Which roles, on which workflow, in which assistant | Analysts and associates, on underwriting, in our licensed assistant |
| Measure | The before and after you will compare | Senior hours per deal, from our own time records |
| Next two workflows | From the inventory, in order | Monthly asset review, then the investor letter |
Common questions
- What does AI implementation mean for a commercial real estate firm?
- Real workflows running on live deals. AI does the reading and the first drafts, in your own models and templates, and a named person signs every output.
- Where should a CRE firm start with AI?
- With an inventory of recurring work, ranked by senior hours per run. Pick the one workflow whose inputs your firm already has and whose output a person can check. Write down what done looks like, then put that one into production first.
- Should we buy AI seats for everyone before we have a plan?
- No. Seats without a named workflow, firm context and a review gate just sit there. License the people who run the first workflow, prove it on live deals, then widen access one workflow at a time.
- Who should own AI at a commercial real estate firm?
- One named person with authority over the workflows being changed. Usually a partner, or the head of the team whose work goes first. They own the firm context, the review gates and the order of what comes next.
- How do we know the first AI workflow is ready for live deals?
- When it's been run on deals you already underwrote, every difference from your team's work is explained, the reviewer can trace each figure to its page, and the person who signs says they'd put their name on the output.
- Which AI assistant should a CRE firm standardize on?
- Whichever one your firm can run under business terms that keep your data out of training. Usually that's the Claude, Copilot or ChatGPT plan that matches the systems you already use. The method's the same in all three.
- What goes in a CRE AI strategy one-pager?
- Nine lines: goal, first workflow, owner, firm context, review gate, data rules, training, the before-and-after measure, and the next two workflows.
- What happens after the first workflow is live?
- Take the next workflow from the inventory, connect the systems the work lives in, and turn repeat work into agents that kick off when a document arrives or a month closes. At our clients, PSV stays on as the embedded AI team that keeps it current.