This is the second tutorial in our new community: AI for Real Estate HQ.

Last week we ran the OM read-and-summarise workflow inside a Claude Project.

Learn the last workflow

This week we go a level up. We’re going to build a Skill.

• Drop in the OM.

• Skill activates automatically.

• Out the other side: a buy box screen, your returns at asking, the offer price that gets you to your target IRR, and an IC-ready note.

• 20 minutes.

The judgment is still yours. The data entry isn’t.

If you’re not already doing this, and you/your team does anything to do with acquisitions, this is for you.

Make sure you're subscribed

Up Next

A few things on the whiteboard at HQ:

1. The community opens this month. This is going to be a private community for real estate pro’s who are serious about learning AI. I’ll be doing

   1. Detailed workflow tutorials

   2. Reviews of AI tools

   3. Sharing my personal AI stack

   4. Weekly live sessions

Reply if you want on the early list. Early members join at a discount.

Now, back to the the workflow…

First, what is a Claude Skill?

If you’ve been following AI news, you’ve probably heard about Skills.

Most write-ups make them sound technical.
They’re not.
Here’s the plain version.

A Skill is a folder of instructions that Claude loads on-demand.

You build it once. From then on, every time you start a relevant task, Claude knows the playbook.

Think of it as the difference between a Project and a Skill:

→ A Project is a workspace with always-on custom instructions. Useful, but only inside that project. You have to remember to open it.

→ A Skill lives in the background. Claude pulls it in automatically when it recognises a relevant task. Works inside any project, any chat, anywhere on Claude.

The clearest mental model: a Project is your office. A Skill is something written down that you carry with you wherever you go.

A Skill is just a folder with a file in it called SKILL.md. Inside that file you tell Claude:

1. The name of the skill.

2. A short description of when to use it.

3. The full playbook (what to do, in what order, with what assumptions).

Claude reads the name and description on every chat. If the task matches, it loads the full playbook. If not, the skill sits dormant. You can have 50 skills installed and the unrelated ones cost you nothing.

Once you understand that, the rest is just writing the playbook well.

Today we will build one together:

The underwriting Skill.

Today’s Workflow: From OM to target offer price in 20 minutes

Most acquisitions analysts I know spend hours on a first-pass underwrite.

• Pull the rent comps

• Find the micro-market narrative

• Build or find the proforma

• Sense-check the cap rate

• Write the IC note

• Sense-check it again because they don’t fully trust their own model.

By 6pm they’ve got a number.

By 7pm they’re back-solving the number to satisfy an MD who’s already seen 12 other deals that week.

This Skill runs the same logic in minutes.

It doesn’t replace you. It runs the floor.

What it does:

→ Screens the OM against your buy box. Hard pass or proceed.
→ Pulls the in-place financials (cap, NOI, units, vintage, occupancy) and cites every page.
→ Runs your levered returns on your assumptions.
→ Back-solves the offer price that gets you to your target IRR, equity multiple, yield on cost, and cash-on-cash.
→ Writes an IC summary as a Word doc.

What it doesn’t do:

→ Replace your comp file. (You bring the comps.)
→ Replace the broker call, the site visit, or the read on the sponsor.
→ Replace the full Excel model you build once a deal clears the screen.

What You’ll Need

• Three of your own files. You probably already have all of them.

  1. Your buy box. A document, memo, or notes describing your fund’s investment criteria, return targets, and standard assumptions. PDF, Word, markdown, anything. If you don’t have one written down, spend 20 minutes typing it out before you start. (More on what to include in Step 1.)

  2. Your underwriting model. Your firm’s actual Excel model. The real one. Not a sanitised version.

  3. Your comps file. An Excel of rent and sales comps for your target markets.

Step 1: Pull your materials into a folder

On your computer, create a folder. Call it underwriting-skill. Drop your three files into it:

• buy-box.md (or .pdf, .docx)

• model.xlsx

• comps.xlsx

For comps, in this scenario we’re using comps we’ve obtained from a broker or from an external source.

But, we can also use deep-research AI tools to find comp data online.
But, this can only collect public data.
(We’ll do a seperate workflow on this in the future).

A note on the buy box file. The richer this is, the better the Skill works. If you’re starting from scratch, write down:

→ Strategy and asset class focus
→ Geography and submarket preferences
→ Deal size range
→ Return targets (IRR, equity multiple, yield on cost, cash-on-cash)
→ Vintage and minimum scale criteria
→ Standard underwriting assumptions (LTV, cost of debt, going-in cap range, exit cap, OpEx ratio, capex bands, rent growth)
→ Hard fails (criteria that should kill a deal automatically)

Quick tip: save it as a markdown (.md) file instead of Word. Markdown is plain text, takes fewer tokens for Claude to read, and the Skill runs faster as a result.

Use Apple notes on your Mac, or Google Docs on your Windows and save as an .md file.

Step 2: Enable Skills in Claude

Go to claude.ai. Click your name in the bottom left. Settings > Capabilities.

Turn on Code execution and file creation.

Then go to Chat > Customize > Skills.
Make sure ‘skill-creator’ is turned on.

This is where the Skill will live once it's built.

Step 3: Let Claude build the Skill

Open a new chat. Drag all three files in.

Type this:

“I want to build a Claude Skill that underwrites property OMs. The three files I’ve attached are my buy box, my underwriting model, and my comps file. I want the Skill to read an OM, screen against the buy box, cross-check the comps, populate my model, back-solve an offer price that hits my return targets, and produce an IC memo. Before you build anything, ask me questions so we map it out the right way.”

Subscribe now

Step 4: Answer the design questions

Claude will come back with 5 to 7 design questions. They shape how the Skill behaves on every deal it ever runs. Don’t rush this.

Here’s how I answered each one. Yours will be different. The point is the type of question Claude asks, so you know what to expect.

→ How should the Skill handle missing data in the OM? Flag it. Never invent. Write “Not stated. Assumption: X. Rationale: Y.” Force the user to confirm before proceeding.

→ Comps file handling? Treat as a living source of truth. Read the latest version every run. If the OM’s submarket isn’t in the comps, flag it.

→ Output depth? Populate the exact model, generate a Word doc IC memo, AND a recommended offer price that clears all four return targets.

→ Sensitivity depth? Single scenario with a sensitivity table at 50bps exit cap shift either way. Don’t over-engineer.

→ Hard fails? Below minimum scale, wrong asset class, or outside core markets stops the Skill immediately. Anything else continues with a conditional flag.

→ Page citations? Yes, on every number pulled from the OM. Non-negotiable.

→ Trigger? Manual drop-in for this version. Any OM PDF should activate it.

Spend 15 minutes here. Better answers now means a Skill you trust on every deal.

Step 5: Watch Claude build the Skill

Now agree with Claude to build the Skill.

Claude will produce a draft SKILL.md and write a build plan. It will reference your buy box, your model, and your comps file. It will lay out the pipeline.

Skim the SKILL.md. Read the description and the output spec. If anything contradicts how you actually underwrite, tell it to fix that section.

Two checks worth doing before you ship:

1. Does the description make it clear when the Skill should trigger? (Should mention “OM,” “underwriting,” “investment memo.”)

2. Does the output spec match what you’d actually want on an MD’s desk?

Step 6: Drop the OM in. Type one line.

Open a fresh chat (no project needed). Drag the OM PDF in.

Type: “Underwrite this deal.”

Claude recognises the task, pulls in your Skill, and runs it.

On a 50-page OM, the underwrite lands in about 60 seconds.

What It Doesn’t Replace

The full model.

Once a deal clears this workflow, you still build the proper proforma in Excel. The reason: your IC will ask questions this Skill can’t answer.

Sensitivity tables across rent growth scenarios. Refinancing exit modelling. Different debt structures. Multi-tranche capex schedules. Tax structuring. Waterfall maths.

What this Skill replaces is the first hour of the underwrite. The bit where you screen, set assumptions, get to a directional number, and decide whether to commit a day to the full thing.

Most analysts I know burn 40% of their week on deals that should have been killed in the first hour.

This kills them in 20 minutes.

Honest Take

Where it’s weak:

• It can be hard work prompting to make tweaks. For busy analysts and execs, the last thing they need is to have another channel to type on.

• Token spend - using skills, uploading docs, generating outputs. It all burns tokens, fast.

• Time - a new feature on Claude is updated every week. So, your setup today might not stand next week (that’s why you have me!)

On the positive, the reason Skills matter more than Projects is portability.

A Project locks the workflow inside that project. You have to remember to open the right one. You have to copy the prompt across teams. If your analyst leaves, your prompt walks out the door with them.

A Skill is a piece of intellectual property your firm owns. Every analyst on every desk gets the same playbook. Every IC memo gets framed the same way. The buy box doesn’t drift between analysts because the buy box is in the Skill, not in someone’s head.

This is what real firms do with AI in 2026. They codify the workflow. They stop relying on each person remembering the rules.

If you’re an MD or director and your team is still pasting buy box prompts into ChatGPT, you’re building on sand.

Subscribe now

The Rundown

We round up the AI announcements that matter to real estate, so you don’t have to.

→ Cushman & Wakefield: AI will add 330 million sq ft of US CRE demand over the next decade

Their new white paper AI Impact on Commercial Real Estate: The Next 10 Years models AI’s effect on office, industrial, retail, and multifamily under three adoption scenarios. “AI widens the range of outcomes for real estate more than it shifts the central path,” says Chief Economist Kevin Thorpe. The takeaway for investors: asset selection and submarket dispersion will matter more, not less. Read it before your next allocation meeting. Cushman & Wakefield

→ Crexi launches Crexi AI and Crexi Create

Crexi this week unveiled Crexi AI, embedded AI capabilities built on Crexi’s proprietary transaction data spanning verified listings, buyer engagement signals, ownership records and deal history. They’re demoing live at ICSC Las Vegas on May 18. The broker-side AI arms race is accelerating. Worth watching if you sit on the buy-side and need to know what brokers are now bringing to the table. PR Newswire

→ The “AI immunity trade” is reshaping REIT correlations

Hazelview Investments’ Samuel Sahn notes REIT correlation with the S&P 500 dropped to near zero in early 2026, from a decade-long average of 0.73-0.75. Capital is moving into asset-heavy sectors perceived as AI-resistant. For capital markets folks: the macro tailwind for real estate from AI displacement is real, and it’s repricing risk in your favour. ConstructConnect

→ Proptech funding hit $16.7 billion in 2025, the biggest year on record

CRETI’s year-end analysis shows proptech and adjacent real estate technology investment grew 67.9% year-over-year, surpassing the 2019 pre-pandemic peak. The capital is going to infrastructure-grade companies embedded in core workflows, not point solutions. If you’re evaluating a proptech stake or deciding which platforms to bet on, this is the lens to use. Multifamily Dive

→ JLL: 61% of institutional investors now using AI for market analysis

Up from 22% in 2023. Site screening cycles compressed from 6-8 weeks to 3-5 days. Goldman Sachs estimates 20-35% reduction in CRE due diligence costs at the institutional level. If you’re not on this curve yet, the gap is widening monthly. The Skill we just built is the personal version of what these firms are deploying at scale. Build

Subscribe now

About me

A bit of context, since this is the first one.

I’ve spent the last decade in real estate - first as an operator and developer with the Parklane Group, then building the infrastructure that sits behind the workflows at the top of the market.

Right now I run two things: Buildable, a proptech platform used by some of the largest institutional investors and consultancies in the world for deal analysis and valuation - we’ve been building this for two years and every feature in it was built to solve a problem I watched an analyst waste time on.

And Rex, a real estate fundraising database that, using AI, tracks institutional dry powder, fund strategies, and deal flow across the globe.

I also write The Real Brief, a newsletter that goes out to thousands of real estate professionals every week. It’s where I think in public about what’s actually happening in the market, not what vendors are pitching.

What I’ve seen over the last two years is that AI isn’t coming to real estate. It’s here. It’s being deployed by the firms that move fastest.

And I started AI Real Estate HQ to help firms master AI, through training, implementation, and free resources. This is what you’ll get:

• One workflow per week, written so you can run it.

• The news that actually matters, in eight bullets you can read in two minutes.

• A community where the prompts, the walkthroughs, and the people who actually ship this stuff live.

• Plus, we work with commercial real estate teams to help them implement AI.

It’s for analysts, brokers, valuers, asset managers, and anyone running an MD-level desk who’s tired of being pitched AI tools and wants to know what works. Mostly UK and Dubai, growing into Europe and the US. If that’s you, you’re in the right place.

Share this with the analyst on your desk who shouldn’t be doing the work this Skill replaces.

Have a great weekend. I’ll be sending another one of these next week.

Hope you found value in this.

Let’s do this AI thing together!

Zakee

Subscribe now

Zakee here. I am super excited to be writing the first issue of AI Real Estate HQ (or “HQ” from now).

I’ll keep this short and let the workflow do the talking. There’s a longer note from me at the bottom about why this exists, who it’s for, and who I am. But first, the thing you came here for.

Up Next

A few things on the whiteboard at HQ:

1. The community opens in the coming weeks. This is a community for real estate professionals (analysts, execs, owner/directors). We will be offering:

   1. workflow walkthroughs

   2. the prompt library - our library of prompts that actually work

   3. weekly live sessions

   4. and a tools database with member discounts

Reply to this email if you want to be on the early list.

Note: Early members will be able to join at a DISCOUNT.

Today’s Workflow: OM → 1-Page Memo in 10 Minutes

Most analysts spend 60 to 90 minutes on the first-pass memo for an OM.

Three quarters of that time is extraction, formatting, and chasing missing numbers.

The judgment work (the bit that actually matters) gets the remaining 15 minutes.

This workflow flips it.
Claude does the extraction.
You do the judgment.

10 minutes from PDF to memo, with citations.

What you’ll need

• Claude Pro ($20/month)

• One OM you’ve already evaluated - PDF, locally saved

• 10 minutes

• Coffee (optional)

Step 1: Pick an OM you’ve already worked on

Don’t run this on a live deal.
Run it on one you you already analyses.
That way you can compare.
If Claude’s memo lines up with yours, you trust the workflow.
If it doesn’t, you find out why before relying on it.

Step 2: Open Claude. Create a Project. Don’t use a normal chat.

Projects let you save instructions Claude follows on every memo.
Build it once, use it forever.
Normal chats forget.
Projects don’t.

If you’re not doing this already, this is your signal to start using projects. I have 10+ projects deeply customised to things I work on.

Projects → New Project.

Call it “Investment Memos.”

Step 3. Paste this into the Project’s custom instructions field.

This is the brief. Claude follows it on every OM you drop in. The structure matters more than the words - copy as-is, change the firm name, save.

This is where you put it:

And this is what you copy in:

Save the Project.

Step 4: Drop the OM in. Type one line.

Open a new chat inside the Project.
Drag the PDF in.
Type: “Generate the investment memo.”

That’s it. On a 60-page OM, the memo lands in about 30 seconds.

Step 5. Don’t trust it. Sanity-check three things.

The memo will look right. That’s the dangerous part.

• Open the OM. Verify three numbers - one from the snapshot, one from the table, one from the risk flags.

• Read the What’s Missing section first. This is where Claude is most useful and most analysts cut corners. Tired humans skip “what’s missing.” Claude doesn’t.

• Check the page citations. If any are missing, ask: ”add page citations to every number, citing the OM page where the figure appears.”

Step 6: Copy. Paste. Format. Send.

Move the memo into Word or Docs. Run your firm’s formatting. The memo isn’t the deliverable — it’s the first draft a senior analyst would have written in 45 minutes. You’ve done it in 2.

This is a very BASIC workflow. We will build on this, incuding:

• Building Claude SKILLS so you can do this analysis anytime in any project

• Using public data to find comps, analyse the micro-market, fetch recent deals

• Using Claude Design to create internal or external IM’s from this that you can share with clients/managers

PLUS:

I will share a recorded walkthrough in the HQ community when it opens.

Reply if you want first access.

Honest Take

Once we’ve done all the workflows related to analysing an OM, you will be able to analyse OM’s in seconds.

Deal screening is one of the largest use cases in AI for commercial real estate right now.

For any team reading more than two OMs a week, this gets you 4 to 6 hours back. For any analyst still doing memos at 11pm on a Tuesday, this is the workflow that ends that.

Memo filed, message sent, deal triaged. You’re back to the rest of your inbox before the kettle’s cold.

The Rundown

We round-up AI related announcements every week, so you don’t have to.

→ https://www.anthropic.com/news/finance-agents

Ten ready-to-run agents for pitchbooks, KYC, and month-end. Claude is now native inside Excel, Word, and PowerPoint. We will break down all of this in the HQ Community.

→ Memory for Managed Agents enters public beta

Cross-session learning for production agents. The architecture every long-running AI workflow will eventually need. This is BIG news, and I’ll tell you why in another post.

→ The Automation Center of Excellence framework

REdirect Consulting names the gap most firms are stuck in: AI awareness with no implementation path. Useful read if you’ve been asked to lead AI rollout internally.

→ AI is fighting AI-driven rental fraud

Vero (now $19M raised, Fifth Wall and Sunriver backing) and Findigs leading the response. Identity and income verification at lease application is becoming the new baseline. I own a property and my previous tenants got away with paying 1 month on a 12 month lease. I would have done anything to have a tool to combat this!

→ PwC and ULI: Emerging Trends in Real Estate 2026

The headline most firm leaders will quote this quarter: AI moving from experimentation to adoption. Plus the return of monster rounds for category leaders.

About me

A bit of context, since this is the first one.

I've spent the last decade in real estate - first as an operator and developer with the Parklane Group, then building the infrastructure that sits behind the workflows at the top of the market.

Right now I run two things: Buildable, a proptech platform used by some of the largest institutional investors and consultancies in the world for deal analysis and valuation - we've been building this for two years and every feature in it was built to solve a problem I watched an analyst waste time on.

And Rex, a real estate capital AI agent that tracks institutional dry powder, fund strategies, and deal flow across the market.

I also write The Real Brief, a newsletter that goes out to thousands of real estate professionals every week. It's where I think in public about what's actually happening in the market, not what vendors are pitching.

What I've seen over the last two years is that AI isn't coming to real estate. It's here. It's being deployed by the firms that move fastest.

And I started AI Real Estate HQ to help firms master AI, through training, implementation, and free resources. This is what you’ll get:

• One workflow per week, written so you can run it.

• The news that actually matters, in eight bullets you can read in two minutes.

• A community where the prompts, the walkthroughs, and the people who actually ship this stuff live.

• Plus, we work with commercial real estate teams to help them implement AI.

It’s for analysts, brokers, valuers, asset managers, and anyone running an MD-level desk who’s tired of being pitched AI tools and wants to know what works. Mostly UK and Dubai, growing into Europe and the US. If that’s you, you’re in the right place.

It’s great to have you on-board. Let’s master this AI thing together!

Have a great week.

Zakee