How should a real estate development team learn to use AI?
It depends on where your team’s hours go. If acquisitions and development staff spend most of the week reading and writing, such as offering memorandums, zoning codes, market reports, consultant studies and investment committee memos, short hands-on training on those tasks pays off fast. If the pain is in the model itself, like rebuilding pro formas from broker files or pulling rent comps from several sources, training helps less and you likely need a connected workflow. Either way, train on your own deals, in approved tools, with a rule that every number is checked against its source.
Who this is for
Heads of development, acquisitions leads, asset managers and COOs at development firms, owner-operators and real estate investment groups who want their team using AI on live work without risking a bad number in an investment committee memo.
The short version
- Train on real documents from recent deals, not generic examples.
- Teach the core habit first: AI drafts and summarizes, people check every figure and every legal or zoning claim against the source.
- Start with reading-heavy work (offering memos, zoning codes, studies), then move to drafting (memos, investor updates), then to repeatable workflows.
- Give each role its own exercises. Acquisitions, development management and asset management use AI differently.
- Measure on named tasks at 30, 60 and 90 days.
Why development teams need their own training
General AI training is built around emails and meeting notes. Development work is different in three ways.
First, the documents are long and dense. A zoning code, a geotechnical report, a purchase agreement and a market study can each run dozens of pages, and the useful facts are buried. AI tools are good at finding and summarizing those facts, which is where most early time savings come from.
Second, the cost of a wrong number is high. An AI tool can misread a table in a broker package, mix up gross and net figures or state a setback requirement that is not in the code. Training has to make checking automatic, not optional.
Third, information is sensitive. Land under contract, partner terms and lender conversations should not go into personal AI accounts. Training needs to cover which tools are approved and what data may go in. If you have no policy yet, start with how to write an AI acceptable use policy.
What to cover, in order
1. Reading and extracting. Have people upload a real offering memo or market study into an approved tool and ask for a structured summary: unit mix, in-place rents, assumptions the seller is making, and anything that looks unusual. Then have them check each item against the document. This one exercise teaches both the value and the limits.
2. Codes and entitlement documents. Practice asking specific questions of a zoning code or a planning staff report, such as height limits, parking ratios or required approvals, and always asking the tool to quote the section it relied on. Treat the answer as a pointer to the right page, not as a legal opinion. Confirm anything that matters with your land use counsel or the jurisdiction.
3. Drafting. Use AI to produce first drafts of investment committee memos, partner updates and lender narratives from your own notes and figures. The person who owns the deal edits and signs off.
4. Research support. Use AI to organize public market information, compare submarkets or prepare questions for a broker call. Teach people to ask for sources and to treat unsourced claims as unverified.
5. Review routines. Agree on a simple checklist: figures traced to source, legal and zoning claims confirmed, no confidential data in unapproved tools, named reviewer on every external document.
A simple training format that works
- Session 1 (90 minutes): the basics in your approved tool, data rules and the reading exercise on a real document.
- Session 2 (90 minutes): role-specific work. Acquisitions works on screening and memos, development management on codes, consultant reports and schedules, asset management on leasing reports and budgets.
- Weeks 2 to 6: short weekly office hours where people bring live work, plus a shared library of prompts that worked.
- Day 90: review the named tasks you measured at the start and decide what to standardize.
Timing will shift with team size. Our page on how long it takes to train a team on AI covers the trade-offs.
Where training is not enough
Some development work is a data problem, not a skills problem. If analysts spend hours rebuilding the same model inputs from broker files or copying rent comps between spreadsheets, better prompting will not remove that work. A connected workflow or custom agent, built with checks a person reviews, does more. See whether AI can help with underwriting and custom AI vs off-the-shelf tools.
Common mistakes
- Training on toy examples. If exercises are not your own deals, little changes the next week.
- Trusting summaries without checking. A confident summary of a code or a broker package can still be wrong.
- Leaving out senior people. If the head of development never reviews AI-assisted work, the team quietly stops using it. See getting partners and senior leaders to use AI.
- No approved tool before training. People default to personal accounts with deal data in them.
- One session and done. Without follow-up, habits fade within weeks. See why AI adoption stalls.
When you do not need us
If your team already has approved tools, clear data rules and someone internally who uses AI well on deal work and is willing to teach, an internal champion program may be enough. See internal AI champions vs an outside trainer.
How Ruby helps
- AI training: hands-on sessions for acquisitions, development and asset management teams, using your own documents, with follow-up office hours.
- AI strategy and consulting: choosing starting workflows, setting data rules and measuring results.
- AI engineering: building connected workflows or agents where training alone will not remove the manual work.
If you want a training plan built around your team’s actual deals, book a call with Ruby.
