Two weeks before a best‑and‑final on a 200‑unit deal, an acquisitions lead told me, “If AI can get us to a first pass in half the time, that is all I care about.” They were getting crushed on bandwidth. Two analysts, three live deals, inbox on fire. Faster sounded like salvation. They got their speed. A generic AI tool cranked out quick deal summaries and “IC‑ready” talking points. The team moved faster. They also missed a tax reset buried in a county document and a lender covenant that behaved differently than they assumed. When IC pressed on the details, the story did not quite hold. They pulled their bid, not because the math killed it, but because confidence did. That is the part most people miss. In CRE, speed helps only when it rides on top of trustworthy, auditable work . If AI gives you fast answers that you cannot stand behind in front of a lender, an IC, or a capital partner, you are not ahead. You are just accelerating into uncertainty. Here is why faster is not always better in CRE AI, and where you actually want to slow down. The Wrong Question: “How Do We Make This 10x Faster?” When partners talk about AI, the first instinct is usually: “Can we cut underwriting time in half?” “Can we generate IC decks in minutes?” “Can we answer brokers immediately?” Those are understandable questions. They are also incomplete. For most core workflows, you do not get paid for speed in isolation. You get paid for: Bidding the right deals at the right price Catching real risk before you sign a PSA or close Showing up to IC with a story, a model, and a risk view that line up If AI helps you move faster in places where you already have tight controls and a shared mental model, it is a win. If AI helps you move faster in places where you are still improvising, you quietly push error into the system. The real question is closer to: “Where can we let AI go fast because we have the right guardrails, and where should we force it to slow down to protect judgment and trust?” Three Ways “Faster” Backfires In CRE AI 1. Fast Drafts That Outrun Your Ability To Check Them The classic trap: use AI to write IC memos, lender narratives, and DD reports “in seconds.” You feed in: The model A few notes from email threads Maybe an OM You get back something that looks polished. Bullet points, sections, even “risks and mitigants.” The problem is not that the draft is bad. The problem is review debt . To trust that document, someone senior now has to: Cross‑check numbers against the model and source docs Make sure the story matches what was actually negotiated Confirm that nothing material got invented to fill a gap If that review step is not explicit and scoped, the team does one of two things: Skims and hopes nothing serious is off, because time is short Rewrites large parts of the draft, which kills the time savings Either way, the “speed” you gained on drafting reappears on the review side. Worse, the firm has now blurred the line on what is human‑vetted and what is AI‑generated. Where to go fast: Use AI for structured, low‑risk drafting where the inputs are already locked: Formatting a standardized IC template from a clean data pack Turning a lease abstract table into a short tenant summary you will edit Generating alternative versions of a paragraph to tighten language Where to slow down Anything that combines numbers, narrative, and risk for external consumption should have: A clear owner A short list of fields that must be checked back to source A checkpoint where someone with signing authority approves it AI can help draft. Human review controls the speed. 2. “Instant” Analysis That Bypasses Source Documents The second failure mode is asking AI for direct answers instead of structured extraction. Examples: “What is the NOI and cap rate on this deal?” “What covenants does the lender require?” “What are the key risks in this environmental report?” If you pass in the actual OM, term sheet, or report and ask the model to lift specific fields and quotes , you have a path back to ground truth. If you paste random snippets or describe the deal in natural language and ask for conclusions, you are effectively saying: “Please interpolate and guess.” AI is good at sounding confident. In CRE that is dangerous. On a live deal, “instant” answers that are one step removed from the PDF or email turn into: Mismatched numbers between the model and the memo Misquoted covenants in internal discussion Vague risk language that does not match the report when someone reads it carefully At that point, speed has just pulled distrust forward. Where to go fast Use AI to: Parse OMs, T‑12s, rent rolls, and term sheets into structured tables Tag and quote specific clauses in leases, PSAs, and loan docs Build quick comparison tables across multiple quotes or deals You still decide what the inputs mean. AI’s job is to get you clean data fast. Where to slow down Anything that smells like: “What should we bid?” “Is this loan structure acceptable?” “Is this risk