One of the most common questions I hear from Multifamily executives right now is some version of: "We're using AI, but are we using it right?"
It's a fair question, because while the industry has largely moved past debating whether AI belongs in multifamily, most teams are still figuring out how to deploy it strategically. A big part of that is understanding something that the AI vendors don't always make obvious: not every model is built for the same job.
At REBA, we've spent a considerable amount of time thinking through how the three primary Claude models — Haiku, Sonnet, and Opus (TBD on Fable, released this week) — map to the real decisions, workflows and people that define our industry. This isn't a product pitch. It's a framework. Because if you're going to make AI a real part of your operating infrastructure, you need to know not just that it can help, but which version to use, when and where a human still needs to stay in the loop.
First, a Quick Orientation on the Models
Anthropic's Claude lineup follows a clear performance and cost hierarchy. Think of it like the difference between a fast, reliable analyst who can handle high-volume routine work (Haiku), a seasoned associate who combines speed with strong reasoning (Sonnet) and a senior advisor you bring in for your most complex, consequential decisions (Opus).
Claude Haiku is fast and cost-efficient. It excels at structured, high-volume tasks where the inputs are clean and the outputs are predictable. Think data extraction, classification, tagging and summarization at scale.
Claude Sonnet is the balanced workhorse. It handles nuanced language, comparative analysis, multi-step reasoning and longer documents without breaking a sweat. For most production workflows in multifamily, Sonnet is the right default.
Claude Opus is the heavyweight. It's slower and more expensive, but it brings the deepest reasoning to the most complex problems — the ones where getting it wrong has real consequences. It's not something you run on every lease renewal or market comp pull. It's what you reach for when the stakes are highest.
Understanding that distinction — and matching model to moment — is what separates teams that extract real value from AI from those that get impressive demos and disappointing outcomes.
The Lifecycle Lens: Where AI Fits Across the Deal Journey
Deal Screening
This is where volume is highest and patience is lowest. Analysts are processing dozens of potential opportunities a week, and the goal is to quickly separate the ones worth deeper attention from the ones that aren't.
Haiku thrives here. Parsing an offering memorandum for key deal metrics, flagging whether a submarket falls within an investment mandate, extracting rent rolls from PDFs and organizing them into a standard format — these are high-frequency, structured tasks where speed matters more than nuance. An analyst running Haiku across a pipeline of 40 OMs can get a first-pass summary of each in the time it would have taken to manually read two.
That said, the human still owns the go/no-go call. Haiku can tell you what a deal looks like on paper. It cannot tell you whether the GP is trustworthy, whether the submarket story makes sense given what you heard at a recent conference or whether the deal fits a portfolio thesis that lives in the head of your CIO. That judgment stays with your team.
Underwriting
Underwriting is where the nuance begins. Assumptions compound. A 50-basis-point change in exit cap rate can swing a deal from a fund-level investment to a pass. This is not a place for raw speed — it's a place for precision and reasoning.
Sonnet is the right model for most underwriting support tasks: reviewing and stress-testing assumptions in a narrative memo, identifying where a sponsor's projections diverge from submarket trends, synthesizing comparable transaction data across multiple sources, or drafting the analytical commentary that sits behind a model. Sonnet handles multi-page documents and multi-step reasoning well, and it can hold the context of a deal across a longer exchange.
For the most consequential underwriting moments — a large-scale recap, a distressed acquisition, a market entry decision — Opus earns its cost. These are the situations where you want the deepest possible reasoning applied to the most complex set of interacting variables. What happens to this deal's debt service coverage if rates move 75 basis points and occupancy softens 300 basis points simultaneously? Opus can reason through that scenario in a way that is genuinely additive.
But here, the human is non-negotiable at the output stage. AI can sharpen an underwriting model. It cannot sign off on one. That's a decision that carries fiduciary responsibility — and fiduciary responsibility does not delegate to a language model.
Acquisition
By the time you're in acquisition mode, the deal has cleared the first two gates. Now you're in due diligence, negotiation, and close. The work becomes more document-intensive, more legal, and more relationship-driven.
Sonnet handles the document-heavy side of acquisition well: reviewing PSAs for non-standard provisions, summarizing third-party reports, organizing due diligence trackers, drafting correspondence and LOIs. For most acquisition teams, this is where AI earns real time savings — not by replacing the attorney or the acquisitions director, but by eliminating the hours spent on the first pass of everything.
Opus has a role in any acquisition scenario involving significant structural complexity — joint ventures with complex waterfall structures, distressed debt acquisitions or deals requiring synthesis across a large and varied due diligence package. When your legal and deal teams are trying to reconcile 15 documents with conflicting representations, Opus is the model you want helping you find the inconsistencies.
The human still drives negotiation, relationship management and final deal structure. AI is a force multiplier for the analytical and administrative work around the edges of those conversations — not a participant in them.
Operations and Asset Management
This is where the volume of AI use cases in multifamily is highest, and also where the temptation to over-automate is most dangerous.
Haiku is well-suited for the high-frequency, structured tasks that define day-to-day operations: classifying maintenance requests by urgency and category, generating first-draft responses to resident communications based on standard templates, summarizing daily leasing activity reports or flagging anomalies in weekly financial reports by comparing actuals to budget. These are tasks where the inputs are relatively clean, the outputs are predictable and speed creates real value.
Sonnet fits the analytical layer that sits above daily operations: month-end variance commentary, competitive positioning analysis, performance benchmarking across a portfolio or synthesizing survey feedback across a community to surface recurring themes. For asset managers carrying 15 to 20 communities, Sonnet is the model that helps you spend your mental energy on the 20% of problems that matter rather than the 80% that are just noise.
The human remains essential in any situation involving resident conflict, lease enforcement, personnel decisions or anything that touches legal compliance. AI can draft. Humans must review, decide and act on anything that creates liability or requires empathy.
Disposition
Selling an asset requires telling a story. And while the numbers need to be right, the narrative needs to be compelling — to brokers, to the market and ultimately to buyers.
Sonnet does strong work here: drafting the investment narrative sections of a broker package, analyzing buyer feedback and synthesizing themes across offers, comparing exit timing scenarios across rate and cap rate environments or building the performance story around NOI growth and operational improvements made during the hold period.
Where the disposition involves significant tax structuring, 1031 optimization, or complex partnership unwinds, Opus brings the reasoning depth to match the stakes. The same is true if the sale is part of a broader portfolio strategy conversation — synthesizing how the disposition affects remaining portfolio metrics, rebalancing toward target allocations or framing the exit in the context of fund-level performance.
The broker relationship, buyer negotiations, and pricing strategy remain firmly in human hands. AI can help you prepare for those conversations. It cannot conduct them.
Capital Raising
Capital raising is a relationship business. It is also an information-intensive one. The distance between those two realities is exactly where AI adds value.
Sonnet can materially accelerate the process of preparing LP updates, synthesizing portfolio-level performance data into a coherent narrative, drafting responses to common LP due diligence questionnaires and organizing the data room for a new raise. For IR teams that are stretched thin across a growing LP base, this is high-value work that often gets done at the last minute.
Opus is the right model for preparing for a sophisticated institutional LP's investment committee presentation — synthesizing fund-level data, anticipating questions based on the LP's known investment criteria, or helping structure the strategic argument for a new fund strategy in a competitive fundraising environment.
The LP relationship, trust-building, and live Q&A are irreducibly human. No model should be in that room — figuratively or literally.
The Persona Lens: Who You Are Changes How You Use AI
The right model isn't just about where you are in the deal lifecycle. It's about who's sitting in the chair.
Analysts are the highest-volume users of AI, and they tend to get the most immediate productivity gains. Haiku handles their extraction, classification and first-draft work. Sonnet handles the analytical synthesis that makes their output look senior. The risk for analysts is over-reliance: using AI output as a final answer rather than a starting point. The discipline of checking, contextualizing and editing AI output is a skill that will separate good analysts from great ones.
Managers — asset managers, revenue managers, property managers — benefit most from Sonnet applied to their core analytical work. They're not running raw data tasks. They're interpreting results, making operating decisions and communicating up and down the organization. Sonnet helps them synthesize faster, communicate more clearly and spend less time in spreadsheets. But they own the decisions.
VPs are where strategic judgment enters the equation, and that's where Opus starts earning its place — not for every task, but for the high-stakes analyses that inform the recommendations they're making to leadership. A VP of Acquisitions preparing a deal memo for the investment committee should not be handing that document to a language model without deep review. But using Opus to stress-test the analytical narrative before they do is an entirely reasonable use of the tool.
Executives — CIOs, CFOs, COOs, CEOs — should be the least frequent direct users of AI and the most important beneficiaries of it. AI's job is to make the information that reaches the executive level cleaner, more synthesized and more actionable. The executive's job is to apply judgment that no model has — institutional knowledge, relationship context, board dynamics, market instinct. That judgment is not replaceable, and anyone who tells you otherwise is selling something.
The Operating Model Lens: Structure Changes Strategy
How your organization is structured fundamentally shapes which AI use cases deliver the most value — and where human oversight is most critical.
REITs operate at scale, with reporting obligations that are both frequent and unforgiving. Haiku and Sonnet together do strong work on the routine reporting layer: summarizing property-level performance, flagging variances, preparing board materials, and generating analyst-ready data packages. The regulatory dimension — SEC filings, disclosure language, investor communications — requires human review of every AI-assisted output before it leaves the building. Full stop.
Vertically integrated owners have the broadest set of touchpoints across the lifecycle, which means they also have the broadest opportunity to deploy AI. The challenge is consistency: when you own the deal from acquisition through disposition, the handoffs between teams — acquisitions to asset management, asset management to disposition — are where information gets lost. AI can play an important role in maintaining continuity of context across those handoffs, provided the underlying data foundation is clean.
Capital allocators — GPs, fund managers — are most acutely exposed to the information asymmetry problem in capital raising and portfolio reporting. Sonnet and Opus together can help close that gap, making LP communications more rigorous and fund-level analytics more accessible. But allocators also face the highest reputational risk from an AI error that makes it into a client communication. Human review of every external-facing output is not optional.
Third-party property managers are managing the tension between scale and customization — operating hundreds of communities across multiple owner clients, each with different reporting requirements, performance standards, and communication preferences. Haiku's strength in structured, high-volume tasks makes it particularly valuable here. The risk is letting the efficiency of AI output reduce the quality of owner-client relationships. The human touch in client management doesn't get automated. It gets protected.
The Through-Line
Across every lifecycle stage, every persona, and every operating model, a few principles hold:
AI does its best work when the inputs are clean. The REBA data foundation isn't just about dashboards and reports — it's about ensuring that when you ask a language model to analyze your portfolio, you're not feeding it three versions of occupancy from four different systems. Garbage in, garbage out is not a new concept. It's just more expensive now.
The model tier matters less than the discipline around it. Operators who get lasting value from AI are the ones who build clear protocols: what gets human review, what can flow through automatically, who owns the output when something goes wrong. That's not a technology conversation. It's an operating model conversation.
And finally: AI is not a replacement for institutional knowledge, relationship capital or fiduciary judgment. It is a multiplier for the people who have those things and know how to use them.
The teams that figure that out — really figure it out, not just in theory but in practice — are going to have a meaningful advantage in the years ahead.
.png?width=70&name=Untitled%20design%20(20).png)
SHARE