AI Underwriting Assistants: Can Machine Learning Spot Deals

04/16/2025

Industry Insights Deal Book

Underwriting commercial real estate deals has always combined art and science. Today, artificial intelligence is shifting that balance by turning data analysis into a superhuman skill. An AI “underwriting assistant” can digest rent rolls, expense statements, market comparables, and even satellite images or social media chatter to build a 360-degree view of an investment opportunity. In seconds, algorithms sift through volumes of information that would take a human analyst weeks—spotting subtle patterns and risks that a person might overlook.

How AI Algorithms Underwrite Deals Differently

Traditional underwriting focuses on financials and comps, but AI casts a wider net. Machine learning models pull in diverse data sources simultaneously—from lease ledgers and operating statements to geospatial imagery and online sentiment—to evaluate a property’s prospects. For instance, an AI platform might analyze high-resolution aerial photos to gauge a building’s condition or parking utilization, while also scanning local social media reviews to detect emerging neighborhood trends. By integrating these unconventional inputs with core metrics, AI develops a richer, more granular picture of asset quality and market momentum ( CBRE, “Harnessing non-traditional data like satellite maps, social media, and credit card transactions can enrich real estate forecasts” ). Crucially, these systems learn continuously: as new rent data or market news arrives, the model refines its underwriting assumptions in real time. In effect, AI serves as an always-on research analyst, absorbing fresh information and recalibrating projections far faster than any manual spreadsheet could.

Another key difference is pattern recognition at scale. AI underwriting tools excel at detecting correlations that humans wouldn’t know to look for. They might discover, for example, that properties within a half-mile of a new transit stop tend to see above-average rent growth, or that a specific lease clause correlates with higher tenant default rates. These insights emerge by training on thousands of transactions and data points. While a person might rely on experience and intuition, an AI combs through historical data to quantify what actually drives performance. Modern algorithms can instantly compare a subject property to a vast universe of similar assets, revealing outliers and anomalies. If an apartment building’s expense ratio is unusually high for its submarket, the AI will flag it for review—highlighting potential inefficiencies or maintenance issues that merit attention. In short, machine learning doesn’t get “gut feelings,” but it can surface hidden indicators in the data, bringing to light both red flags and hidden gems that would otherwise remain buried.

The Speed and Scale Advantages of AI-Powered Underwriting

One of the most immediate benefits of AI in deal analysis is pure speed. A task that once kept an analyst busy for days—scrutinizing financial statements, researching comps, double-checking formulas—can now be performed in minutes by an AI model. In practice, this means an algorithm can deliver a preliminary underwrite in the time it takes a human just to gather the data. According to industry experts, what might take three or four weeks of human due diligence on a complex property, an AI platform can handle in a fraction of that time ( GrowthFactor.ai – AI underwriting delivers analysis in minutes vs. weeks ). This rapid turnaround doesn’t just save time; it allows investors and lenders to screen far more opportunities. Instead of evaluating, say, 10 deals per month, a team armed with AI might feasibly scan 50 or 100 deals and quickly zero in on the most promising ones. That kind of scalability can be a game-changer in competitive markets, where being early (or simply not missing a deal) is crucial.

Beyond speed, AI offers unparalleled consistency and objectivity in analysis. Every deal is measured against the same data-driven criteria, free from the fatigue or cognitive biases that can affect human underwriters. The model isn’t swayed by a seller’s slick marketing or by an analyst’s personal attachment to a property class—it simply follows the numbers. This data-centric approach can actually help counteract common blind spots in human underwriting. For example, a human might subconsciously discount opportunities in an unfamiliar submarket or rely on “rules of thumb” that don’t hold true universally. An AI, by contrast, evaluates each deal on its merits and comparable data, which can lead to fairer assessments of properties off the beaten path. In fact, there have been cases where AI models identified overlooked neighborhoods poised for growth that veteran investors initially dismissed. By focusing on evidence (occupancy trends, rent spikes, new permits filed nearby) rather than reputation, the algorithm can flag “sleeper” locations before they become obvious. In short, AI’s pattern recognition and neutral lens can augment decision-makers’ vision—helping them see opportunities and risks with greater clarity ( GrowthFactor.ai – AI models find data-driven patterns investors might miss ).

It’s also important to note how AI improves as it gains experience. A traditional underwriting model in Excel is static—only as up-to-date as the last manual input. But an AI underwriting system continuously learns from each deal it processes. Over time, it can identify which signals most strongly predict success or failure. For instance, if certain combinations of economic indicators and sponsor track record consistently lead to above-average returns, the AI will assign more weight to those factors in future analyses. This adaptive learning means the tool becomes smarter and more accurate with each passing quarter, potentially recognizing shifts in the market (such as a sudden change in interest rates or consumer behavior) faster than a human team could. The result is an underwriting approach that not only moves quicker, but arguably gets more insightful as it ingests more data.

Real-World Applications: From Deal Sourcing to Credit Risk

The rise of AI underwriting isn’t just theoretical—it’s already playing out in real deals. One prominent example is the use of AI to identify mispriced assets. Skyline AI, a pioneering real estate AI firm, demonstrated this capability by mining data on multifamily properties and pinpointing buildings priced below their algorithmically predicted value. In one case, the AI flagged an apartment asset whose asking price was significantly under its true market value based on thousands of data points. Investors using the platform were able to capitalize on that discovery, acquiring the property at a bargain and locking in outsized returns that traditional analysis might have missed. By crunching everything from local rent trajectories to demographic trends, the system highlighted opportunities to “buy low” with a confidence that would be hard to replicate manually ( Commercial Observer – Skyline AI platform revealed undervalued deals via predictive analytics ). Other AI-driven systems are doing something similar on the sell side: predicting which properties are likely to come to market. By detecting subtle signals like a spike in occupancy coupled with below-market rents and an approaching loan maturity, an algorithm can infer that an owner may be preparing to sell, giving buyers an early heads-up to pounce on a potential off-market deal.

AI underwriting has also proven valuable in assessing tenant quality and credit risk—an area that came into sharp focus during the COVID-19 upheaval. Consider the challenge landlords and lenders faced in 2020: how to know which tenants would keep paying and which were likely to default as entire industries went into lockdown. Some forward-thinking firms turned to machine learning for an edge. They fed their models data on tenants’ financial health, industry trends, and real-time economic stress indicators to generate predictive “tenant default probability” scores. For example, a retailer’s risk score might factor in not just its rent payment history, but also the plummeting foot traffic in its sector and the credit card spending trends of its customer base. One European real estate portfolio company deployed an AI solution to do exactly this, after finding that traditional credit ratings no longer captured tenants’ post-pandemic financial realities. The machine learning model analyzed each tenant’s cash flow patterns and industry outlook, successfully distinguishing those under temporary strain from those at high risk of default ( Matics Analytics – Post-Covid case study: AI-generated tenant risk scores improved default prediction ). Armed with these insights, landlords could proactively engage at-risk tenants (offering lease modifications or tapping replacement tenants early), and lenders could adjust loan covenants or reserves in anticipation of trouble. In essence, AI became a radar for emerging credit issues that might not surface in standard underwriting until it was too late.

Even in more routine times, AI-driven risk scoring is helping underwriters paint a fuller picture of deal risk. Modern lending platforms now use AI to synthesize borrower financials, property KPIs, and market data into a single composite risk score or “confidence rating” for each loan application. This can involve running thousands of scenario simulations—What if interest rates jump 200 basis points? What if the anchor tenant leaves?—to see how resilient a deal is under stress. By quantifying uncertainty, the AI effectively highlights which assumptions matter most. For instance, if an office acquisition’s viability hinges on achieving above-market lease-up within 12 months, the model will make that clear in its scoring. Lenders still make the final call, but they do so with a dashboard of AI insights: a deal ranking, key risk drivers identified, and even suggestions of comparables that influenced the model’s view. This level of analysis was previously available only for the largest institutions with armies of analysts; now it’s increasingly accessible through off-the-shelf AI underwriting tools. The net effect is that underwriters—whether at a private equity fund or a community bank—can focus their expertise where it counts, reviewing the riskiest assumptions and negotiating deal terms, rather than grinding through data assembly.

Limitations: Why the “Black Box” Won’t Replace Human Insight

For all its promise, AI is not a crystal ball, and savvy real estate professionals recognize its limitations. First and foremost, an AI’s output is only as good as its input data. Commercial real estate is notorious for patchy and proprietary data; if an algorithm is fed inaccurate rent rolls or out-of-date market comps, its conclusions will be off-base. There’s also the risk of hidden biases in the data. If a model trains mostly on prosperous times or certain geographies, it may overestimate performance and underestimate risk when conditions change. In fact, a known challenge with some AI models is that they can inherit historical biases—for example, underpricing properties in traditionally lower-income areas due to less robust data, or giving too much weight to past growth that might not repeat. Ensuring high-quality, representative data is an ongoing struggle; as one industry report noted, biased datasets can lead to flawed predictions that miss the mark if not corrected with human oversight ( CBRE – Data bias and “black box” transparency are key concerns with AI forecasts ).

Another limitation of AI underwriting tools is their inability to capture qualitative nuances—the “soft” factors that can make or break a deal. An algorithm can analyze a hundred metrics about a location, but it can’t yet drive through the neighborhood to sense its character or up-and-coming vibe. It won’t smell the dampness in a building’s basement or notice that the lot next door is a community eyesore. Seasoned investors often cite these intangibles: the feel of a street, the quality of management, the buzz in a lobby. These are aspects that defy easy quantification but weigh heavily on real-world performance. Today’s AI, for all its data prowess, lacks true context. It doesn’t understand politics, local sentiment, or cultural factors beyond what’s reflected in the numbers. As a result, purely algorithmic underwriting might miss a zoning issue that a local broker knows by heart, or fail to anticipate community opposition to a development that an experienced developer would see coming. This is why most practitioners treat AI’s findings as informative rather than definitive. The model might flag a deal as “low risk” on paper, but a human still needs to sanity-check that against on-the-ground reality.

Transparency is yet another concern. Many advanced AI models, especially those based on deep learning, function as black boxes—it can be hard to explain exactly why the algorithm arrived at a given conclusion. In regulated areas like lending, underwriters need to justify their decisions to credit committees and regulators. A recommendation from an opaque AI can raise uncomfortable questions: Which assumptions drove that output? Has the model been stress-tested for worst-case scenarios? Without clear answers, there’s understandable caution. Some AI providers are addressing this by incorporating explainability features (for instance, highlighting the top factors influencing a risk score), but it remains a hurdle to full trust. And of course, there is the simple reality that even the best algorithm is based on historical data—it has never seen the future. If an unprecedented event hits (a pandemic, a new regulatory regime, a black swan economic shock), the model’s predictions could falter just when we most need judgment. Real estate moves in cycles, and many AI models have not been truly tested across a full boom-bust cycle. Caution dictates that we not over-rely on any single tool, no matter how advanced.

The Power of Human–AI Collaboration (“Centaur” Teams)

Far from rendering human underwriters obsolete, the advent of AI is making their expertise more valuable—when paired together effectively. The best results in deal-making seem to come from a hybrid approach often likened to a “centaur,” where human insight and artificial intelligence work in tandem. In this model, the AI does the heavy lifting of data crunching, unbiased pattern detection, and routine number-checking, while the humans focus on strategy, creativity, and judgment. Think of the AI as a highly skilled junior analyst that never sleeps: it will tirelessly surface facts and figures, perform complex simulations, and even suggest avenues of inquiry. But it still takes a seasoned professional to ask the right “big picture” questions and to interpret the findings in context. As one proptech CEO put it, AI underwriting isn’t about replacing people—it’s about making smart people even more effective. Instead of spending days entering data, an underwriter can spend that time evaluating deal strategy, negotiating terms, and understanding the story behind the numbers.

In practice, human–AI teams can guard against each other’s blind spots. The AI provides diligence and consistency, ensuring no key metric is overlooked and mitigating the risk of human error. Meanwhile, the human can override the model when qualitative red flags appear, or adjust inputs to test different scenarios the model didn’t consider. For example, an algorithm might flag a loan as too risky, but a human underwriter who knows the sponsor’s exceptional track record might decide to make an exception—after all, relationships and reputations still matter in real estate. Conversely, the AI might nudge the team to revisit an assumption (“Are we sure rents will grow 5%? The model’s confidence drops if growth is even 3%.”) thereby injecting a healthy challenge to groupthink. Many organizations now deliberately keep a “human-in-the-loop” for major AI-driven decisions, requiring that a person reviews and signs off on any automated recommendation. This maintains accountability and leverages AI as an assistant, not an autocrat.

Ultimately, the goal is to leverage AI’s strengths—speed, scale, analytical rigor—without losing the wisdom and prudence that come from human experience. High-net-worth investors, brokers, and fund managers are finding that AI can dramatically narrow the funnel of deals to those truly worth a closer look. It’s a powerful bias-checker too: if an analyst has a pet hypothesis about a market, the AI can either validate it with data or contradict it, sparking deeper investigation. The organizations that thrive will be those that integrate these tools thoughtfully, training their teams to interpret AI output and to feed it better data, all while continuing to visit properties, talk to local experts, and apply their intuition. In the end, real estate is a people business, and numbers rarely tell the whole story. But used wisely, an AI underwriting assistant can ensure that when the humans step into the boardroom to make the big decisions, they’re armed with the best insights data can offer.