By: KeyCrew Media
General-purpose AI tools can produce a clean summary of a 30-page offering memorandum in seconds, and brokers are increasingly using them to do exactly that. But according to Dan Mosher, CEO and co-founder of DealGround, this approach doesn’t solve real problems and creates a data problem that compounds over time.
A Summary Is Not a Database
Mosher’s core argument is that summarizing a document and storing searchable, updatable data from that document are fundamentally different functions, and that brokers who conflate the two are building on a foundation that will fail them.
The distinction matters because a summary is static. Once a broker uploads an offering memorandum to a general AI tool and receives an output, that output reflects the document as it existed at that moment. Once the leases in that document mature for two additional years, the rent figures, vacancy rates, and lease durations generated two years ago become obsolete, and the user has no recourse other than to rerun the same document.
“If I put it in ChatGPT and it’s a 10-year-old document, it could tell me how much rent they’re paying today,” Mosher says. “But once I do that, that’s a static output. In five years, that output is still telling me it was $100,000 rent when in fact they’re now paying $110,000.”
For prospecting purposes, where brokers evaluate dozens or hundreds of properties against specific financial criteria, this kind of data drift can produce materially misleading results.
How Brokers Actually Search for Properties
The one-document-at-a-time model of general AI tools is misaligned with how commercial brokers actually work. Brokers do not typically sit down with a single offering memorandum and ask an AI to explain it. They work on assignments and client mandates that require them to identify properties that meet specific criteria across a broad market.
Mosher describes a representative workflow: a broker looking for multi-tenant strip centers in a specific region, of a certain square footage, with a particular roster of tenants, a certain range of rental rates and lease terms, and that haven’t sold in the past five years. That kind of search is only possible if the underlying data from dozens of offering memorandums has already been extracted, structured, and stored in a queryable database. Uploading documents to ChatGPT one at a time does not provide that capability.
“There’s no efficient way to find the 30 OMs that solve that search if you haven’t already uploaded them and already extracted the data,” Mosher says.
He adds that brokers working from client assignments need to see the full set of available properties that meet their criteria, not a summary of the single document they happen to have in hand. The difference is between reactive reading and systematic prospecting.
The Guardrails Problem
Even for one-off document summarization, Mosher says general AI tools produce inferior results compared to purpose-built platforms, not because the underlying models are weaker, but because they lack the domain-specific context needed to interpret commercial real estate documents accurately.
His team builds what they call “harnesses”: scaffolding and a structured set of execution parameters that guide an AI model to operate with commercial real estate-specific insights. Without these, a general model may misinterpret terminology, conflate different lease structures, or produce technically accurate but practically misleading outputs for a broker evaluating a deal.
“You have to put the harness around it to get it to accurately convey the information, because if you’re just putting it straight in without the context, without the harness, you’re going to get inferior results,” Mosher says.
In an industry where small errors in rent figures or lease terms can affect underwriting decisions, the gap between a general AI summary and a domain-specific extraction may not be visible to a broker who lacks the context to spot it.
DealGround’s Approach to Document Intelligence
DealGround’s platform addresses both the static data problem and the guardrails problem by extracting structured data from offering memorandums and populating a searchable, filterable, and updatable database. The system tracks rent increases and other changes, so the data a broker retrieves reflects current conditions rather than the state of a document when it was uploaded.
Mosher says DealGround’s data extraction operates at a high level of accuracy, which he considers appropriate for prospecting, where the goal is to identify a set of properties for further evaluation rather than to produce audit-ready financials. If a property’s rent is $205,000 but the search threshold was $200,000, the inclusion of that property in a prospect list does not harm the broker’s workflow.
“More information is better than less,” Mosher says. “Brokers appreciate having more options to choose from, as they can always curtail the list.”
For brokers experimenting with AI tools for commercial real estate research, the risk is not adopting technology too slowly. It is mistaking a useful shortcut, a one-off summary of a single document, for a durable prospecting workflow. A summary answers a question about one property at one moment in time. A structured database answers questions about an entire market as conditions change.
About DealGround: DealGround is an AI-powered intelligence command center that helps commercial real estate professionals prospect smarter. The platform brings together property intelligence, ownership research, and comp data to help brokers generate qualified leads and move faster from prospecting to closed deals. For more information, visit www.dealground.com.
