04 / AI
AI.
We run it on our own capital first. Then we build it into client shops, train the teams that run it and publish what works.
We run it on our own capital first. Then we build it into client shops, train the teams that run it and publish what works.
Commercial real estate is a documents business wearing a relationships costume. Rent rolls, trailing twelves, leases, loan documents, offering memoranda: the deal lives in paper, and the paper arrives in fifty formats. That is precisely the work machines are now good at, and precisely why most shops that adopt AI badly get burned. They let the machine decide instead of draft.
Our rule is one sentence long. AI drafts, humans decide and every number that reaches an investor gets reconciled against the source once. Intake, screening, abstraction and first drafts move at machine speed. Judgment, structure and the final number stay human. The discipline is not anti technology. It is pro accountability.
We hold our own tools to the same standard. Our underwriting runs on proprietary AI and on models that are verified against an independent replica before release, because a fast wrong answer is still a wrong answer, delivered sooner. The second engine principle from our audit practice applies to machines too: nothing is trusted because it is confident, everything is checked because capital is at stake.
The result is boring on purpose. Fewer heroics, fewer surprises, a pipeline where the top three deals earned their place and the model that priced them can prove its own arithmetic. The model decides the deal. AI just gets it to the decision faster.
The same machinery that underwrites our own deals. Every system below runs under the doctrine: the machine drafts, a person decides and nothing is trusted because it is confident.
Every deal that enters the pipeline is read, standardized and scored against our hurdle rates before an analyst opens the file. The machine ranks. The pipeline meeting decides.
Leases, loan documents and offering memoranda abstracted to verified registers: every term, date and covenant with a source-page reference beside it.
First-pass models drafted in our own format: locked engine, unlocked inputs, audit rows already checking the draft before a human touches it.
Committee-ready drafts assembled from the live model, never from a parallel copy that drifts. Humans rewrite the judgment, not the arithmetic.
We underwrite industrial real estate with proprietary AI: screening, document intake and draft underwriting at machine speed, judgment held by people with capital at risk.
We help clients implement AI where it earns its keep: intake, lease abstraction, screening and reporting, built into your workflow with the checkpoints where they belong.
The AI Brief, every Friday: practical AI for real estate shops, written from implementation work, not from keynotes. Paired with The Underwriting Memo on Tuesdays.
Implementation is a discipline, not a subscription. Five stages, in order, every time, and your team is trained to run the system before we leave.
We audit the workflow before we touch a tool: where the week actually goes, which hours are drafting and which are judgment. The drafting hours are the opportunity. The judgment hours are off limits.
Before anything is built we write the constitution: what the machine may draft, what only a person may decide and where the checkpoints sit. Guardrails are designed first, not patched in after the first mistake.
The system lives inside the software your team already runs, not in a new tab nobody opens. Intake, abstraction, screening, drafting and reporting, wired to your files and your formats.
The new system runs beside your current process against ground truth until the output earns trust. We document the error rate instead of asking you to take our word for it.
Your people run it, with the checkpoints documented and the failure modes named. We teach the operators, not just the owners, and The AI Brief keeps them current after we are gone.
We use an essential cookie to remember your preferences and privacy-first analytics that collect no personal data. By continuing you accept our Cookies Policy.We use one essential cookie and privacy-first analytics. Read the Cookies Policy.