66%
of CRE professionals use AI weekly¹
AI is arriving in CRE underwriting whether your firm invites it or not. The question that decides who benefits is not whether you adopt it — it is whether a number an AI touched can still be proven when it reaches your investment committee. Fifteen questions locate your firm on that map.
About your models, your AI usage, and what happens when someone asks you to prove a number. Every option is a real practice — including the embarrassing ones.
A score, an archetype, and a map of where your stack sits on the two axes that matter: how much AI is in the workflow, and whether its output can be proven.
Your three biggest gaps, what the top 5% of firms do differently, and a 30-day punch list you can start Monday — whether or not you ever talk to us.
Most AI diagnostics measure adoption — how many tools, how many seats, how much enthusiasm. In a business where a committee votes eight figures on a spreadsheet, adoption is the wrong axis. This instrument measures the one that decides outcomes: whether the numbers your firm produces can still be reconciled, reproduced, and proven once machines help produce them.
Choose the closest to your firm — not the best answer
Two axes decide everything: how much AI is actually in your workflow, and whether a number it touches can be proven afterwards.
The market today: 66% of CRE professionals use AI weekly, 5% trust it for a deal decision¹ — most of the industry is drifting right along the adoption axis without moving up the proof axis. The firms that win the next cycle move diagonally.
The further each point reaches, the stronger the practice. Hover a vertex for the dimension name and grade.
| Dimension | Category | What it measures | Grade |
|---|
Ranked by score. Each one comes with the first concrete move — none of them needs a budget line.
The firms that trust AI with deal work didn't get there by adopting harder. They got there by making the output provable first.
Any figure in an IC pack walks back to the cell and the source document that produced it, in one motion. Where a platform computed it, the calculation opens for inspection — an engine you can see into, not a verdict you take on faith.
Approved versions are sealed read-only, and every change afterwards — an assumption, an override, a renegotiated waterfall — carries who changed it, when, and why. The methodology the IC voted on is signed and frozen; drift surfaces and is re-approved, not discovered in month fourteen by an LP's advisor.
When an investor pushes on the split, the team shows a different pref, promote tier or structure and compares outcomes at the table — not as an overnight Excel exercise. The structure that wins lands in the change record like everything else.
DSCR and the lender's other tests are computed in the model, per period, at the events the loan cares about — post-development, full lease-up, refinance — with the binding constraint named. Not a side calculation rebuilt whenever the lender asks.
"Show me the model the committee approved, and prove nothing has changed" is answered in an afternoon, not a bad week. The approved version re-runs and reconciles a year later, provably unchanged.
AI is let loose on the mechanical layer — extraction, drafting, assembly — which is 60–70% of analyst time². Nothing it touches reaches an IC pack until it passes reconciliation. The boundary is enforced by the system, not by vigilance.
This diagnostic was built by IC FIDES. We put an analyst workbench under a firm's own underwriting: an open calculation engine where every committee number traces to the cell and the source that produced it, sealed versions with a record of who decided what and who changed what, a Waterfall Picker that shows an investor alternative profit splits live in a negotiation, and lender covenants proven at the events the loan cares about. The engine emits to universal Excel in FAST format — working files your analysts keep, carrying the logic with them.
On top sits the layer most models never reach: three statements built from your P&L and Sources & Uses, giving investors a Net Asset Value read on the value building in their ownership.
The next step is a Bench Audit: a working session on your actual model stack — where the mechanical hours go, where a number would fail a proof, where AI can safely take the load. You keep the gap map whether or not we ever work together.