Ask any acquisitions analyst what consumes their week and the answer rarely involves judgement. It involves retyping. A trailing twelve-month operating statement arrives as a scanned PDF with merged cells. The rent roll comes from a property management system nobody on the team uses. The offering memorandum quotes a pro forma that assumes a loss-to-lease recapture no one can trace. Three days later, a model exists — and the bid deadline is tomorrow.
That bottleneck is where artificial intelligence is making its most measurable difference in commercial real estate, and it is a long way from the glossy narrative of algorithms picking winners. The real change is quieter and more structural: documents are becoming machine-readable inputs rather than human transcription jobs, and that single shift alters how many deals a team can credibly assess, how often errors creep into a capital stack, and how tightly underwriting assumptions stay connected to what actually happens after closing.
Why speed matters more than it sounds
Speed in underwriting is often dismissed as a convenience argument. It isn’t. In a competitive multifamily market, the constraint on returns is frequently selection, not execution. A team that can only properly underwrite six deals a month is choosing from six deals a month. A team that can get to a defensible first-pass view on thirty is making a genuinely different decision.
There is a second, less obvious benefit. When a model takes three days to build, nobody wants to rebuild it. Assumptions harden early because revisiting them is expensive. When the mechanical work compresses to hours, sensitivity testing becomes routine rather than ceremonial — and the questions get better. What happens if concessions persist for eight more months? What if the insurance quote comes back forty percent above the trailing number, as it has across much of the Sun Belt? What if the tax assessor reassesses at the purchase price in a jurisdiction that reassesses on transfer?
The diligence errors that actually hurt
Catastrophic underwriting mistakes are rarely exotic. They are usually arithmetic and attention. A few recurring offenders:
- Expense normalisation. Taking a seller’s trailing payroll at face value when the property has been self-managed, or carrying forward an insurance premium bound three renewal cycles ago.
- Tax treatment. Property tax mechanics vary enormously by jurisdiction — caps, millage changes, transfer-triggered reassessment, abatement expiries. A single wrong assumption here can move year-one NOI by double digits.
- Rent roll versus financials. In-place rent from the rent roll and rental income from the T12 should reconcile. When they don’t, the gap usually explains something important about occupancy, bad debt or concessions.
- CapEx granularity. A blended per-unit renovation figure hides the fact that the interiors budget, the roof, and the parking lot have entirely different timing and entirely different effects on lease-up.
- Version drift. Three analysts, four spreadsheet copies, one investment committee memo citing a number that exists in none of them.
Automation addresses these differently than most people expect. It does not improve judgement about whether a submarket will absorb 400 new units. It improves the odds that the number in cell F42 is the number the document actually said, and that everyone on the deal is looking at the same version of it. Marcus & Millichap’s own commentary on the science and the art of technology in commercial real estate makes the same distinction: the technology handles pattern and volume, while relationships, local knowledge and negotiation remain stubbornly human.
Financing scenarios are where models break
Debt has become the swing variable in most multifamily deals, and it is the part of the model most often handled badly. Agency versus bridge, fixed versus floating with a cap, interest-only periods, supplemental proceeds, earn-outs tied to a debt yield test — each of these changes not just the cash flow but the shape of the equity waterfall and the timing of the promote.
Modelling that properly in a spreadsheet means maintaining parallel scenarios that each touch dozens of linked cells. Most teams simplify, which is reasonable until the simplification becomes the decision. The more useful approach is to treat the capital stack as a configurable layer: hold the operating assumptions constant, swap loan terms, and look at what happens to levered IRR, equity multiple, refinance capacity and downside DSCR coverage side by side. The conversation with a lender changes when you can show how their structure performs against two alternatives under the same operating case.
Closing the loop between the model and the asset
The most underused idea in this whole area is continuity. An underwriting model is a hypothesis: these rents, this renovation pace, this expense load, this exit. Once the deal closes, the model usually gets filed and the asset management team starts from a fresh budget template. The hypothesis is never tested.
Carrying underwriting assumptions forward into quarterly performance tracking is how a firm actually gets better at underwriting. If renovated unit premiums have landed consistently below projection across four assets, that is institutional knowledge — not a one-off miss. If payroll keeps coming in above the underwritten figure, the assumption is wrong, not the property manager. This feedback loop is the difference between a team with a model and a team with a view.
The same discipline applies further down the ladder
None of this is exclusive to institutional capital. Private buyers assessing a single residential purchase face a smaller version of the same problem: developer projections presented as facts, service charges that drift, and incentives that quietly change the effective price. Much of the thinking in these practical pointers for buying a 2026 new-launch condo mirrors institutional diligence — verify the comparables yourself, normalise the running costs, and stress-test the exit rather than trusting the brochure.
Where the tooling has landed
A handful of platforms now treat the whole chain — intake, model, debt, committee, ownership — as one workflow rather than five disconnected ones. Noal is a reasonable example of the category: an AI-powered commercial real estate platform that converts deal documents such as T12s, offering memoranda and rent rolls into a complete underwriting model, then keeps going into collaboration, financing scenarios, IC-ready reporting and post-close asset management. Its approach includes underwriting on- or off-market deals from just an address, automatic comps with jurisdiction-specific tax context, dedicated expense agents for taxes, insurance and payroll comparisons, line-by-line CapEx budgeting inside the model, capital stack and equity waterfall modelling with live loan options and lender comparisons, and shared real-time deal workspaces with audit trails. For operators underwriting at volume, the audit trail matters almost as much as the automation — it is what lets a committee interrogate an assumption six weeks later and find out who changed it and why.
Practical steps for a team considering the change
- Run a parallel test. Underwrite three live deals both ways and compare outputs line by line. Differences are information, not failures.
- Fix your assumption library first. Automation applied to inconsistent house standards just produces inconsistency faster.
- Decide what remains human. Submarket conviction, sponsor quality, business plan credibility and negotiation should stay with people who own the outcome.
- Insist on traceability. Every populated figure should point back to a source document and page.
- Instrument the feedback loop. Agree, before closing, which underwritten assumptions you will track quarterly for the first two years.
The judgement premium
As the mechanical portion of underwriting gets cheaper, its value as a differentiator falls. Nobody will win a deal in 2027 because their analyst builds a tidier spreadsheet. The premium moves to the parts that resist automation: knowing which operator can actually execute a heavy renovation, reading a submarket’s supply pipeline with local context, and having the nerve to pass on a priced-to-perfection asset when the model technically clears the hurdle. The firms that benefit most from these tools will be the ones that spend the recovered hours on exactly that.