Underwriting a commercial building has always been a paperwork problem disguised as a judgment problem. Before anyone can argue about exit cap rates or whether a submarket will absorb another 300 units, somebody has to open a trailing twelve-month operating statement, a rent roll with 240 lines, an offering memorandum written by a broker who is paid to be optimistic, and a tax bill that may or may not reflect what a new owner will actually pay. That data entry stage is where days disappear. It is also, increasingly, where software is taking over — and that shift changes who can credibly analyse a deal, how fast capital moves, and what investors should expect from the sponsors they back.
The old workflow, and why it broke
The traditional acquisitions pipeline runs on a firm’s proprietary Excel model. It encodes years of institutional knowledge: how the firm treats replacement reserves, what it assumes for bad debt, how aggressively it trends insurance in coastal markets. That model is an asset. The problem is the labour required to feed it. An analyst retypes figures from PDFs into a spreadsheet, reconciles categories that never match, builds a comp set by hand, and emails three lenders for indicative terms. By the time a first-pass proforma exists, the seller has already received competing offers.
When deal flow is heavy, firms triage. They screen on a back-of-envelope basis — price per unit, broker-stated cap rate, a gut feel for the street — and only underwrite properly what survives. That triage is where good deals get discarded and bad ones get through, because the screen uses the least reliable numbers in the package. Automation’s real value is not that it produces a model faster; it is that it makes full underwriting cheap enough to apply to everything, including the deals a human would have skipped.
What AI is genuinely good at in a deal file
Extraction and normalisation
Reading a T12 and mapping 60 inconsistent expense line items into a standard chart of accounts is a pattern-recognition task, and language models handle it well. The same applies to rent rolls: identifying lease expiry concentration, loss-to-lease, concessions buried in a notes column, or units marked “down” that the OM counts as occupied. This is unglamorous work with outsized consequences, because a single misread category compounds through every year of a ten-year hold.
Comparables and jurisdictional context
Sales and rent comps are retrieval problems with judgment layered on top. Software can assemble a defensible comp set in minutes and — crucially for US-style underwriting — attach the tax treatment that applies in that specific jurisdiction. Reassessment on sale is one of the most common underwriting errors in multifamily, and it is entirely mechanical to catch.
Scenario volume
Once the model is populated, running 40 debt and equity permutations costs almost nothing. That is where the analytical gain shows up: capital stack structures, waterfall splits, refinance timing, and CapEx phasing can be tested rather than assumed.
Where human judgment still decides the deal
Extraction accuracy is not the same as conviction. Investment committees are still asked to approve eight-figure equity cheques, and no committee signs off on an output it cannot trace. Industry reporting on how technology and uncertainty are reshaping underwriting practice has captured the mood plainly — practitioners are enthusiastic about speed and sceptical about delegating the final call, as this examination of tech uncertainty changing underwriting decisions illustrates. The useful framing is division of labour: machines do extraction, comparison and permutation; humans own assumptions, local knowledge and the decision.
That boundary matters for a second reason. Automated underwriting is not financial, tax or investment advice, and no software approves a loan or guarantees a return. It produces a more complete, more consistent basis for a decision that remains entirely the sponsor’s.
Tooling that works with a firm’s own model
The practical objection most acquisitions teams raise is sensible: they do not want to abandon a spreadsheet refined over a decade in favour of a black-box calculator with someone else’s assumptions baked in. The more workable approach is to automate the inputs and keep the model. Platforms such as NOAL AI take that route, operating as an AI underwriting tool that ingests a firm’s existing Excel model, assumptions and deal documents — T12, OM and rent roll, or in some cases just an address — then auto-populates the model, flags risks, pulls instant sales and rent comps with jurisdiction-specific tax context, runs dedicated expense agents for taxes, insurance and payroll comparisons, builds line-by-line CapEx budgets, models the capital stack and equity waterfall against live loan options, and generates IC-ready reporting and a final proforma. Shared real-time deal workspaces with audit trails address the traceability point, and asset management tracking compares actual performance against the original underwriting, including refinance timing — which is where most sponsors discover whether their assumptions were skill or luck.
Why Indian investors should care
India’s commercial property market has been quietly institutionalising for a decade, through REITs, structured debt and fractional platforms that let individuals hold a slice of a Grade A office floor or warehousing asset. Anyone considering passive income via fractional ownership of commercial real estate is, in effect, outsourcing underwriting to a sponsor. The quality of that sponsor’s analysis is the investment.
So the questions worth asking change. Not “what IRR are you projecting?” but: where did the operating expense assumptions come from, and what comparable set supports them? What happens to the distribution schedule if the refinance slips 18 months? Is actual performance tracked against the original underwriting, and can you show me that comparison for a past asset? Sponsors running automated, auditable workflows can answer those questions in a sitting. Sponsors relying on a single analyst’s unversioned spreadsheet often cannot.
Practical steps for a firm modernising its process
- Document your assumptions before automating anything. Software amplifies whatever logic you hand it, including the sloppy parts.
- Run automation in parallel with manual underwriting on 10 to 15 live deals and compare line by line. Disagreements are where you learn.
- Insist on audit trails. Every number in an IC memo should be traceable to a source document.
- Treat risk flags as prompts for human review, not verdicts.
- Close the loop annually: compare realised NOI, expenses and refinance timing against the original model, and feed the gap back into assumptions.
The competitive shift worth watching
The firms that gain from this are not the ones replacing analysts. They are the ones whose analysts stop retyping PDFs and start arguing about assumptions, submarket dynamics and structure — the work that actually distinguishes one buyer from another. For investors on the other side of the table, the signal to look for is no longer how confident a sponsor sounds about a projection. It is whether they can show you the paper trail behind it, and whether they have bothered to check how their last ten underwritings turned out.
