The Spreadsheet Is Not Dying — It Is Being Wired Into Something Smarter
Ask an acquisitions analyst what they actually do all day and the answer is rarely “analysis.” It is typing. Rent roll figures retyped into a tab. Twelve months of trailing operating statements re-categorised line by line because the seller’s broker used their own chart of accounts. Tax assumptions chased down from a county website. By the time the model is populated, the bid deadline is tomorrow and the thinking happens in the last two hours.
That imbalance is what artificial intelligence is now attacking in commercial and multifamily real estate — not the judgment, but the data entry that crowds judgment out. The shift matters well beyond the US market where it started. As institutional capital flows into Indian commercial assets, warehousing and rental housing, the underwriting discipline that supports those flows is being rebuilt with the same tools.
Where the Hours Actually Go
Most investment committees believe their deal pipeline is constrained by capital or by seller expectations. Look closer at a mid-sized acquisitions team and the constraint is usually throughput. A team of three analysts can properly underwrite perhaps six to ten deals a month at institutional quality. Everything beyond that gets a “quick screen” — a cap rate applied to a broker’s stated NOI, which is another way of saying the broker underwrote the deal for you.
The cost of that compression is invisible until it isn’t. Deals that deserved a real look get passed on. Deals that should have been passed on get pursued because nobody had time to normalise the T12 and notice that repairs and maintenance was running at a third of comparable properties, or that half the “other income” was non-recurring lease break fees. Automation’s first contribution is not insight. It is capacity — the ability to take a serious look at thirty opportunities instead of eight, with the same headcount.
Document extraction is the unglamorous breakthrough
Reading an offering memorandum, a rent roll and a trailing twelve-month statement sounds trivial for software until you remember that none of them follow a standard. Rent rolls arrive as PDFs exported from a dozen different property management systems. Operating statements bury capital items inside operating expenses. Modern extraction models handle this reasonably well because they read context, not fixed field positions — they can tell that “RUBS recovery” belongs in income and that a roof replacement does not belong in R&M, then map both into the firm’s own line-item structure.
From Faster Models to Better Questions
The more interesting development is what happens after extraction. Once a model populates in minutes, the analyst’s day shifts from assembly to interrogation. Why is insurance quoted at twice the submarket average — is it coastal exposure, loss history, or a lazy placeholder? Why does payroll per unit sit so far above comparable assets of similar vintage? Are the sales comps genuinely comparable on unit mix and year built, or merely nearby?
Industry practitioners describe this as a change in the shape of the job rather than its disappearance. In a wide-ranging conversation on artificial intelligence and commercial real estate’s next transformation, the recurring theme is that the technology removes the mechanical layer and raises expectations of what a competent analyst should be able to see. Firms that treat automation purely as a cost-cutting exercise tend to get a cheaper version of the same output. Firms that treat it as a capacity multiplier get more deals examined and sharper reasons for the ones they decline.
Why This Lands Differently in India
Indian commercial real estate has spent a decade professionalising its data. Rent rolls for Grade A office assets are now reasonably standardised, escalation clauses are documented, and institutional owners maintain audited operating histories. That is precisely the groundwork automated underwriting requires. The growth of listed real estate vehicles has accelerated it further — the disclosure discipline that makes REITs attractive to Indian investors also creates the structured, comparable performance data that models can be trained and benchmarked against.
Residential rental and co-living portfolios are further behind, and that is where the practical friction sits. Municipal property tax regimes vary by city and are often revised unpredictably. Insurance pricing is fragmented. Tenant-level collections data may live in WhatsApp threads rather than a system. Automation does not erase those gaps; it exposes them, which is uncomfortable but useful. A model that flags “insurance assumption not supported by comparable data in this jurisdiction” is doing its job even when the answer is inconvenient.
Keeping the Firm’s Own Model at the Center
The biggest objection acquisitions leaders raise to underwriting software is reasonable: our model is our intellectual property. Twenty years of judgment is encoded in how a firm treats loss-to-lease, bad debt, exit cap spread and renovation absorption. Replacing it with a vendor’s generic template is a downgrade, not an upgrade.
This is where the newer generation of platforms positions itself differently — by automating around the existing spreadsheet rather than displacing it. NOAL AI, for example, takes a firm’s own Excel model, assumptions and deal documents and turns them into an automated engine for multifamily real estate underwriting, reading uploaded T12s, offering memoranda and rent rolls (or working from just an address) to populate the model, surface risk flags and generate instant sales and rent comps with jurisdiction-specific tax context. Dedicated expense agents benchmark taxes, insurance and payroll; CapEx is budgeted line by line inside the model; capital stack and equity waterfall modeling runs alongside live loan options so financing structures can be compared rather than guessed. Shared deal workspaces carry audit trails, output flows into IC-ready reporting and a final proforma, and asset management tracking then measures actual performance against the original underwriting — including refinance timing. The point is continuity: the analyst’s assumptions stay the analyst’s, and the labour of assembling them does not.
Practical Steps for a Team Making the Shift
- Clean the model before you automate it. If three partners maintain three different versions of the acquisition template, pick one and document its logic. Automation amplifies whatever structure you feed it.
- Run a parallel period. Underwrite ten live deals both ways. Compare where extracted figures diverge from manual entry and investigate every gap — it will usually reveal an ambiguity in your own definitions, not a software error.
- Insist on audit trails. An investment committee is entitled to ask where a number came from. Any system that cannot trace an assumption back to a source document creates governance risk.
- Keep a human sign-off. Extraction confidence is not certainty. Someone senior should still read the lease abstracts on the top five tenants and the actual tax notice.
- Close the loop post-close. Comparing realised NOI against the underwritten case is the only honest feedback mechanism a firm has. Most teams skip it; the ones that don’t improve their assumptions every cycle.
The Skill That Gains Value
If the mechanical work compresses, the premium shifts to people who can read a market, pressure-test an assumption and argue a position in front of a committee. Formatting a model is becoming a commodity. Explaining why a submarket’s rent growth will disappoint, or why a lender’s structure makes a thin deal workable, is not. The analysts who thrive in the next cycle will be the ones who used the reclaimed hours to look at more deals, more closely — and who can say, with evidence, exactly why they walked away from most of them.