In most professional financial analysis, the document you receive is not the document you underwrite. The numbers on the page are reported numbers — they reflect how the current owner operated the asset, not how a market participant would operate it. The work of analysis is closing the gap between the two.

This normalization process is rarely visible in the output. A final underwriting model shows adjusted numbers. What it doesn’t show is the reasoning behind each adjustment: why the management fee was changed, what was added for capex, where the payroll line was modified to reflect market rates. That reasoning lives in the analyst’s head and in a set of conventions that experienced professionals know and newer ones have to learn.

Owner-operated properties are a common example. When an owner manages their own building, there’s often no management fee on the income statement — they don’t pay themselves a formal fee, so it doesn’t appear as an expense. The property looks more profitable than it would under professional management. An experienced analyst adds a management fee at market rates before underwriting. A model that doesn’t perform this adjustment will systematically overstate NOI on owner-operated assets, which means it will systematically overpay.

The same pattern appears across multiple expense categories. Capital expenditure reserves rarely appear in operating statements — they’re not a cash expense in any given year, so they don’t show up in a trailing 12 months of actuals. But a buyer who doesn’t reserve for future capital needs will face those costs eventually. The convention is to add $200 to $400 per unit per year as a reserve, adjusting the underwritten NOI downward accordingly. Reported statements also frequently understate payroll for properties where the owner’s family provides informal labor, or insurance for properties where the owner has portfolio policies that don’t reflect standalone asset costs.

The normalization problem is why domain expertise matters in financial AI tools. A tool that reads the reported numbers and presents them accurately is doing something real. A tool that reads the reported numbers, identifies the categories where reported differs from market, and applies the appropriate adjustments — with the conventions that experienced professionals use — is doing something categorically different. The first is useful. The second is professional-grade.

The adjustments that matter most are almost always the ones that aren’t in the document. They’re in the domain knowledge of the analyst who knows what to look for and what to add back. Building that knowledge into a tool is the actual work.

+++