There is an obvious objection to assessing a building without going to it, and it is the right one: if nobody visited the property, how does anyone know what is there?
It deserves a real answer rather than a marketing one. The honest version is not that a desk study knows everything an engineer on site would know. It is that a defined set of public and commercial data sources can establish most of a building’s physical facts to a documented standard, that the remaining gaps are knowable, and that a competent study discloses them instead of quietly modelling an average.
Here is what that actually consists of, source by source. The application is US cost segregation — reconstructing a building’s component costs so each can be depreciated over its own life rather than the whole property over 27.5 or 39 years — but the measurement problem is general, and so is most of the method.
Parcel geometry, not the listing
The first requirement is the land, and specifically the recorded parcel boundary rather than whatever square footage a listing claims.
County and state parcel services publish this as geospatial data: a polygon, an assessor parcel number, a recorded acreage. Cost Seg Smart resolves parcel boundaries through a chain of statewide and county sources covering the large majority of US orders, falling back through progressively broader sources where a county publishes nothing.
This matters more than it first appears. Site improvements — paving, kerbs, landscaping, drainage, fencing, exterior lighting — are frequently the largest short-life category in a study, and their scale is a function of how much lot surrounds the building. A method that starts from building floor area cannot see them at all. A method that starts from the parcel computes parcel area less building footprint and sizes the site works against the space that actually exists.
It also produces a useful failure mode. If the building footprint exceeds the parcel, or the implied site coverage is physically impossible, the parcel record is wrong. An arithmetic impossibility indicts its own input — a check a desk study can run continuously and a site visit generally cannot.
Imagery, with its limits stated
Aerial and satellite imagery establishes what sits on the parcel: roof form and material, storey count from shadow and oblique views, paved extent, parking layout, pools, decks, outbuildings, rooftop plant.
The limit is worth being blunt about. Imagery reads exteriors. It will measure a 1,400 square foot asphalt driveway to a defensible tolerance. It will tell you nothing about whether the kitchen has stone worktops. Any provider claiming otherwise is overselling, and the correct response to that gap is to go and ask the owner rather than to assume a mid-range specification.
The permit record
Building permits are the most under-used source in this field and often the most valuable.
A permit filing carries a scope of work, a declared valuation, a date and frequently a contractor. For a property that has been renovated, the permit record is the difference between modelling a 1974 building and modelling a 1974 building with a 2019 kitchen, a 2021 roof and a 2022 mechanical replacement — three separate assets with three separate in-service dates, which in this application is not a detail but the entire structure of the answer.
Permit coverage across US jurisdictions is genuinely uneven. Some counties publish decades of filings in structured form; some publish nothing. A study that uses permits should state which it found, and one that found none should say so.
Assessor records, treated sceptically
The county assessment roll gives assessed land and improvement values, and their ratio informs the land allocation — the portion of a purchase price that is not depreciable at all, and therefore the single largest determinant of how much basis the study is working with.
The scepticism is the important part. Assessment rolls routinely carry placeholder allocations that look like valuations. A county publishing exactly 50.0000% land, year after year, while the assessed dollars inflate, is applying a flat allocation convention rather than valuing the parcel. The right response is to detect ratios sitting on a round grid and holding stable across assessment years, hold them, and route the file to a human — not to adopt a convention because it arrived from a government source.
Two commercial data providers reselling the same county roll are also not two independent sources. They are two readers of one document, and their agreement proves the record was read consistently, not that it is correct.
Owner photographs
Interior finish level is an important residential input, and the one no public source carries. That gap is closed by asking the owner for photographs — kitchens, bathrooms, flooring, built-in joinery, plant rooms.
This is a genuine dependency on the client, and a study that receives no photographs is working with less information. Saying so in the report is better practice than modelling an average and presenting it as a finding.
Costing and reconciliation
With components identified and measured, each is priced at regionally indexed construction unit costs typical for the property type and specification tier. Anyone who has used a published cost manual will recognise the structure: base unit rate, regional index, specification adjustment.
Two details are where these studies most often go wrong, and both are construction facts before they are tax facts.
Site work is not shell work. Asphalt over compacted subgrade and a heavy reinforced concrete slab do not share a cost band. Applying building finish assumptions to yard paving overstates it substantially — on one distribution warehouse, by roughly 85% on the paving line alone.
The estimate is reconciled, not asserted. Component estimates are scaled so their total equals the property’s actual depreciable basis. The output is an allocation of a known total rather than an independent cost opinion, which is the step that most surprises people from a construction background.
That reconciliation regularly lands well away from 1.0, and it deserves an explanation rather than a scalar the reader is expected not to notice. A modelled replacement cost prices the direct construction cost of the improvements. What a buyer pays for a completed, permitted, in-service property, net of land, can also embed indirect and financing costs, entrepreneurial incentive, above-typical specification not observable to a desk model, and conditions at the transaction date. They are different quantities and rarely coincide. Age-related depreciation does not explain the gap either, because the allocation is built on replacement cost new.
What the evidence trail looks like at the end
A parcel polygon with an APN. Imagery with a date. Permits with numbers. An assessment roll with a year. Photographs with timestamps. Unit costs with a regional index. Every input traceable to something a reviewer can independently check.
The finished report runs 40+ pages with component-level schedules, and more than 100 automated quality checks run against each study before release. It includes a table mapping the thirteen principal elements described in the IRS Cost Segregation Audit Techniques Guide against the report sections addressing them — which is a reasonable thing to ask any provider in this field to produce, and a surprisingly effective filter.
What it does not buy
An engineer standing in the mechanical room. For a complex industrial asset with unusual process equipment, that is still the correct answer, and any provider suggesting otherwise is selling.
For many residential, short-term rental, small multifamily and straightforward commercial properties, these sources can provide enough evidence to identify and cost the relevant components without an on-site visit — at a fraction of the time and cost of many traditional engagements, with most residential studies priced between $495 and $1,995 and delivered the same day.
The useful question was never whether software was involved. It is what the study can show you, and whether you can check it.
