The Cost Segregation Playbook

Every study begins where the money actually landed: the client's own accounting record of the property. That record, not a reconstruction of it, establishes what there is to depreciate. A study that quietly re-derives basis from construction paperwork when the books are sitting right there has already introduced a difference it will spend the rest of the engagement defending.

Getting from the books to depreciable basis means separating out everything that cannot be depreciated — the land, the costs that never entered a recovery period at all, and the items capitalized outside the MACRS system. Each of those is a category with its own treatment and its own traps, and the amounts are rarely labelled helpfully.

This is where our reconciliation engine earns its place. It performs that separation across the whole record and then proves the result against the document's own internal arithmetic, which is redundant by construction — a set of figures that does not tie is surfaced as an open question rather than silently absorbed into a total. It does that in seconds across thousands of lines, which is the part no team does reliably by hand at three in the morning before a filing deadline.

It does not, however, decide anything. An accountant who has spent a career in fixed-asset ledgers confirms the resulting basis before a single dollar is allocated. The most expensive error we find on prior studies lives exactly here: a basis taken from a total that looks authoritative and quietly carries land or non-depreciating costs inside it. Every figure downstream inherits that error, and no amount of careful classification later will remove it.

A study is only ever as good as what it is built from, and the instinct to ask for everything is the wrong one — it delays the engagement and buries the few documents that carry real weight. We ask for six categories of record: the accounting record of the asset, the contractor's cost documentation, the design and construction drawings, dated evidence of the property as it was actually built, the acquisition paperwork where a property was bought rather than built, and the property's own prior depreciation history if it has one.

Which specific documents satisfy each category depends entirely on the property, the delivery method and how the owner keeps records — a ground-up development and a twenty-year-old acquisition answer the same six questions with completely different paper. Part of the job is knowing which substitutes are acceptable and which are not, and that is a judgement, not a checklist.

Our extraction models read across the whole set and assemble a candidate component register in which every line is traced back to the page and position it came from. Two things about that matter more than the speed. Nothing is presented as a fact unless the document proves it — where a figure cannot be anchored to a specific row of a specific page, it is raised as an open item rather than estimated into place. And the register is a proposal. It is read line by line by an engineer whose career has been spent in the built environment, who knows what a mechanical system costs because they have priced one, and who will strike a line the models were confident about and confident wrong about.

That review is the gate. Volume is a machine problem and we treat it as one; knowing that a line item labelled one thing on an invoice is physically something else in the building is not, and never has been.

Classification is where the study earns its money. Each identified component is assigned a recovery period that traces to a specific IRS reference, not to a convention:

A finished study exists in more than one place: the underlying record, the working schedules a CPA will read, and the written report that explains the reasoning. Those are produced independently of one another, and the whole point is that they should therefore say exactly the same thing. A number that appears differently in two of them is not a formatting issue. It means something moved, and nobody noticed.

So before anything is delivered, all three are generated fresh and compared against each other, figure by figure, down to the cent. Totals must reconcile to the basis established in step one, the first-year position must be identical everywhere it appears, and the written narrative must describe the same property the schedules do. Any disagreement stops delivery. Not flags it — stops it.

The comparison itself is automated because it has to be: no person reliably eyeballs thousands of figures across three documents, and the errors worth catching are precisely the ones too small and too consistent to notice by hand. What is deliberately notautomated is what happens next. When the surfaces disagree, a specialist finds out why — whether a reclassification was applied in one place and not another, whether a rerun moved something quietly, whether the change was correct and the report simply stale. The software can tell you two numbers differ. Deciding which one is wrong, and why, is the work.

This gate exists because we have watched studies drift. It is the reason a reclassification made on a Tuesday cannot reach a client on a Wednesday without every downstream figure moving with it.

Bonus depreciation under §168(k) depends on when the property was acquired:

Every study reports bonus at the rate the acquisition-date test determines, per component. There is no blended rate. The placed-in-service date drives the depreciation start; the acquired date drives bonus eligibility.

§754 / §743(b) partnership buy-outs: purchase-triggered basis adjustments after Jan 19, 2025 qualify for 100% bonus on the short-life slice per Treas. Reg. §1.168(k)-2(b)(3)(iv)(D). Inheritance- triggered §743(b) adjustments do not. Full mechanic: §754 / §743(b) partner-basis primer.

A completed study ships:

• Component classification list, tied to source documents.
• MACRS depreciation schedules for every classified component.
• Excel workpaper + PDF narrative report.
• Form 3115 / §481(a) support for look-back studies (Rev. Proc. 2015-13 automatic method change). No amended returns required; the catch-up deduction is taken in the year of the change.

For CPAs preparing the client's return, we provide a Form 4562 crosswalk sheet — study figures → return-line numbers. The CPA transcribes into their own filing software. We never file for them.