THE CLIENT
About the organization
A tax group in the United States serving U.S.-based clients.
THE CHALLENGE
The problem
A cost segregation study separates a building's cost into the components that qualify for shorter depreciation periods, so the owner can take those deductions sooner. Engineers inspect the property, put a cost on each component and build the cost workbook. That work stays with the engineers. The write-up that follows took the group 1–2 weeks per study. It meant carrying hundreds of figures from the workbook into a report of 40 pages or more, building the tables, describing the property and checking that everything tied.
The group wanted the write-up done faster. Every figure still had to trace to its source, and every judgment call had to stay with its engineers and reviewers.
OUR APPROACH
How we solved it
Split the work between code and AI
Divided the report between the two. Code handles the figures, the tables, the documents and the checks. AI writes only the paragraphs that are specific to the job.
Code reads and computes
Code pulls every value out of the cost workbook and records the tab and cell it came from. It confirms that the schedules tie, then works out the totals, percentages and depreciation tables.
AI writes, code assembles
In a single step, AI drafts the job-specific paragraphs, such as the executive summary and the property description. It never types a number. It leaves a marker where each figure goes, and code fills it in. Code then builds the Word report and the Excel workbooks in the group's house format.
Automated checks, then expert review
A checker runs 47 checks on the package before anyone sees it. It fixes what has only one right answer in the draft and does not change a figure or a classification. Missing facts print as marked placeholders with reviewer notes. The solution does not guess. A reviewer sheet lists what still needs a person, largest dollar effect first. The engineer and the tax reviewer make every call and sign off.
Let code handle every figure and the automated checks, and use AI only for the writing that is specific to the job. That split turned a 1–2 week write-up into a draft that is ready in about a minute and that engineers and reviewers can check figure by figure.
More results