Property Revaluation Experts · Co-Founder & COO case study
What the deed actually said
I built two systems for one county: a recalibration pipeline that runs Georgia's statutory sequence for correcting property values, and a separate reconciliation workflow that checks the sales data behind a ratio study against the actual recorded deed. The pipeline moved the numbers. The reconciliation workflow is what held up when the resulting study got challenged.
The result
The same digest correction also reduced what the county owes annually toward its state school-funding formula obligation, since that calculation runs off the same digest and timber figures. That relief is real but genuinely hard to isolate to a clean dollar figure — it spans two different study years and is gross of a statewide cap that isn't independently confirmed — so it's described here rather than quoted as a number.
The recalibration pipeline
O.C.G.A. § 48-5-343 doesn't leave the sequence up to interpretation. Every scenario this pipeline produces has to clear it in order, and every one gets validated against Georgia statute and IAAO dispersion standards before it's even eligible for delivery — not reviewed after the fact, gated before.
Verify the input data
Parcel and sales records get rebuilt and reconciled before anything downstream trusts them.
Diagnose the ratio
Every sale is stratified by property class and neighborhood; the study runs at the stratum level, not one countywide average.
Calibrate land first
Land value gets corrected before anything built on it does, so downstream cost corrections start from a true base.
Model cost & depreciation
Replacement cost starts from national tables, then gets corrected class by class against real local sales. Depreciation is fit from what older properties actually show, with several methods scored against each other.
Validate for compliance
Coefficient of dispersion, price-related differential, and assessment-ratio band all have to clear Georgia's own thresholds, and the same check runs no matter which model produced the scenario.
Deliver board-ready
A partial-data result gets flagged as a pilot or blocked outright — never delivered quietly as final.
The reconciliation workflow
When a county's sales-ratio study gets challenged, the state doesn't take a correction on faith — it wants the evidence behind every disputed sale. I built a separate Claude-driven workflow for exactly that: pull the disputed sample set, check each one against the actual recorded deed instead of the field note that first went into the file, and document what the record actually supports before anything gets resubmitted.
Run against this county's study, it found three samples resting on assumptions the deed didn't support: one sale logged as spanning two counties based on the buyer's mailing address, when the deed showed the whole transaction sat in one; what looked like three separate sales in one file, which turned out to be a single conveyance and an unrelated release, misfiled together; a sale checked as one parcel when the deed actually conveyed three lots priced as one. Reconciling those, along with eight other disputed samples, moved the study's ratio to 28.58. The state auditor's office reviewed the correction and put its own position in writing: the study, as revised, is now correct.
Standing rules
- No method grades its own homework. The validation layer runs identically regardless of which valuation model produced the scenario.
- Holdout, every time. A portion of sales is held out and never touched during calibration, so the fit gets checked against data it hasn't seen.
- Document the correction, not just the number. Every class-by-class cost correction carries its data support, method, and confidence — never applied as one blanket adjustment.
- Coverage before delivery. Nothing reaches the county without clearing a coverage threshold first.