# Scoring data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → with the metrics vendors don't advertise
A rent roll lands in your inbox. Ninety-two units, twelve columns, formatted in someone's house style. It looks clean. It is not clean. Three tenants show lease end dates in the past. One unit appears twice with two different rents. The "last verified" column is blank for a quarter of the rows. Nobody flagged this before it hit the underwriting model, because nobody was scoring it. That is the gap this lesson closes.
Vendors sell you coverage: how many parcels, how many comps, how many markets. They rarely sell you the quality metrics that determine whether that coverage is usable. This lesson builds a scorecard you can run on any rent roll, comp set, or lease abstract before it becomes an input to a valuation or acquisition decision.
Real estate data is fragmented by design. Property records sit in county recorder offices. Leases sit in PDFs and property management systems like Yardi or RealPage. Comps come from brokers, CoStar, or courthouse filings. Nothing forces these sources to agree with each other.
The result: valuation models routinely run on data nobody has formally graded. A cap rate assumption built on five "comparable" transactions is only as good as whether those transactions were actually comparable, current, and correctly attributed. Cap rate here means the ratio of a property's net operating income to its purchase price, a valuation input, not the subject of this lesson. We care about the data feeding it, not the ratio itself.
Completeness measures the share of required fields that are actually populated, not blank, null, or placeholder text like "TBD."
For a rent roll, required fields typically include: unit ID, square footage, lease start/end, current rent, tenant name, and renewal option flags. A rent roll missing lease end dates on 15% of units is not 85% reliable, it is unusable for lease rollover analysis until fixed, because rollover risk is exactly what those missing fields would reveal.
Simple formula:
Completeness % = (Fields populated / Fields required) × 100Worked example: a 90-unit rent roll needs 8 fields per unit = 720 total fields. If 648 are populated, completeness = 648/720 = 90%. Industry practice generally treats anything below 95% completeness on core underwriting fields as requiring manual remediation before use, though thresholds vary by firm.
Freshness measures how old the data is relative to when a decision is made. A comp set pulled six months ago in a market where rents moved 8% in that window is stale, even if it was perfectly accurate at capture.
Freshness matters differently by data type:
Metric: age in days since last verified, per record, not per file. A file "as of January" can still contain individual line items last touched a year earlier.
Lineage means being able to trace a data point back to its origin: who captured it, from what source, when, and what transformations happened along the way. This is the metric vendors advertise least, because it exposes how much of their "proprietary" data is actually aggregated from public filings, broker submissions, or scraped listings of unknown reliability.
Ask, for every comp: is this a closed transaction from a public deed record, a broker's asking-price estimate, or a self-reported figure from a survey? These are not interchangeable. The Uniform Standards of Professional Appraisal Practice (USPAP) requires appraisers to document data sources for exactly this reason: unverifiable lineage undermines the entire valuation.
A lineage score can be as simple as a three-tier tag per record:
1. Verified primary source (recorded deed, audited rent roll)
2. Verified secondary source (broker confirmation, tax record)
3. Unverified or self-reported (survey response, scraped listing)
Match-rate measures how well records from different sources link to the same underlying asset. This is the quiet killer in real estate data because addresses are messy: "123 Main St Unit 4B" and "123 Main Street, Apt 4B" and "123 Main St #4B" may all refer to the same unit but fail an exact-string join.
Match-rate is critical when merging a rent roll against a property management export, or when reconciling a comp set against county assessor records using parcel IDs.
# Simplified match-rate check
matched = df_source_a.merge(
df_source_b, on="parcel_id", how="inner"
)
match_rate = len(matched) / len(df_source_a) * 100
print(f"Match rate: {match_rate:.1f}%")A match-rate below roughly 90% between two sources describing the same portfolio should trigger investigation, not assumption that 10% simply "didn't exist" in the other source. Often it means inconsistent parcel IDs, merged/split units, or address normalization failures.
Combine the four into a single working grade per dataset, before modeling:
| Metric | Weight (example) | Threshold for "pass" |
|---|---|---|
| Completeness | 30% | ≥ 95% core fields |
| Freshness | 25% | ≤ 60 days (rent roll) |
| Lineage | 25% | ≥ 80% Tier 1 or 2 sources |
| Match-rate | 20% | ≥ 90% cross-source |
Weights are illustrative, not universal; adjust based on what the model downstream is most sensitive to. A discounted cash flowdiscounted cash flowDiscounted Cash Flow (DCF) is a valuation method that estimates an asset's value by projecting future cash flows and discounting them to present value using a required rate of return.Voir la définition complète → model is highly sensitive to freshness of rent data; a portfolio-level risk screen may weight lineage more heavily.
Score each metric 0 to 100, multiply by weight, sum. Anything scoring below 70 overall should be remediated, not modeled.
Vérification des acquis
1. Why does the lesson describe data quality as a real estate-specific problem rather than purely an IT issue?
2. A rent roll has 92 units and is missing lease end dates on 15% of rows. What does the lesson imply about simply reporting this as '85% complete'?
3. Why does the lesson distinguish cap rate (the valuation ratio) from the data quality metrics it teaches?
4. Select ALL correct answers about why an unflagged rent roll (like the one in the opening example) poses a risk to underwriting.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers about what 'completeness' as a data quality metric actually captures.
Sélectionnez toutes les réponses correctes.
Three common failure patterns worth naming:
Silent overwrite. A property manager updates a rent roll monthly, but the file name never changes and no version history exists. Freshness looks fine; lineage is actually broken because you cannot prove which version fed last quarter's model.
Comp contamination. A comp set mixes closed sales with active listings without a status flag. Listings are asking prices, not transaction prices, this materially skews averages if unflagged. This is a lineage and completeness failure simultaneously.
False match confidence. Two datasets merge cleanly at a 98% match-rate, but the join key (say, street address) silently matched the wrong building in a market with duplicate street names across submarkets. High match-rate numbers can mask wrong matches, not just missing ones. Spot-check a sample manually, always.
🎬 [VIDEO: "Data QualityData QualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → Fundamentals" - youtube.com/results?search_query=data+quality+fundamentals+completeness+freshness - search for recent explainer content on completeness, timeliness, and lineage concepts applicable across industries, since real-estate-specific videos on this exact framework are rare]
There is no single, universally cited industry-wide benchmark database publishing "average completeness scores for US rent rolls." Firms build internal thresholds (as shown above) based on their own model sensitivity and risk tolerance. Be skeptical of any vendor or course claiming a hard external benchmark here; the honest answer is that these thresholds are largely firm-specific and should be documented, tested, and revisited, not treated as fixed industry standards.
What is well established: regulatory frameworks like USPAP for appraisal and, in Europe, valuation standards from the Royal Institution of Chartered Surveyors (RICS) require documented data sourcing and verification steps. These are governance requirements, not quality scores per se, but they reinforce the same discipline: know your source, know your age, know your match confidence, before you trust the output.