Auditing a data vendor before you sign the contract

Why vendor audits matter more in real estate than in other asset classes

Real estate is structurally opaque. Unlike public equities, there is no centralized tape of every transaction. Data comes from a patchwork of county recorders, brokers, surveys, and scraped listings. That fragmentation means two vendors covering "the same" market can show cap rates 50 to 100 basis points apart, simply because of different sample construction.

Contrast this with public markets, where a Bloomberg terminal and a Reuters terminal will show you the same closing price. In real estate, the vendor's methodology *is* the product. If you don't audit it, you're licensing someone else's blind spots.

The three-part audit framework

1. Update cadence: how fresh is "current"?

Ask for the actual timestamp logic, not marketing language.

• Transaction data (sale prices, cap rates): typically lags 30 to 90 days because deals close, then get recorded at the county level, then get ingested by the vendor. CoStar and Real Capital Analytics (now part of MSCI) are considered industry-standard for US commercial transactions, but even they carry a reporting lag.
• Listing and asking-rent data: can be near-daily if scraped, but "near-daily" scraping still misses off-market and negotiated deals, so it systematically overstates asking rents versus effective rents.
• Vacancy and fundamentals data: often quarterly, tied to broker surveys (common for CBRE, JLL, Cushman & Wakefield research).

Audit test: pull the same metric (say, Manhattan office vacancy) from two vendors dated the same week. If they diverge by more than 200 to 300 basis points, ask both for their as-of date and sample size. The divergence is usually a cadence or sample problem, not a market mystery.

2. Sample bias: whose buildings are actually in the dataset?

This is the audit step most buyers skip, and the one that causes the most expensive mistakes.

Ask three questions:

1. What's the universe? A vendor covering "US multifamily" might only track properties over 50 units, missing the small-balance stock that dominates many secondary markets.

2. How is the sample sourced? Broker-submitted data over-represents institutional-grade, professionally marketed assets. Publicly recorded deed data over-represents smaller, less complex transactions (because large portfolio deals are often structured to avoid a single recordable event).

3. Is there survivorship bias? Repeat-sales indices (like the Case-Shiller Home Price Index, a well-established methodology) only include properties that sold at least twice. That excludes new construction and long-hold assets, which can bias the index toward more liquid, more traded submarkets.

Worked example of sample bias distorting a benchmark:

Suppose a vendor reports average US industrial cap rates at 5.8% (illustrative, not a real current figure) based on 200 tracked deals. If 150 of those 200 deals are Class A logistics assets near major ports (because those are the deals brokers push hardest to data vendors), the "national average" is really a large-asset, coastal-market average. A Class B infill industrial building in a secondary market could realistically trade 100 to 150 basis points higher, and the vendor's number would mislead you about it.

The fix isn't to reject the vendor. It's to ask for a sample breakdown by asset class, geography, and deal size before you apply their benchmark to your specific deal.

3. Methodology disclosure: will they show you the recipe?

A vendor confident in its data will publish (or at minimum, share on request under NDA):

• Data sources and how they're weighted
• Index construction method (repeat-sales, hedonic regression, appraisal-based)
• Revision policy: do they restate historical figures when late data arrives?
• Outlier handling: how do they treat a clearly mispriced or distressed sale?

Red flag: "proprietary methodology, cannot disclose" with zero white paper, zero methodology PDF, and zero named data scientist or economist backing the number. Reputable vendors like NCREIF (National Council of Real Estate Investment Fiduciaries, which runs the widely cited NPI, the NCREIF Property Index) publish detailed methodology documents precisely because institutional buyers demand it.

A simple scoring rubric to run in the sales meeting

| Dimension | Question to ask | Green flag | Red flag |

|---|---|---|---|

| Cadence | "What's the median lag between event and data availability?" | Specific number, by data type | "Real-time" with no specifics |

| Bias | "What percentage of your sample is broker-submitted vs. public record?" | Clear percentage breakdown | "We aggregate everything" |

| Methodology | "Can I see the index construction white paper?" | PDF exists, peer-reviewed or industry-audited | Verbal explanation only |

| Coverage | "What's excluded from this universe?" | Named exclusions (e.g., sub-$5M deals) | "We cover the whole market" |

Score each 0 to 2. Anything below 5 out of 8 is a license fee you're paying for a marketing narrative, not a defensible dataset.

A quick technical gut-check: comparing two vendor feeds

If you have APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.Voir la définition complète → or CSV access to a trial dataset, a basic reconciliation script forces the bias question fast:

This won't tell you who's "right." It tells you where to point the methodology questions.

🎬 [VIDEO: "How Real Estate Data Actually Gets Collected" - https://www.youtube.com/results?search_query=how+real+estate+data+gets+collected - search for recent explainer content from CRE data providers or NCREIF walking through transaction data pipelines, useful for visualizing the lag between deal close and data availability]

Negotiating the contract with your audit results

Once you've scored the vendor, use it in the negotiation, not just the decision:

• Low cadence score: Push for a discounted rate on the lagged tiers, or ask for a "fast feed" add-on if one exists.
• Low bias transparency: Request a sample metadatametadataDonnées sur les données, informations décrivant le contexte, la structure, la provenance et les caractéristiques d'un asset de données (auteur, date, format, source, définition). field (asset class, deal size, source type) delivered with every record, not just the aggregate number. This turns their black box into your own segmentable dataset.
• Low methodology disclosure: Ask for an indemnification or accuracy clause tied to specific error rates, or a right to audit sample composition annually. If they refuse both disclosure and contractual accountability, that itself is the answer.

Also check regulatory exposure on data sourcing. In the EU, vendors aggregating property or ownership data may touch GDPR (General Data Protection Regulation) if any personal data (individual owner names, for instance) is embedded in "commercial" datasets. In the US, county recorder data is public record, but scraped listing data can raise contractual issues with the source platform (MLS, Multiple Listing Service, terms of use). Ask your vendor how they're licensed to resell scraped data, not just how they collect it.

Key Takeaways

• Test cadence with a real timestamp, not a marketing claim: ask for median lag by data type (transactions, listings, fundamentals) and verify with a same-week cross-vendor comparison.
• Sample bias is the single biggest hidden cost in real estate data: always ask what's excluded (deal size floors, geography, broker-submission bias) before applying a "national average" to your specific asset.
• Demand methodology disclosure in writing. Reputable providers (NCREIF, MSCI/RCA, Case-Shiller-style indices) publish their construction logic; "proprietary, can't share" is a legitimate reason to walk away.
• Use a simple scoring rubric (cadence, bias, methodology, coverage) in the sales meeting itself. It converts a vague "does this feel trustworthy" gut check into a documented basis for price negotiation.
• Bake audit rights into the contract: metadatametadataDonnées sur les données, informations décrivant le contexte, la structure, la provenance et les caractéristiques d'un asset de données (auteur, date, format, source, définition). fields per record, annual sample composition review, and disclosed revision policy are cheap contractual asks that pay for themselves the first time the vendor's number and reality diverge.