AI property valuation in Quebec: how CoteQC works

AI property valuation in Quebec is not a magic algorithm guessing a price. It's a methodical cascade that leans on real recent sales, normalized by neighborhood and type. CoteQC, the valuation engine powering CourtiConnect and the public Estimator, sits on 10,240 sales (as of April 25, 2026) across 27 areas and 131 neighborhood × type groups. Real statistical accuracy exceeds 90% on 92% of groups. To value your property right now, use the free CoteQC estimator.

The principle: 6-level cascade

CoteQC queries the most statistically relevant data first. Level 0: neighborhood sales at exact type over the last 12 months. Level 0.5: full-city sales at exact type. Levels 1 and 2: active listings (market_listings) at exact neighborhood. Level 3: broad residential ratio for the neighborhood. Level 4: full city at exact type. Level 5: fallback if no reliable data is available.

The engine descends the cascade automatically as long as the sample doesn't allow acceptable statistical accuracy (n ≥ 3 minimum). Benefit: no fragile median on 1 or 2 sales, no overvaluation from a single atypical comparable.

Statistical accuracy: official formula

CoteQC doesn't just display a number. Each estimate ships with a precision score: precision = (1 − stdErrPct / 100) × 100, where stdErrPct = (stddev / sqrt(sales count) / mean) × 100.

Display thresholds are strict: under 5% error, label is "95%+". Between 5 and 10%, "90-95%". Between 10 and 15%, "85-90%". Above that, "Indicative estimate". Below 3 sales, the engine displays "Insufficient data" rather than risking a false signal.

The 5 standard types: condo, house, duplex, triplex, plex

CoteQC deliberately restricts types to five normalized categories: condo, single-family house, duplex (2 units), triplex (3 units) and plex (4+ units). The "other" Centris source category is explicitly excluded because it aggregates mistyped listings (commercial, land, partial sheets) that would skew neighborhood medians.

That discipline is what allows reaching 90%+ accuracy on 92% of groups. Without it, the estimate would slide toward a biased, unusable average.

Living area sources: 4 levels

To compute a reliable price per square foot, CoteQC uses a hierarchy of living area sources. Source 1: explicit Matrix Centris value (17% of listings). Source 2: building dimensions in meters extracted from the listing then converted (additional 39%). Source 3: City of Montreal open data — 512,288 property assessment units (additional 14.5%). Total current coverage: 71%.

For listings with no identifiable area, the engine relies solely on the neighborhood × type median rather than fabricating a speculative price per square foot. Methodological honesty over advertised performance.

Owned limits: what CoteQC does not do

CoteQC does not replace a municipal assessment (used to compute property taxes) or a chartered appraisal (required for bank financing, estate, divorce or litigation). It's a quick framing tool to prepare a brokerage mandate, calibrate a purchase offer or estimate a real estate portfolio.

Hamza Taleb, OACIQ broker at RE/MAX (438 877-8525), uses CoteQC alongside a manual comparative analysis and an in-person property visit on his mandates across Quebec.

Conclusion: transparency first

A credible AI valuation publishes its methodology, sources, real accuracy and limits. CoteQC ships those four elements with every estimate. That's what lets a broker, buyer or seller decide on facts rather than absorb an opaque number.