A Calgary seller arrived at a listing appointment with an automated online valuation $61,000 above what recent comparable sales supported, and asked her agent to justify the difference.
At a glance
A Calgary homeowner preparing to sell pulled up an automated online home-valuation tool the night before her listing appointment and arrived with a number in hand: $618,000. Her agent’s own comparative market analysis, built from four recent comparable sales within 500 metres and adjusted for the seller’s finished basement and updated kitchen, supported a list price closer to $557,000 — a $61,000 gap the seller wanted explained before she would agree to anything.
An automated valuation model works from public records and recent sale prices in the area — it has never been inside the specific house. It cannot see a finished basement that was never permitted and therefore never recorded, a kitchen renovation completed after the last public sale, or a busy road the model’s radius-based comparables happen to avoid. None of that makes the number fraudulent; it makes it structurally incomplete in a way a walk-through comparative market analysis is built specifically to correct.
CREA itself has taken a public position on exactly this kind of AI-assisted output reaching a client’s hands: “The adoption of AI does not alter a REALTOR®’s obligations” — a member remains fully responsible for the information and services provided to a client regardless of whether AI produced or assisted with them. That principle cuts both ways here: the agent could not wave the discrepancy away by blaming the tool, and she also could not treat her own professional judgment as automatically superior without being able to explain, specifically, what the automated number had missed.
Automated online valuation: $618,000. Agent’s comparative market analysis, from four comparables adjusted for condition and finished square footage: $557,000. Gap: $61,000, or roughly 11% of the lower figure.
CREA’s national release for the same month put the non-seasonally adjusted average home price at $674,819, with the MLS® HPI down 3.3% year over year — context the seller had not seen, and a useful anchor for why a single online number, detached from any specific market’s current direction, can drift from what comparable local sales actually support.
CREA explains why it built the HPI as the industry’s own answer to single-number confusion in the first place: “Average or median prices can change a lot from one month to the next and paint an inaccurate or even unhelpful picture of price values and trends.” The HPI uses “more than 15 years of MLS® System data and sophisticated statistical models to define a ‘typical’ home based on the features of homes that have been bought and sold,” tracked by neighbourhood and housing type every month — a normalized benchmark, not a single automated point estimate from one night’s public data.
The agent walked the seller through both figures side by side: what an online AI tool can and cannot see about a specific property, and what CREA’s own benchmark methodology is built to correct for at the market level. CREA’s stated position that “transparency, accuracy and accountability” govern how AI is used in the profession shaped how she framed it — not as the tool being wrong, but as incomplete input that a proper comparative market analysis exists specifically to complete.
The seller listed at $562,000, five thousand above the agent’s comparative market analysis to leave negotiating room, and accepted an offer within eleven days at $558,000. She kept the online valuation printout and, months later, told her agent she had shown it to a friend selling a house across town — this time asking, before listing, what the local comparables actually supported.
The tell was not that the AI number was wrong in some detectable, mechanical way — there was nothing in the printout itself flagging its own blind spots. The tell was structural: any automated valuation built from public records alone will miss whatever a public record does not capture, and an unpermitted basement finish is exactly the kind of improvement that never makes it into one. An agent who can name, specifically, what a given tool could not have seen turns a pricing disagreement into a five-minute conversation instead of a week of back-and-forth.
Related reading: what the MLS® Home Price Index actually measures, and a related file where two of CREA’s own published figures pointed in opposite directions for a different reason: benchmark and average telling two stories.
A 30-minute call is enough to tell you whether AI pays for itself here.
