Type an address, answer the questions an appraiser would ask, and get a number with its evidence: the comparable sales it rests on, how each was adjusted and weighted, the market index that brought them to today, and how confident the model is. Built the way appraisal is taught, run on the sales in your own area, at the moment of the request.
The engine follows the direct-comparison method: take each recent sale that resembles the subject, bring its price to today, adjust it for the ways it differs, weight it by how much it can be trusted, and reconcile. Every rate it uses is derived from the subject's own local sales at run time. Nothing comes from a fixed regional table.
Recent sold candidates are constrained to the subject's property class and a nearest-first ladder: everything within 1.6 km, widened to 3 km and then 6 km only while fewer than eight sales qualify. Six kilometres is a hard ceiling.
Closings from the last 24 months · prefers eight or more · never past 6 km
Comparable prices are indexed to the effective date using monthly sold prices for the city and property class, when both months have adequate volume. A sale from ten months ago is no longer treated exactly like one from last week.
If the segmented index lacks volume, no market-condition adjustment is invented and the report says so
Feature rates are medians of robust locally matched-pair slopes: what the market actually paid for one more garage space, one more bathroom, a finished basement, in this area. A feature the local pool cannot price still improves comparable selection, but receives no invented dollar adjustment.
A rate appears only when at least six matched pairs agree on direction
Each comparable's indexed price is adjusted for how it differs from the subject. Weights combine similarity, recency, data completeness and gross-adjustment reliability, so a close, recent, well-documented sale counts for more.
A comparable needing more than 25% total adjustment is not comparable, and is discarded as appraisal practice requires
The conclusion is the median of the weighted mean, the weighted median and the trimmed weighted mean, then cross-checked against a local hedonic estimate and the subject's own indexed previous sale. When approaches disagree by more than the model expects, the report flags it.
Smaller agreement between approaches supports confidence; large spread lowers it
Comparable count, average similarity, recency, data completeness, gross adjustment, evidence coverage and approach agreement each contribute to one score, graded low, medium, high or very high. The likely range is the weighted 20th to 80th percentile, never narrower than an uncertainty floor that widens as confidence falls.
A disagreeing previous sale lowers the score rather than being ignored
Two things make the number defensible: a model that learned how certified appraisers reason, and live market statistics for the subject's own area at the moment you ask.
We brought a handful of certified property appraisers' working knowledge together into a curated training dataset: how they qualify a comparable, which adjustments they will defend, when they discard a sale, how they reconcile competing indications. The valuation model was trained on that material, and it is what the engine applies to every request.
At run time the model pulls the live picture of the subject's own market from the board feed your data agreement covers, and uses it to find the closest match and bring every comparable to today.
Segment by segment: the index is built for the city and the property class, and same-building sales get the strongest preference for a condominium.
The layer your visitor talks to is a language model. It takes an address from the hero or the assistant bubble, asks the appraiser's questions in plain language, hands the answers to the valuation engine, and explains the result, the range and its confidence in words a homeowner understands. The number itself always comes from the engine.
An estimate, not an appraisal. Estimates are informational and must not be presented to a consumer as a professional opinion of value. Sold comparable details are shown only after a signed-in visitor accepts the MLS® Terms of Use your board requires.
Four steps on your own website, in your own colours. Each question either improves comparable selection or supports a defensible adjustment. Nothing is asked for decoration.
Classify the subject property. Start typing a Canadian address; add a unit if there is one.
Describe the building. Above-grade living area only; finished basement space is collected separately.
Describe the site and features. Frontage and depth stay separate because shape matters beyond lot area.
Review the appraisal inputs, then run. Renovations are treated through condition and matching; project cost is not added directly to value.
The homeowner keeps a report. The agent sees the same one in the CRM, and the presentation builder prefills from it. Nothing in it is a black box.
| Comparable | Sold | Distance | Adjusted | Gross adj. | Weight | |
|---|---|---|---|---|---|---|
| A | 17 Woodfield Rd | $1,349,000 | 0.6 km | $1,301,000 | 6.1% | 0.21 |
| B | 88 Ashdale Ave | $1,262,500 | 0.9 km | $1,288,400 | 7.4% | 0.19 |
| C | 203 Coxwell Ave | $1,190,000 | 1.1 km | $1,266,900 | 11.8% | 0.15 |
| D | 41 Rhodes Ave | $1,318,000 | 1.4 km | $1,279,600 | 8.9% | 0.17 |
| E | 9 Craven Rd | $1,235,000 | 2.2 km | $1,271,300 | 10.2% | 0.14 |
| F | 120 Kent Rd | $1,405,000 | 2.7 km | $1,296,800 | 13.6% | 0.14 |
The estimate sits inside its likely range with a marker, and the confidence grade sits beside the label, because confidence is a state of the estimate, not a footnote.
Sold price, distance, adjusted price, gross adjustment and weight for each sale used, with the ones discarded for over-adjustment named and struck out.
Direct comparison, a local hedonic cross-check and the subject's own indexed previous sale, with the spread between them shown.
For each feature the model priced: the median contribution, the interquartile evidence, how many matched pairs supported it, and its confidence. Rates appear only when at least six robust local pairs agree on direction.
Every limitation is written down. Sold MLS® records stay hidden until the signed-in visitor accepts your board's Terms of Use; the estimate itself remains usable.
Every surface that runs a valuation goes through one place, so a brokerage never has a funnel giving one number while its listing pages quietly give another.
What is my home worth? on your domain, in your colours, with a report the visitor keeps.
"What's my place at 88 Yorkville worth right now?" gets the range and the confidence in chat, then offers the full report.
An estimate box on each property, so a buyer sees value next to the asking price.
"What your home is worth" and "Pricing strategy" prefill from the same engine for the listing appointment.
Saved valuations in the client's account, a downloadable report, and an external endpoint for your own tools.
AI Valuation V2 is the model every Growth and Dominance AI site runs by default, on your own domain and under your own brand. Every valuation request lands in Leads with what the visitor was looking at, and the seller can track it from their account. Dominance AI adds the AI Assistant that starts valuations from chat.
Unlimited agents, board-approved property search, alerts and CRM. AI Valuation V2 included and on by default.
Also included in Dominance AI (CAD $999 a month), where the AI Assistant, Marketing Studio and the presentation builder prefilled from the engine come with it. Comparable sold data comes through your board's VOW agreement.
No. It is an estimate produced by a valuation model, for information. It must not be presented to a consumer as a professional opinion of value, and the report says so. What makes it useful is that every figure carries its evidence: the comparables, the adjustments, the index and the confidence.
From recent sold closings in your board's data feed, under your VOW agreement, within a nearest-first ladder of 1.6, 3 and then 6 kilometres around the subject and the same property class. Sold details are shown only after a signed-in visitor accepts the board's Terms of Use; the estimate itself stays usable.
The range is the weighted 20th to 80th percentile of the adjusted comparables, never narrower than an uncertainty floor. The grade combines comparable count, similarity, recency, data completeness, gross adjustment, evidence coverage and how closely the three approaches agree. Few similar sales means a wider range and a lower grade, and the report says why.
Not directly. Version 2 treats renovations through the reported condition level and comparable matching, because the market pays for condition, not for receipts. The report states this in its limitations.
On a curated dataset built from certified property appraisers' working knowledge: how they qualify comparables, defend adjustments, discard sales and reconcile approaches. We add to that dataset and refine the model continuously, and each candidate version is run side by side against the live one on real subjects before it replaces anything. At run time the model is combined with live market statistics for the subject's own area.
AI Valuation V2 is the default model on Growth and Dominance AI. Under Settings › Valuation you can choose which model your site runs. Existing saved valuations are never re-run or reinterpreted, so reports clients already hold do not change.
Start on Dominance AI, connect your board feed, and What is my home worth? runs on your domain with the comparables, the adjustments and the confidence shown to every visitor.