Most "AI deal finder" tools won't tell you what they're doing. We will. This page is the full ruleset — comp matching, outlier handling, confidence ratings, the three lenses (equity, flip, cashflow), and the bidding-war guard that stops us from flagging a strategic underprice as a discount. If something here doesn't pass your sniff test, push back. The rules are tunable and we'd rather hear it.
Once a week (and on-demand when I refresh it manually), the engine pulls every active TRREB MLS listing in the cities it's scanning. For each listing it tries to answer three independent questions:
A listing can win on one lens, two, or all three. The card you see on /deals shows whichever lens it wins on most strongly, with a confidence chip telling you how much to trust the math. Nothing here is an appraisal. It's a high-throughput first-pass filter so you don't waste a Saturday on the wrong listings — the real work starts when you and I open the file together.
The single biggest weakness of most "AI deal finder" tools is they apply one rigid comp filter and drop anything that doesn't match. That works great in a downtown Toronto condo building with 40 identical units sold last quarter. It works terribly in Brampton when a 4-bed semi has only one tight comp in the last 90 days.
Deal Hunter runs a three-tier search and stops at the first tier that returns at least 3 valid comps:
Each comp gets price-per-sqft normalized (we use TRREB's LivingArea + bracket midpoint when range is bucketed), then we adjust for bed count, bath count, age, and condition tier (turnkey / light / cosmetic / heavy). The adjustments are capped at 25% of comp value — beyond that we don't trust the algorithm to handle the difference and the listing falls out.
Once we have a comp set, we throw out the outliers using the interquartile-range rule. Concretely: sort the comp prices, find Q1 and Q3, compute IQR = Q3 - Q1, drop anything below Q1 - 1.5×IQR or above Q3 + 1.5×IQR.
Why bother? Because TRREB has data-entry mistakes (a $2.4M family of four mis-typed a $4.2M sale), forced sales (estate liquidations that go for 20% under market), and the occasional dual-agency lowball that doesn't reflect the actual market. Without IQR filtering, one bad comp pulls the estimated value 5-8% in the wrong direction. With IQR, the median of the survivors is what we trust.
Every deal card shows a confidence chip. It's not vanity — it gates the math. A Low-confidence Equity flag uses a stricter % threshold than a High-confidence one, because we know we might be wrong.
| Confidence | What earned it | What it changes |
|---|---|---|
| HIGH | 5+ Tier-1 comps survived IQR · low price variance (CoV < 12%) | Equity threshold relaxes to 8% below comp value |
| MEDIUM | 3–4 Tier-1 comps, OR 5+ Tier-2 comps · moderate variance | Equity threshold tightens to 12% below comp value |
| LOW | Tier-3 fallback only · high variance · sparse data | Equity threshold tightens further to 16% below comp value — we need a real margin of safety before we flag this |
We compute equity_pct = (comp_value − list_price) / comp_value. If that's above the confidence-aware threshold (8% / 12% / 16%) and the bidding-war guard doesn't fire (see below), the listing is flagged as an Equity deal.
The card shows the dollar gap ("$184,018 under market") and the percentage ("23.5% below comp value"). The comps behind the number are visible to paying members on the card — click "8 comps back this value" to expand.
We score this when a listing is in cosmetic or heavy condition and either:
We deliberately don't enforce a hard "70% rule" the way some US flip tools do. In the GTA, retail homes never trade at 70% of ARV — that rule of thumb just kills the whole feed. Discount + condition signals + your own walk-through tell the real story.
For each listing we estimate market rent using TRREB rental comps in the same FSA / subtype / bed-count band over the last 180 days. We compute:
(rent × 12 − operating costs) / list price.A listing wins the Cashflow lens if monthly cashflow ≥ $0, OR cap rate ≥ 5.0%, OR cash-on-cash ≥ 6%. Honest note: in 2026 GTA most resale residential does not cash-flow with 20%-down financing. When the feed shows 0 cashflow deals it's because the market is what it is, not because the engine is broken.
In Toronto and Mississauga it's standard practice to list a $1.4M home at $1.19M and hold offers for 7 days. To a naive comp-match algorithm that's "16% below market — buy it!" In reality the seller is fielding 9 offers and the home will sell at $1.45M.
Deal Hunter detects this pattern and refuses to flag the listing as an Equity deal. The guard fires on either of:
When the guard fires the listing still appears in the feed, but it gets a "⚠ Likely underpriced for offers" tag and won't carry the Equity lens badge. We're explicit about why — so you don't lose two weeks chasing a phantom discount.
It's important to be square about what we can and can't do here:
Last methodology change: 2026-05-26 — added the bidding-war guard and the confidence-aware equity thresholds. If you notice a deal that you think the engine got wrong (in either direction), email me — that's how the rules get better.
The latest scan surfaced 197 deals across the GTA. Take a look.
Open this week's deals →We use cookies for analytics and, with your OK, advertising to show you relevant homes. See our Privacy Policy. You can change this anytime.