Type your address into any major real estate site and you'll get a number in three seconds. The problem: in Brantford, that number is often wrong by 10–20%. On a $600,000 home, that's a $60,000–$120,000 mistake.
These tools have their place. But sellers who use them as a list price — and buyers who use them as a ceiling — are making one of the most expensive mistakes in real estate. Here's why algorithms get local pricing wrong, and what actually determines what your home is worth.
What an automated estimate actually is
In plain English: it's a regression model running over recent public sales data. It looks at sale prices for homes with similar square footage, bedroom count, and lot size in your area, and it spits out a weighted average.
That's it. There's no judgment, no inspection, no knowledge of your specific property beyond what's in the public record.
The model doesn't know:
- Your kitchen has been renovated.
- Your basement is finished, or unfinished, or partially finished with a strange layout.
- Your neighbour's dog barks all night and the school bus stop is on your front lawn.
- Your street is the one in your neighbourhood that commands a premium because of mature trees.
- The MLS listing description for your "comp" said "needs total reno" — and that's why it sold $80,000 below market.
It can't see any of that. So it averages over it.
The five things online estimates can't see
1. Condition and upgrades. A $600,000 home that's been gutted to the studs and rebuilt with new mechanicals, finishes, and layout is not the same as a $600,000 home with original 1990s everything. The algorithm sees both as "$600K, 3-bed, 1,400 sqft, Brantford" and prices them identically.
2. Micro-location. A corner lot, a cul-de-sac, a busy road, a backyard onto a school yard, a backyard onto green space — these can swing value 5–15% in either direction. The algorithm groups them by postal code or neighbourhood. They look the same to it.
3. Recency. Most automated estimates lag the market by 60–90 days. In a fast-moving market, that's a meaningful gap. In March 2026 specifically, prices were down 11% YoY, and online estimates were still anchored to averages that included the slower decline of late 2025.
4. Market shift. Algorithms average over time windows. If your "comp" sold in September 2025 in a slightly different market environment, that sale is being treated as roughly equivalent to a sale from last week. It isn't.
5. Property type segmentation. This is the one I see the most. The model lumps detached, semi, and townhouse together for "homes near you." Detached in Brantford is down ~13.5% YoY. Townhouses are down ~3.6%. Mixing them gives you a number that's wrong for both.
A real Brantford example
I'll use a recent one with the details anonymized.
Three-bedroom detached, north end, 1,500 sqft, full basement, original-condition kitchen and baths, decent lot.
The major online estimates put it between $585,000 and $620,000. Average across three platforms: ~$600K.
Actual recent comps for that exact pocket — same property type, same size band, same condition tier, last 90 days — pointed to $545,000–$565,000.
It sold for $552,000.
The algorithms were $40,000–$70,000 high. Why? Two reasons:
- They blended in some renovated comps that had no business being in the comparison set.
- They lagged the YoY decline that's been working through detached homes in 2025–2026.
If the seller had listed at the algorithm's number, they would have sat for 60+ days, eventually corrected, and sold for less than $552,000 — because of the stale-listing penalty. Instead, they listed at $549,000, had three offers in eight days, and went $3,000 over ask.
That's the cost of trusting the algorithm. Not the $40K markup — the $20K under the right number you'd have eventually settled at, after a month of deepening discounts.
What actually determines value
Five things. None of them are inputs the algorithm has good access to.
- Recent (last 90 days) comparable sales. Same property type. Within 3 blocks if possible, or within the same micro-neighbourhood at minimum.
- Same size and bedroom count. A 1,200 sqft 3-bed and a 2,400 sqft 4-bed are not comparable, even if they're three doors down.
- Same or similar condition. Renovated comp ≠ original comp. Don't average across them.
- Time-adjusted. A comp from October 2025 needs to be discounted for the market shift since then.
- Lot, location, and exposure adjusted. Corner lot premium or discount, busy road, school proximity, view, backyard size.
A real CMA does all five. An algorithm does roughly the first one, badly.
How a real CMA differs from an algorithm
A few things a human (or a tool that's been built carefully) does that the typical online estimator does not:
- Comp curation. Picks the right comps and rejects the wrong ones. Excludes the renovated outlier. Excludes the estate sale that went for 80% of market.
- Adjustments for condition. Most online tools apply zero adjustment for condition because they have no way to assess it. A real CMA adjusts upward for upgrades and downward for deferred maintenance, with line-item math.
- Knowledge of off-market activity. Some sales never hit MLS. Some deals fall apart at conditions and get relisted at a lower number. Some neighbourhoods have private listing communities. None of this is in the algorithm.
- Buyer behaviour right now. What's pulling offers in the current month vs. what was pulling offers six months ago. The algorithm doesn't update its model that fast.
What algorithms get right
I want to be fair. They're not useless.
- Wide-area trend. "Prices in Brantford are down YoY" — directionally correct most of the time.
- Median price direction. Useful for context, not for pricing your house.
- Ballpark for completely standard property. If you have a brand-new, perfectly typical, freshly listed comp two doors down, the algorithm is probably close. If any of those conditions don't hold, treat it as wrong until proven otherwise.
What to do instead
- Pull real comps for your specific street and property type. Last 90 days. Same type. Filter by size and bedroom count.
- Talk to someone who's actually been inside homes in your area recently. They know which kitchen reno commanded a $40K premium and which one didn't. The algorithm doesn't.
- Use online tools for ballpark, not for strategy. They're a starting point. Treating them as the answer is one of the most expensive mistakes sellers make.