Your Zillow Zestimate is wrong. Here’s why.

A homeowner opens a consumer app and treats the automated number as carved in stone. Automated valuation is a $4.2 billion industry. It is not one number. The same house will print wildly different values depending on which machine is guessing, and for whom.

There are three tiers. Institutional and lender models — Cotality, ICE / Black Knight, Veros, Quantarium, Clear Capital — are built for risk and capital markets. Investor models — HouseCanary, PropStream, and their cousins — hunt ARV and yield. Consumer portals — Zillow, Redfin, Realtor.com — are lead-generation engines. They are not the first two tiers wearing a friendlier typeface.

The trick is anchoring. On-market, Zillow’s advertised median error looks tight (about 1.77%) because the model can smell the human list price on the MLS and sit next to it. Off-market, that error jumps to about 7.06%. Redfin and HouseCanary degrade the same way. On a $1,000,000 Southern California house, a 7% miss is $70,000. That is not a rounding error in a probate net sheet.

Machines fail in four places humans do not: they cannot see a 1990s kitchen versus a chef’s kitchen on the same floor plan; they starve in data-thin neighborhoods; they cannot smell mold or see settlement; and they train on closings that were negotiated 30 to 90 days ago. A probate-aware CMA is a walkthrough, a nowcast, and a conversation about authority — IAEA versus confirmation — not a screenshot.

Stop letting a Silicon Valley lead form set the value of an estate’s largest asset. Let the machines count the bones. Let a local specialist price the house. Then bring that number to counsel before anyone prints a flyer.