Selling models and market structure
The Wrong Opponent
6 min read
The dominant story in residential valuation technology is a contest between the agent and the algorithm. It is a good story and it describes the wrong contest.
An automated estimate does not take a listing. It cannot sit at a kitchen table, agree with a seller, or sign an agreement. Whatever it does to the market, it does by changing what a seller expects before any agent arrives. The listing itself is always lost to a person.
Where the anchor becomes a signature
The sequence is worth separating into its two moves, because they are usually collapsed into one.
First, the estimate sets an expectation. The seller arrives holding a figure formed over several weeks of screen time, and it functions the way any anchor functions: every subsequent number is evaluated as a distance from it rather than on its own evidence.
Second, an agent converts that expectation into a listing agreement by agreeing with it. This move requires a human, and it requires one with a commercial incentive to please. The estimate has no such incentive, which is precisely why it is the more tractable half of the problem.
Conflating the two produces an industry conversation aimed at the half that cannot be negotiated with.
TWO DIFFERENT PROBLEMS
| The automated estimate | The competing agent |
|---|---|
| Sets the expectation | Converts it into an agreement |
| No commercial incentive | Paid to win the listing |
| Wrong for reasons that can be shown | Wrong for reasons that can be denied |
| Cannot be persuaded | Will not be persuaded |
| Addressed with evidence | Addressed with a durable record |
Most tooling and most published advice targets the left column. The listing is lost in the right one.
Why the misdiagnosis persists
Three reasons, and none of them are stupid.
The algorithm is a safe adversary. Writing against a model offends nobody and costs no relationships, while writing about agents who present unsupported numbers implicates the readership. The comfortable target gets the coverage.
The algorithm is also legible. Its method can be described, its failures demonstrated, its blind spots enumerated. A competitor's pricing decision leaves no artefact at all and is not observable from outside the room.
And a product can be sold against a model. There is no product that stops another agent agreeing with a seller. What exists instead is a record that stays legible after the meeting ends, which is a harder thing to build and a harder thing to market.
What follows from getting the opponent right
If the loss occurs in a room, against a person, then the tooling objective changes shape. The goal is not to produce a number that beats an estimate. Beating the estimate is easy and does not decide anything.
The goal is to produce a case that survives the agent's absence. A seller who signs elsewhere carries away whatever they can still reconstruct, and what they can reconstruct is the only thing operating on their behalf at day 45 when the listing has not moved. A figure with no visible derivation reconstructs as nothing. A figure with its adjustments and exclusions attached reconstructs as an argument.
This also reframes what an instant offer competes with, since the instant offer is a priced convenience rather than a market read, and the agent who loses to one has usually lost to the certainty rather than to the number.
Do automated valuations cause agents to lose listings?
Indirectly. The estimate sets the expectation a seller brings to the appointment, but it cannot enter into a listing agreement. The listing is awarded to an agent, and where price disagreement decides it, the winning agent is frequently the one who agreed with the expectation rather than the one who tested it.
Why is price disagreement such a common reason for losing a listing?
Because agreeing costs the competing agent nothing at the moment of decision and the consequence arrives weeks later. By then the agreement is signed. The incentive structure rewards validation at the table and penalises it during the marketing period, and only one of those two moments is visible to the seller when they choose.
What competes with a competitor's higher number?
A derivation the seller can still follow without the agent present. Opinion against opinion resolves in favour of the more agreeable opinion. A visible comp set with stated adjustments and stated exclusions continues to operate after the meeting, and it is what makes a return call possible when the higher number fails to produce offers.
A record, not an argument
The industry has spent several years learning to explain why a model is wrong about a particular house. That skill is real and it is aimed at something that was never going to sign the agreement.
The estimate is Context Blindness™ in its familiar form: confident, quoted to the dollar, missing the facts that decide the answer. What the framing misses is that the damage is done by whoever agrees with it for commercial reasons, and that person is not blind at all. They are looking at the same comparable sales and choosing differently. A reasoned valuation, the approach CMAflow builds, is not primarily a better answer than the model. It is a record that keeps making the case after the person who made it has left the room, which is the only place the actual contest is decided.
This article is general information and analysis, not financial, lending, or appraisal advice. Verify any home value with a licensed professional before acting.
The Independent Agent
Substack | Spotify | CMAflow FAQ | YouTube | Free CMA | Home valuation | Insights | Blog