What does a good first success look like, and how do you engineer it?

Larkhaven Realty sells PriceLantern to independent brokerages. Ondine Fabris has worked as an agent for nine years and just turned on the tool for her first new listing of the month, a two-bedroom bungalow two streets from a row of freshly renovated flips.

How to answer this, stage by stageSix moves, in the order you'd actually say them out loud.

Let's learn

PriceLantern is a tool that reads a new listing's address and pulls a suggested asking price with a short comp rationale, in about twenty seconds.

Before it, Ondine spent close to forty-five minutes on a routine listing: pulling comps from the MLS by hand, driving past two or three of them if anything looked off, adjusting for a renovation she remembered from a previous walk-through, then writing up her reasoning for the seller.

Four manual steps, and every one of them lived entirely in her own judgment, built over nine years.

With PriceLantern, that twenty-second number can look almost too easy. And the turn here isn't that the tool gets a number wrong sometimes. It's what a new agent does with a confident-looking number on day one, before they've ever seen it fail.

At its worst: a first-time user sees one clean number, believes it because it looks tidy, and prices a listing off a comp that was actually a gutted flip house with a totally different finish level. The seller's home sits on the market three extra weeks at the wrong price.

What I would leave alone: for a routine tract-home listing with a dozen near-identical comps on the same block, the full three-comp breakdown barely changes her trust either way. The ceremony earns its keep on unusual properties, not cookie-cutter ones.

The lesson: the first thing a probabilistic tool shows a new user is not a demo. It's a lesson in how to use it, and it teaches that lesson whether you designed it on purpose or not.

Now here is the same thing as a storyThe short version is above. Read this for how the anchor actually got picked.

Ondine can tell a listing that will sell itself from one that needs real work within the first five minutes of walking through the front door. Nine years of it.

The first two weeks with PriceLantern went well. Most of her listings were ordinary two- and three-bedroom homes on streets she already knew cold, and the suggested price matched her own gut within a few thousand dollars almost every time.

This one screen is the whole anchor. Every part of it is doing a job.

Then came the bungalow near the row of flips. PriceLantern's suggestion looked just as clean and confident as every other one that month, one bold number, no caveats attached to it in the early version of the design.

The design decision only matters on the right side. That's the whole point of building it before day one.

In the version with match tags, she opens the same listing and sees one comp marked "stretch," with a one-line note that its listed square footage jumped since the last sale. She checks it in ten seconds, drops it from her reasoning, and prices the home using the other two.

None of these three make the anchor stronger. They only make sense once trust in the basic number is already there.

Without the tags, in the version we nearly shipped, she has no way to tell the shaky comp from the solid ones. She either has to redo the whole comp pull by hand anyway, which erases the entire time saving, or she trusts the clean-looking number and prices the home wrong.

The habit that matters, a quick, confident check, gets built by use four, not use one.

The real cost was never the twenty minutes saved or lost on one listing. It was whether Ondine walked away from her first hard case trusting the checking process, or trusting a number she had no way to question.

I would take back the single bold number on that first screen. It made a great five-minute product demo to a brokerage owner deciding whether to buy the tool at all. It never once made sense for the agent actually using it on a hard case.

SPARK, the anchor in one screenFive letters. One of them is the whole design decision.

The recap, one line per letter: situation is the forty-five-minute manual comp pull, payoff is trading rebuilding-from-zero for spot-checking, anchor is the three-comp-plus-range first screen, risk is the flip-house comp the model can't tell apart from an untouched one, and keep out is holding back auto-publish and personalization until trust is earned.

And if you want to be sure it really works, try it somewhere elseSame five letters, a public library instead of a brokerage. A completely different field, and a completely different anchor.

Bramwell Ferry Library uses CatalogFathom, an AI tool that drafts catalog records, subject headings, call numbers, a short summary, for newly donated items before a librarian reviews them. Corrin Voss catalogs new donations three mornings a week.

Mapped onto SPARK: situation is Corrin manually researching an unfamiliar item's subject and call number, sometimes twenty minutes for anything outside the ordinary. Payoff is trading full manual research for a fast confirm on records the model is actually confident about. Anchor is CatalogFathom's first-ever record showing a confidence tag next to the subject heading and call number separately, since a model can nail one and miss the other. Risk is a genuinely rare item, a local zine or a small-press pamphlet, getting the same high-confidence look as a mass-market novel. Keep out is holding off on auto-shelving locations and public-facing summaries until the confidence tags have earned trust on ordinary donations first.

The library's real risk sits in the top-left corner: rare items where a confident-looking record is the most dangerous kind.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "show the comps and a range, not one number, so the first check feels easy," and stop.
Cost: there's no budget to build match-quality tagging this quarter. Say so honestly, and start with a single manual flag, any comp older than eighteen months gets a plain text warning, since a rough signal still beats a silent one.
The model gets better, for real: if PriceLantern's accuracy improves overall, that's still not a reason to drop the comps and range. A better average model still gets unusual properties wrong, and those are exactly the cases a first-time user needs to learn to catch.

Where people run it wrong.
They engineer the first success to be impressive instead of checkable, chasing a number that looks right rather than a habit that holds up.
They treat "the model was right the first time" as the whole win, without noticing what the user learned to do because it was right.
They build the confidence and personalization features before the basic number has earned any trust at all.

How to use it live. When someone asks what a good first success looks like, ask yourself one question first: what habit does this moment teach, not what number does it produce. Say the habit out loud before you say anything about accuracy.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0