Zillow Cut 500 on an 18% Growth Quarter. The Zestimate ML Pool Is 43.

On Aug 4, 2026, Zillow CEO Jeremy Wacksman announced the company's second and largest layoff of the year: 500+ people, roughly 7% of the workforce, in the same quarter revenue grew 18% year-over-year. If you source ML talent, the interesting number is not 500. It is 43, the approximate size of the Zestimate valuation ML subset that just came loose from the only company in the world that runs hedonic pricing models against every US home on a daily refresh.

What actually happened on Aug 4

Zillow cut 500+ employees, about 7.1% of its ~7,058-person workforce, in a "positioning for the path ahead" reduction that pointedly denied AI as the driver. Wacksman told staff the cuts hit "nearly all levels" and direct managers only learned alongside their reports, which is unusual and worth flagging when you reach out to candidates who are still processing it.

Three details reframe how you should source this pool:

• It is Zillow's second cut of 2026. January took 200. August took 500+. Cumulative 2026 reductions sit around 700.
• Only 91 of 500 were in Washington state, per WARN filings, meaning ~82% of cuts fell on remote and distributed staff. The "Seattle-first" sourcing default will miss most of them.
• Senior roles bore the brunt, per GeekWire's follow-up. Expect L6/L7-equivalent comp anchors, not junior pricing.

This is not a distressed-company diaspora. Q1 revenue was $708M (up 18% while the broader residential real-estate industry grew 2%), and net income climbed to $46M from $8M a year earlier. In May, Wacksman told investors, "We are rapidly becoming an AI-native company." The layoff dropped roughly 90 days later. Read that sequence however you want, but do not sell candidates the "you got laid off from a failing company" narrative. They will bounce.

Why "ex-Zillow" on LinkedIn is the wrong query

The useful subset of ex-Zillow ML talent is roughly two orders of magnitude smaller than an "ex-Zillow" LinkedIn search returns, and the wrong query wastes a full sourcing week.

In Refolk's index of professional profiles, filtering on eight ML/DS-family titles (MLE, Senior MLE, Staff/Principal MLE, Applied Scientist, Data Scientist, Research Scientist, ML Infra, Research Engineer) crossed with "Zillow" in current or past employment returns:

405

Of those 405, only a fraction ever shipped against the Zestimate model itself. The rest worked on recommendations, semantic listing search, floor-plan generation, textual home insights, and Premier Agent match optimization - five distinct ML sub-specialties documented on Zillow's own tech blog. All useful, none the differentiated hire. The Zestimate subset is closer to 43.

That is the whole game: separating five ML sub-specialties inside one company from the one sub-specialty that trades at a premium in the poaching market. LinkedIn's filter grammar cannot express "worked on the Zestimate valuation stack specifically." Boolean strings on title or headline fields miss most of it because almost nobody puts the product name in their headline. This is the exact gap Refolk closes: describe the person in plain English ("ML engineers who shipped against the Zestimate valuation model at Zillow, past or present, Seattle or remote") and get a ranked shortlist instead of a keyword guess.

The dataset, in one table

Here is the full sizing picture, with sources, so you can defend it to a skeptical hiring manager.

The 18.2% WA share is the number most sourcers get wrong. Zillow went "Cloud HQ" all-remote in 2021, so the engineering org spread nationally years before the layoff. If your ATS auto-tags "Seattle" on every ex-Zillow profile, fix that first.

The Zestimate skill set, and why it is non-substitutable

Zestimate ML is hedonic pricing models running on nationwide real-estate feature data at daily-refresh cadence, and the skill set exists at maybe four companies globally. That scarcity is why 43 profiles is a lot, not a little.

Hedonic pricing decomposes a good's price into implicit prices of its attributes: for a home, that means square footage, lot size, school district, comparable-sale distance, and hundreds of other features per property. Zillow ships this against every US home with a daily refresh. That combination (national coverage, feature engineering against MLS and tax records and imagery, calibrated confidence intervals, latency budgets) does not exist elsewhere at the same scale.

The four companies that run substitutable stacks:

1. Redfin (Redfin Estimate). Direct competitor, similar hedonic approach, smaller footprint.
2. Opendoor (offer pricing). Different objective (inventory risk, not consumer-facing estimate) but shares the feature stack.
3. HouseCanary (institutional AVM). B2B, deeper commercial focus.
4. CoreLogic (institutional AVM). Legacy incumbent, mortgage industry.

If you are hiring for a proptech Series B or a mortgage-tech incumbent, these are the four names in the "worked at" filter. If you are hiring for a marketplace ML role that just needs someone who has shipped calibrated pricing models to millions of users, the Zestimate 43 is the pool with the best combination of scale exposure and availability window.

The org lineage that tells you who is worth chasing

Every Zestimate engineer traces back to Stan Humphries, Zillow Group's chief analytics officer and the model's original creator. Following that lineage is the fastest way to rank a shortlist.

Three signal layers, in order of predictive power:

• Reported into the Zestimate/AI org. Job history mentions "valuation," "Zestimate," "home value," or "AVM." Strongest signal, hardest to extract from LinkedIn without reading every profile.
• Zillow Prize participation. Jordan Meyer's three-person team beat 3,800 teams from 91 countries to win the $1M Zillow Prize, and Zillow hired from the leaderboard (Meyer himself was one of those hires). Kaggle profiles list the placement publicly. The top finishers are pre-vetted valuation-ML talent, and many never worked at Zillow - that is the shadow pool.
• Bylines on Zillow's tech blog. Posts on model serving, feature stores, and floor-plan generation credit authors by name. Cross-reference with LinkedIn.

The Zestimate poaching window opens once per decade. This is that window.

How to structure the outreach

Lead with the AI-margin framing, not the layoff. Senior engineers cut from a company posting 18% revenue growth and net income up from $8M to $46M year-over-year do not want condolences. They want to know you understand what happened.

A message shape that has worked in the first 72 hours after similar cuts:

1. Line 1: name what happened accurately. "Saw the Aug 4 cut. Rough sequence given Wacksman's May 'AI-native' comment."
2. Line 2: name the specific work. Reference a Zestimate paper, a tech-blog byline, or a Zillow Prize placement. Not "your impressive background."
3. Line 3: name the parallel problem you are hiring for. Marketplace pricing, inventory valuation, dynamic supply matching. Not "an exciting opportunity."
4. Line 4: name comp range or ask about it. Senior Zillow anchors are high. If your budget is junior, do not send this message.

Recruiters running this at volume will hit the same friction: InMail caps and the impossibility of expressing "Zestimate specifically" in LinkedIn's search grammar. Refolk's plain-English queries let you describe the person the way you would describe them to a coworker ("ex-Zillow ML engineer who worked on the valuation stack, published on the tech blog, based anywhere in North America"), and the ranking uses signals LinkedIn cannot see, including GitHub activity, Kaggle placements, and paper authorship.

The trap: Zillow is still hiring MLEs

Zillow's Agentic AI team is still posting Principal MLE roles during the layoff, per current listings on The Muse. That confirms the "shift the ML org from valuation to GenAI" thesis and it should change how you talk to candidates.

Some of the Zestimate engineers were not laid off. They were offered internal transfers to Agentic AI, and some accepted. The ones who left, whether pushed or having declined the transfer, generally fall into two camps:

• Declined the pivot. They wanted to keep doing valuation ML, not switch to LLM-flavored agentic workflows. Best hires for proptech and mortgage-tech roles.
• Wanted the pivot but were cut anyway. Best hires for GenAI-native marketplace roles, because they already made the mental jump.

Ask which camp on the first call. It sorts the pool in one question.

What to do this week

Move fast on Zestimate. Move slower on the rest. The 43-person subset will be signed inside 30 days.

• Days 1 to 3: Pull the Zestimate 43. Cross-reference with the Zillow Prize leaderboard and tech-blog bylines.
• Days 4 to 10: First outreach with AI-margin framing. Skip the condolence language.
• Days 11 to 21: Second wave to the broader 405 ML/DS pool for adjacent roles (recommendations, semantic search, floor-plan CV).
• Days 22 to 60: Long-cycle nurture of the Zillow Prize shadow pool - top Kaggle finishers who never took a Zillow offer.

The full 500-person layoff includes PMs, designers, agents, and ops. Those are separate sourcing exercises with separate playbooks. Do not mix them into the ML pipeline.

FAQ

How many ex-Zillow ML engineers are actually on the market after Aug 4?

Roughly 43 with Zestimate/valuation experience, drawn from a broader 405-profile ex-Zillow ML/DS pool in Refolk's index. The 43 figure is derived from Zillow's total ML headcount, filtered to the subset that ever shipped against the valuation stack specifically, and adjusted for the senior-heavy layoff mix reported by GeekWire.

Is this a distressed-company situation or something else?

Something else. Zillow's Q1 revenue grew 18% to $708M while the residential real-estate industry grew 2%, and net income rose to $46M from $8M year-over-year. The layoff came about 90 days after Wacksman called Zillow "AI-native" on the earnings call. Frame outreach around AI-margin repositioning, not company distress, or your reply rate will crater.

Why not just use LinkedIn Recruiter with "ex-Zillow" as the filter?

Because the useful subset is 43 and LinkedIn's filter grammar cannot express "worked on the Zestimate valuation model specifically" - almost nobody puts product names in their headline. You have to read profiles individually or use a sourcing tool that ranks on signals LinkedIn cannot see. Refolk indexes GitHub commits, Kaggle placements, and tech-blog authorship, which is where Zestimate work actually shows up.

Which companies are most likely to poach the Zestimate 43 first?

Redfin, Opendoor, HouseCanary, and CoreLogic run the four substitutable AVM stacks globally, and their recruiters knew the layoff was coming the same day you did. Expect competitive offers inside two weeks. If you are outside proptech but hiring for marketplace pricing or dynamic supply matching, your pitch has to translate the skill set explicitly, because these candidates will otherwise default to the four names they already know.