Zillow Offers: pricing forecasts and business risk
Zillow announced the wind-down of Zillow Offers in November 2021, citing unpredictable home-price forecasts and excessive balance-sheet volatility. This historical example concerns a capital-intensive buying business. It does not establish that one AI model caused every loss. Zillow is not presented as Paul Okhrem’s client.
What Zillow reported
| Measure | Status in the disclosure | Limit |
|---|---|---|
| About $304 million Q3 inventory write-down | Reported | Not an isolated estimate of loss attributable to AI. |
| $240–265 million additional Q4 losses | Expected at that date | A forecast, not a realized amount in this release. |
| About 25% workforce reduction | Planned during wind-down | A company-wide restructuring measure. |
Original company disclosure, reviewed 29 October 2026.
Forecasts affected real inventory exposure
Zillow’s disclosure links the decision to forecasting uncertainty and earnings volatility. It also identifies renovation and resale capacity constraints. The source does not isolate algorithm design from those operating and market factors.
Review the decision and its downside
Paul Okhrem’s analysis: a forecasting system needs a business limit as well as an accuracy target. Before it influences spending, define who can stop it and what evidence triggers that decision.
- How much capital may the process commit?
- What happens when forecast errors persist?
- Who checks the link between a model output and an approved action?
These are useful questions in AI due diligence.
Limits of the example
This is an illustrative governance interpretation, not an independent forensic finding about Zillow’s software. A historical property-buying outcome does not predict the economics of a different AI workflow.
Zillow Offers: common questions
Can this disclosure establish the model’s sole causal role?
No. It reports a business decision and financial effects. Separating model, market and operating causes would require evidence beyond this disclosure.
How should a board use the example?
Review decision authority, capital exposure and stop conditions. Avoid using one historical example as a universal AI failure rate.
Change log
- : Distinguished the reported 2021 write-down from the following-quarter forecast and removed unsupported causal claims.
Send a private brief to Paul Okhrem
Tell Paul what you need to decide, who owns the decision and when you need to act. Review his background and published terms.