What’s actually holding up building permits in Seattle? I ran the data on 54k+ recent projects.

Hi all,

I recently finished a data project looking into the city’s building permit process and thought this community might find the results interesting.

I downloaded the public records for 54,389 building permits in Seattle dating from 2018 up to 2025. Using machine learning, I analyzed the data to see where things get bottlenecked, why projects get caught in correction loops, and if approval timelines can be predicted.

Here are the biggest takeaways from the data:

• I looked at what I call “multi-cycle risk” (projects that get sent back for multiple rounds of corrections). “Middle housing” is the worst: 75.6% of these permits require multiple correction cycles with a median review time of 181 days. On the contrary, single-family alterations/additions are the easiest to get through (only 31.8% need multiple cycles, with a median time of 76 days).

• Zoning and addressing have the highest volume of reviews, but they actually get processed quickly. The biggest slowdowns happen in three areas:

  • Drainage: 69.6% of these need corrections. For the slowest 10% of projects, it takes reviewers 40+ days just to reply.
  • Geotech (ECA and geo soils): 66% require corrections, also frequently hitting 40+ days for a response on the slower end.
  • Housing: 61% need corrections, and the bottom 10% of these take almost two full months (58 days) just to be reviewed.

• I sampled reviewer notes to find the most common triggers for corrections. The recurring themes are life safety codes, zoning/massing, trees/landscaping, critical areas/geotech, and structural design (which generates the longest comments, averaging 413 characters).

• Can we predict timelines? Pinpointing an exact approval date is practically impossible. However, I built a model that assigns projects into time-range “buckets.” It gets the correct time range within its top two guesses about 64% of the time. Even better, my model that flags “multi-cycle risk” before you even submit is highly accurate (89% ROC-AUC).

If you are curious about the methodology or want to run your own project parameters through the models, I put together a free interactive dashboard here: seattlepermit.vercel. app

Happy to answer any questions! Also curious to hear your experience with permitting in Seattle 🙂