If you run SiteAI twice on the same parcel and get different yield counts, that's expected behavior — not a bug. SiteAI is a generative algorithm, which means it optimizes lot placement and road layout based on your inputs, but the specific arrangement it produces can vary slightly between runs even when the inputs look the same. Here's what drives the differences, and how to get a stable number you can use for underwriting.
Why Yield Can Vary Between Runs
The most common reasons yield changes between runs:
You changed an input. Even a small change — adjusting minimum lot width by a foot, adding a custom avoidance area, changing the road entry point — can meaningfully affect how many lots fit. Before assuming the algorithm changed, check whether your inputs match your previous run exactly.
Algorithmic variation. SiteAI's generative algorithm explores many possible lot arrangements and selects a high-performing one. On complex sites with irregular boundaries or environmental constraints, different runs may find slightly different arrangements that produce yields within a small range of each other. A difference of 1-3 lots on a site is normal.
Environmental data updated. Wetland boundaries, FEMA flood zone maps, and other environmental data sources are periodically updated. If an environmental boundary shifted since your last run, your buildable area may have changed.
What to Do When Yield Changes Significantly
A difference of 1-3 lots is within normal variation and doesn't require investigation. If your yield count changes by more than that between runs with identical inputs, check:
Whether your avoidance areas are the same as the previous run — custom avoidance polygons don't persist between sessions by default
Whether your road entry points are in the same location
Whether the parcel boundary loaded correctly — on some complex parcels, SiteAI may load a slightly different version of the boundary if the parcel record was updated
Getting a Stable Number for Underwriting
For a yield count you can take to an investment committee or use in an offer, run SiteAI two or three times with the same inputs and use the average — or the conservative end of the range. This gives you a defensible number that accounts for the natural variation in the algorithm. Save the plan that produced the most representative output and note the yield range in your project record.
SiteAI vs. Engineering
SiteAI produces a planning-level estimate, not a final engineering design. The yield count is accurate enough to underwrite a deal and make an offer, but the final site plan prepared by a civil engineer will reflect actual grading, utility routing, and regulatory submittals that may shift the number. Think of SiteAI as your pre-offer feasibility tool and engineering as your post-contract verification.
SiteAI is available on Expert and Enterprise plans. Not on Expert or Enterprise yet? Talk to your account team.