After decades of working with models for retail real estate, I've watched AI transform how we evaluate sites — and watched smart operators get burned by trusting the score over the story.
Let me say the quiet part loud: the best site selection tools today are genuinely impressive. They process hundreds of variables — mobility data, demographics, competitive density — in the time it used to take an analyst to build a trade area map. That's not hype. That's real.
But here's what decades of location modeling has taught me: the score is the beginning of the analysis, not the end of it.
The model sees the data. It doesn't see the strip center that can't lease up, the anchor tenant quietly negotiating an exit, or the difficult left turn with no light. I've seen high-scoring sites underperform in the short term due to road construction. I've seen low-scoring sites overperform because population growth was so rapid, no demographic projection had caught it.
The model works from a rearview mirror. Experience and judgment tell you when the road ahead looks different.
Where This Matters Most
This dynamic is especially pronounced in destination retail — optical, dental, specialty services. These categories don't behave like impulse retail. A strong co-tenancy signal for fast-casual can be actively misleading for an eye care provider. Trade area shape, residential density, and competitive voids matter in ways general-purpose models aren't calibrated to capture.
Pattern recognition fills that gap. Knowing this market's customers shop 30% farther than the model predicts. Recognizing that a landlord consistently misrepresents co-tenancy. Remembering that the last three times you opened near a certain anchor, lease comps spiked 20% within 18 months.
None of that lives in a dataset. It lives in the analyst who's been doing this for ten years and has earned the right to say "the model says yes, but I'm not sure."
The Right Framework
Use AI to screen the universe, eliminate obvious misses, and surface candidates you'd never find manually. Then apply the experience layer where it counts: ambiguous signals, contested markets, local dynamics no algorithm has touched yet.
The score tells you what the data saw. Experience tells you what the data missed.