There's a better way to think about a commercial real estate market than rows in a spreadsheet. Think of it the way a physicist would think of a pond. Every event in a market — a utility upgrade, a loan maturing, a permit filed — is a stone dropped in the water. The ripples radiate outward, lose energy with distance, and, if you're patient enough, you watch them arrive at every parcel in the neighborhood.
That's the model behind AI CRE parcel scoring. Not a black box. Not a magic 0-to-100 number that falls out of the sky. Something closer to physics.
§1 · The Premise
Changes Act Like Gravity Waves
Every time something happens in a market — an owner transfers a building, a city approves a utility extension, a bank originates a loan — the effect propagates. Nearby parcels feel it first and most. Distant parcels feel it later and less. Eventually, if enough time passes without reinforcement, the wave fades.
That's it. That's the whole model on a napkin. Changes radiate. Nearby matters more than far. Time erodes effect.
AI CRE parcel scoring takes this idea and makes it concrete. Every event is an emitter. Every parcel is a receiver. The algorithm measures, for every parcel at every moment, the sum of every wave that has arrived — weighted by how far away the emitter was and how long ago it fired.
One wave by itself is rarely interesting. The math gets interesting when multiple waves from different sources arrive at the same parcel inside a short window.
§2 · Convergence
When Waves Converge, They Create Signal
One wave arriving at a parcel is, by itself, not enough to change a decision. A single permit filed three blocks away might mean nothing. A single loan maturity across the street might be a one-off.
What matters is when multiple waves, from different sources, arrive at the same parcel inside a short window. That's convergence. That's the moment a parcel goes from noise to signal.
The loudest parcels on the Gravity Map are never the ones with a single big event. They're the ones where three or four or twelve different waves are all stacking on top of each other — a utility extension and a permit cluster and a distress opportunity and a demographic tailwind and a capital inflow — all within 0.3 miles, all within the last 12 months.
A single wave is a rumor. Convergent waves are a story.
Convergence · Three Emitters, One Parcel
Auto-loops
§4 · The Math (in plain English)
Delta, Distance, and the Compression
Once you accept that changes behave like waves, the rest of the model is bookkeeping. Careful, disciplined bookkeeping — but bookkeeping.
Connecting data points across time
A parcel isn't a static number. It's a trajectory. On January 1, it has one state — pricing, ownership, zoning, surrounding activity. On April 1, it has a different state. The question AI keeps asking is: how did we get from the first state to the second?
That's the first concept: delta. The distance, in whatever units matter, between two observations of the same parcel at two different moments.
Comparing deltas across parcels
One parcel's delta on its own is information. Ten thousand parcels' deltas, side by side, is a market. You can see which parcels are accelerating. You can see which are stalling. You can see which ones are surprising — moving faster than the submarket average, or moving in the opposite direction.
The scoring engine is, at the simplest level, a machine that computes deltas and compares them.
Compressing the 3D spreadsheet
Here's the part that matters for anyone who's ever opened a CoStar export. The underlying data is three-dimensional: parcels × metrics × time. Oklahoma County has 504,000+ parcels, Signal Intelligence tracks 526 metrics, and the time dimension goes back 25 years. Multiply those out. That's a spreadsheet no human opens twice.
The job of scoring is to compress that cube into a single filter. One signed number from -100 to +100, per parcel, that tells you the same thing the cube does — where this parcel sits in the full context of every other parcel, every other metric, every other moment.
Why One Number Is Useful
A broker can't scan 504,000 parcels on 526 dimensions. A broker can absolutely sort a list by one column. Compression isn't dumbing down — it's how you make 526-dimensional reality fit into a workflow.
§5 · Theta
Theta: The Decay of Data Over Time
Not all data ages equally. A permit filed last week is a live signal. The same permit filed in 2014 is historical context. The AI has to weigh them differently — otherwise yesterday's events get drowned out by a decade of noise.
The variable for that weighting is theta. It's borrowed from options pricing, where it measures the rate at which an option loses value as expiration approaches. In AI parcel scoring, theta is the rate at which a data point loses influence as time passes.
A high-theta signal decays fast: yesterday's permit is huge, last month's is strong, last year's is a whisper. A low-theta signal decays slowly: a 20-year transaction history is almost as informative today as it was five years ago.
Drag the slider to see how a signal's weight changes with different half-life assumptions. A permit signal might use a 3-month half-life. A zoning change might use 10 years.
Here's the subtle part: theta doesn't mean ignore the old data. It means weight the new data more heavily while still using the old data to tell the trajectory. You can't understand why a parcel is at $175/SF today without knowing it was at $75/SF two years ago. The old data explains how we got here even when the new data dominates what we do next.
That distinction is the difference between a scoring engine that forecasts and one that just reports.
The best opportunities in a market don't announce themselves. They're the ones where the math changed before the price did.