Methodology: How Inktrue Calculates Fair Market Rent

The Inktrue Methodology

How does Inktrue build the most accurate rent analysis?

Every property has two kinds of value: what's true about the unit itself, and what's true about where it sits. Inktrue combines real attributes gathered directly from tenants and owners (the kind of detail no government or MLS database tracks) with genuine market data and precise location, then runs it all through a machine learning model trained to learn how every factor actually behaves, rather than assuming one fixed rule applies everywhere.

Property Attributes + Market Forces → Estimated Rent ± range
1The Honest Starting Point
value(feature) ≠ constant across markets
Why there's no single formula
Most valuation tools apply one fixed rule everywhere (a flat dollar value for an extra bedroom, applied the same way in every neighborhood). That's not how real markets work: the same bedroom is worth vastly different amounts depending on everything else about where it sits. Getting that right meant rethinking what "the math" should even look like (more on that below).

Part 1: Two ingredients

True market rent is calculated from two different kinds of information.

One side is about the unit itself: things that stay pretty fixed. The other is about the world around it: things that shift week to week. A good rent estimate has to account for both at the same time.

Fixed side

Property Attributes

Physical facts about the specific unit and building: verified, not guessed.

  • Size, layout, ceiling heightstructure
  • Kitchen & bathroom qualityfinishes
  • In-unit laundry, parking, gymamenities
  • Floor, view, walk scorelocation signal
Moving side

Market Forces

Conditions in the area right now: they can change even if nothing about the unit itself ever does.

  • Vacancy rate nearbysupply
  • How many renters are lookingdemand
  • Recently signed leases close bycomps
  • Season & local inflationtiming
+both sides are measured, then combined into one number
2What's Baked In, Today
vacancy rate + current rent level + exact coordinates
Real market conditions, not just the unit
Current vacancy rate and current local rent levels are genuine inputs to every estimate today (built directly into the number, not just shown alongside it). So is precise geographic location: exact coordinates, not just a ZIP code, since value shifts gradually across a neighborhood rather than jumping at a line drawn for mail delivery.

Part 2: The math, in plain terms

Think of the analysis as a dynamic process, not a fixed number.

Start from what similar places nearby actually go for, factoring in real current conditions (vacancy, prevailing rent levels). Then adjust for this unit's specific attributes and its exact location. The confidence range is honest about any missing data or unusual attributes.

Step 1
Start from real market data (current vacancy, current rent levels, exact location)
+
Step 2
Adjust for this unit's attributes (parking, laundry, view…)
=
Result
Estimated Fair Rent (with a +/- confidence range)
3The Real Differentiator
gradient boosting: hundreds of small, corrected stages
Why not just one clean formula?
A single formula assumes an extra bedroom is worth the same amount everywhere; it isn't. Inktrue uses gradient-boosted machine learning, the same class of technique used across modern tech and finance, to learn hundreds of these small, specific patterns automatically. Instead of one rule applied everywhere, the model builds its understanding in stages: an initial estimate, a check for where it was wrong, a small correction; repeated hundreds of times until the picture sharpens. Not one equation you could write on a napkin. Closer to how an experienced local appraiser actually thinks.
4What Actually Goes In
real market data + attributes no database tracks
Trained on real properties, not assumptions
The model is trained on real market data, combined with attributes gathered directly from tenants and owners (condition, finishes, amenities, and dozens of other details no external database has). Not every attribute has an equally confident weight yet: some are already well-established, others are actively gaining influence as more real reports flow through the system. Your report shows exactly which is which.

Part 3: Worked example

Watch it happen on one real apartment.

Say the average studio nearby rents for $1,500. Here's an illustrative example of how one unit's specific attributes move that number. Showing this breakdown in each report helps both tenants and landlords see if a unit is priced fairly (or where there's room to negotiate or improve).

Rent Estimate

Studio · 1 Bath · 4th Floor · Approximate Contribution
Nearby market baseline$1,500
In-unit laundry+$45
Reserved parking spot+$60
Updated kitchen+$35
No outdoor space−$20
Current vacancy conditions+$5
Estimated fair rent$1,625
confidence range: $1,575 – $1,675
6Constant Improvement, on Purpose
tested → found real errors → fixed → verified
Our model is constantly evolving
It started simpler, trained on a smaller set of active listings. Testing it against real, independently-verified rents (properties we already knew the true numbers for) showed it was underperforming, in some cases by a wide margin. Rather than ship that result, we rebuilt the approach: found and fixed real errors in the underlying data, tested a fundamentally different modeling technique, and verified every claimed improvement against real held-out data before trusting it. The model we use for every report is the result of that evolutionary process, and the same process runs continuously, not once.

Part 4: Why there's a range, not one number

The more you know, the tighter the range gets.

Two separate things shrink your uncertainty; and one can't make up for the other.

Only 35 of 126 attributes known
Wide rangeNarrow range
108 of 126 attributes known
Wide rangeNarrow range
5Why The Range Isn't Decorative
range reflects real, measured model uncertainty
An honest range, not a guess dressed up as one
The two gauges above show the real relationship: the range narrows as more of a property's attributes are known. It's built from the model's own tested accuracy against real, held-out data (not a decorative +/-). A single confident-looking number that's actually a coin flip is worse than useless; the range exists so you know exactly how much to trust it.

Part 5: A model that keeps improving

This isn't a formula we computed once and shipped.

The model is retrained on an ongoing basis as more real market data and real outcomes become available; today's numbers get better over time, not stuck at whatever they were on day one.

7The Actual Commitment
retrained on a regular, ongoing basis
Every report reflects the current version
Every report shows exactly when its underlying model was last updated (not a number frozen in time). As real reports and real outcomes accumulate, the model gets retrained and re-verified against held-out data, the same rigorous process described above, running continuously rather than once.

Takeaways

The short version.

1

Start with better property data.

Every attribute nudges the rent up or down from a starting comparable price; it's not a single formula pulled from thin air.

2

Real market conditions, actually baked in.

Current vacancy rate and current rent levels are genuine inputs to every estimate, not just context shown next to it.

3

Precise location, not a rough zone.

Exact coordinates, not just a ZIP code, so value reflects the specific block a unit sits on, not a citywide average.

4

Confidence range beats a guess.

An honest estimate says how sure it is, not just what the number is.

See this methodology run on a real address.

Free, independent, and ready in about ten minutes.

Get My Free Report