For more than two decades, real estate technology companies have built consumer-facing search tools and agent platforms. Yet a core problem persists: the data powering these systems is often unreliable, inconsistent, and fragmented across states and counties.
Andy Taylor, founder and CEO of RetroRate and former VP of Product at Redfin, says the challenge of building accurate mortgage data systems is far more complicated than he anticipated. The consequences of flawed data are immediate and lasting.
“When I first started out, I thought, this is going to be an easy process,” Taylor recalls. “I thought there’d be some off-the-shelf data I could just plug in and go. But you really have to clean it up pretty hard to make it useful and interesting.”
The Data Fidelity Problem
Taylor says real estate data comes from a “patchwork quilt” of APIs and data sources, each varying in quality. “Some have really great fidelities. Some have really poor fidelities. And it also depends on which states you’re working in – some are great in certain states, and some are not so good in others,” he explains.
The problem begins with how real estate information passes through multiple intermediaries before reaching end platforms. “Like all real estate data, it’s been touched by 1,000 agents and 1,000 county clerks,” Taylor notes. Each transfer is an opportunity for errors, which compound as data moves through the system.
Building national platforms intensifies the challenge. Some jurisdictions maintain high-quality, accessible records; others have poor data quality and limited access. “You really have to clean it up pretty hard to make it useful and interesting,” Taylor emphasizes.
For RetroRate, which identifies assumable mortgages and their interest rates, the stakes are especially high. Real estate professionals spot inaccuracies instantly. “If you don’t do that, sure, you can get something up and running,” Taylor says. “But the moment an agent looks at it and says, ‘Well, I know for a fact that this home has a 4.25% and you’re showing a 5.875%,’ they’re never going to trust you again.”
The Waterfall Model Solution
To address data quality, Taylor relies on multiple redundant data sources for every criterion. Rather than trusting a single source, RetroRate uses what Taylor calls a “waterfall model.”
“We have a waterfall model that shows, based on all the sources we’re looking at, here are the top sources for each, and here’s our confidence score on how good the data is on that particular home,” Taylor explains.
No single data source is complete or accurate across all markets. Instead, the system identifies the best available data for each property and data point, then assigns a confidence score to help users gauge reliability.
This process requires ongoing improvement. “We’re constantly looking at it and saying, all right, what additional data can we source to fill those gaps and make this data even better?” Taylor says.
Why Accuracy Matters More Than Speed
Many proptech startups are tempted to launch quickly with imperfect data, planning to improve quality later. Taylor argues this approach is fatal when serving professionals who know the ground truth.
“It really has to, right from the get-go, be clean and accurate,” Taylor insists. “There’s a lot of work. It’s a lot harder than I thought it was going to be, for sure.”
The difference between consumer and professional users drives this need for accuracy. Homebuyers may not know if a listed interest rate is correct. Agents working with specific properties know the actual loan terms and spot errors immediately.
Once trust is broken with professional users, Taylor argues it cannot be rebuilt. The agent who sees incorrect data will not return, regardless of later improvements.
The API-First Architecture
RetroRate’s answer is to build everything around an API that delivers property-level mortgage information. “Everything we built is built around an API, so there’s API access to all the loan information and home information,” Taylor explains.
The API provides both basic information, like interest rates, and comparative analysis. “That API includes the basic information around what’s the rate on the assumable loan, but also, how much better is that loan than the prime rate at that point?” Taylor notes.
This architecture allows the data to be used in multiple ways—integrated into MLS systems, accessed by brokerages, or used in other applications—while maintaining a single source of truth that can be continuously improved.
Broader Industry Implications
Taylor’s experience highlights a persistent infrastructure problem in real estate: data quality remains unsolved despite years of proptech investment. The industry still lacks standardized data formats and validation protocols that would enable reliable information from source to consumer.
“It’s a lot harder than I thought it was going to be,” Taylor admits, reflecting on the data infrastructure challenge.
Whether the industry will develop better data standards or individual companies must keep building proprietary data cleaning systems remains uncertain. But Taylor’s experience indicates that proptech firms prioritizing data quality from the start, even at the cost of slower product launches, build more trustworthy and defensible platforms than those that focus on speed at the expense of accuracy.
