Scaling Real Estate Underwriting with AI

HatchWorks AI designed and delivered a machine learning underwriting system for Grandview Homes, a Real Estate business in the US Midwest. The system augments Grandview's existing human-owned underwriting process with three machine learning models: ARV (After Repair Value), Rehab Cost, and Time to Sell, surfaced directly inside Salesforce.

Drawing on Grandview historical data and third-party data, the system produces model-driven predictions that feed Maximum Allowable Offer (MAO) calculations. It replaces a manual deterministic benchmark as the primary analytical input to offer pricing decisions.

Grandview Homes acquires residential properties, renovates them, and resells them. Because deal margin is heavily influenced by the initial purchase price, underwriting decisions directly affect how effectively Grandview can deploy its capital.

As the company looked to scale, Grandview wanted to increase the number of opportunities its team could evaluate without adding underwriting cost at the same rate. It also wanted to improve the quality and consistency of those decisions by making better use of its proprietary historical data and current market data.

The existing underwriting process relied on a deterministic model to estimate ARV, rehab cost, and time to sell. While useful as a benchmark, the model could not effectively learn from Grandview’s historical transactions or adapt as market conditions changed.

The system is live in production and accessible directly within the Salesforce workflow Grandview’s underwriters already use.

Independent holdout evaluation showed the new models outperforming Grandview’s existing benchmark on two of the three core predictions, with near parity on the third. ARV prediction error was reduced by 31%, while Time to Sell prediction error fell by 69%.

Because these predictions feed Grandview’s Maximum Allowable Offer calculations, the system gives underwriters stronger analytical inputs when determining what to offer for a property. It creates the foundation for Grandview to evaluate opportunities more consistently and scale underwriting capacity without relying on a proportional increase in overhead.

Just as importantly, the system was designed to improve over time. Automated ingestion, monthly retraining, evaluation, and drift monitoring allow new transaction and market data to inform future model versions, helping Grandview build on its proprietary data advantage as the business grows.

Grandview has introduced the machine learning predictions alongside its existing deterministic model, allowing underwriters to use both when determining an appropriate offer amount. This approach gives the team an opportunity to understand where the models are strongest and build trust through real-world use.

Underwriters also have access to historical predictions and monthly evaluation metrics. Combining those results with their own observations will help Grandview identify reliable patterns, improve future features, and continue refining how AI informs underwriting decisions.

As Grandview captures more data from future purchases, renovations, and sales, that information can feed back into the system and improve the signals available to its models. Over time, this creates a compounding advantage: Grandview can use what it learns from each investment to make more informed decisions about where to deploy its limited capital next.

“Offer pricing is where our margin lives. Having model-driven predictions inside Salesforce gives our underwriters a sharper, more consistent starting point than the benchmark we relied on before.”