By: Alex Klarfeld, Cofounder / Engineer @ Divvy Homes
The process of buying one home is difficult in itself — helping thousands of families buy their first home is formidable and creates an almost impossible scaling problem. Fortunately, we’re working on a solution.
In June of 2017, I helped co-found Divvy Homes, a company solving this problem. Our model is straightforward: we help renters become home owners by allowing them to build equity in their home as they rent. Our customers aren’t confined to a fixed inventory — they can Divvy any home on the market.
As an engineer, I often find myself explaining to my peers how software fits into this solution. I’ve generalized our problem space into three distinct buckets, each of which run deep: Data Intelligence, Operations, and Financial Engineering. I’ll give a high level overview of these buckets below.
To qualify for our program, customers apply online at www.divvyhomes.com and provide us with information about their credit and financial history. It’s a delicate balance trying to make our application process as painless as possible, while still having enough information to underwrite our customers fairly.
Adding even more complexity, we’re looking at our customers through a different lens than a traditional mortgage provider. Rather than assessing their ability to get a mortgage right now, we’re predicting their trajectory to get a mortgage in 3-5 years.
We get thousands of applications, and the sheer variety of data required to process each customer is staggering. We’re dealing with not only structured data from providers like Plaid, but semi-structured and unstructured data in the form of full credit reports, pay stubs, and bank statements. All of this information is provided to us by humans and is naturally prone to error. Blurry pay stubs, cut-off bank statements, and even random selfies make their way into our data pipeline. We need to clean and process all this data and reach a decision quickly.
The application is only the tip of the iceberg. Once a customer is accepted into our program and receives a budget, the home matching process begins. Since customers can choose from any home on the market, we underwrite each home as they shop. Underwriting an individual home requires even more data regarding the home’s projected appreciation, fair market rent, and list price to ensure our customer will be able to comfortably afford the monthly Divvy payments. This pricing calculation needs to be perfect, as it directly impacts the financial success of our customers.
The home shopping process quickly blends in to the home buying process once a customer finds a home they love. At this point, we start getting additional data about the home through our inspection team. A thorough inspection completed without sophisticated software takes around 4 hours per home. Most of this time is spent manually aggregating information and generating reports.
We’ve taken the first steps to build out custom applications for our inspectors to help them breeze through an inspection without sacrificing on quality, while automatically organizing the inspection data and photos in a consumable format. While most of our inspection reviews are still software-assisted, we are quickly automating parts of the process through computer vision and machine learning.
Inspections are just one part of the home buying process. An average home purchase requires around 10–20 unstructured documents to close and requires coordination with many different parties including our partner agents, seller’s agents, and legal teams. There are a handful of interesting tech companies helping streamline parts of this process, like Spruce’s title insurance API, but on the whole this entire workflow is fertile ground for software.
As customers settle into their homes, they are quietly on-boarded into Divvy’s ledger product. The finances of each individual Divvy household are complicated enough to warrant their own accounting software, let alone tens of thousands at scale. Rental payments, equity builds, and maintenance costs must be continuously categorized and accounted as money flows through the system. Additionally, as large financial institutions interact with smaller FinTech companies such as Divvy, airtight accounting is a necessity. Transactions must be able to be replayed, reported on, and audited. Unfortunately, there is not an existing software product on the market that can manage the needs of a company like Divvy at scale. In lieu of hiring an army of accountants, we require a home grown Ledger product.
Ledger systems are not unique to Divvy, nor solely contained to blockchains. Many financial institutions and FinTech companies, like Affirm, have large engineering teams dedicated to the complexity of automatic accounting.
This article only scratched the surface of the breadth of engineering problems that Divvy is working to solve. Buying one home for one family is difficult — buying thousands of homes for thousands of families becomes really interesting.
We think these problems are exciting, and if our mission resonates with you, reach out — we’re hiring! Also, if you like this blog post, and want to hear more from our engineering team, please sign-up for our newsletter here.