DataFlik helps real estate investors and wholesalers find motivated sellers before properties hit the market, using AI-powered list stacking and predictive modeling to focus marketing on the homeowners most likely to sell.
DataFlik needed machine learning expertise to build a predictive model capable of processing millions of property data points to identify houses likely to sell before listing. Lacking an in-house ML team, the startup brought in NineTwoThree to augment their capabilities.
Built a custom ML model that predicts which properties will list for sale before they hit the market.
Combined AI-driven list stacking with predictive modeling to target homeowners most likely to sell off-market.
Built a mini model from sample data to prove the concept before DataFlik committed to purchasing the full dataset.
Continued as an embedded engineering partner, working alongside DataFlik daily to iterate on the model.
Building an effective predictive model meant proving it would work before DataFlik could justify the cost of the data behind it.
The real estate data needed to train an effective ML model was extremely expensive. Millions of US properties had roughly 1,700 data points, and the full dataset totaled 42 billion data points. DataFlik had to justify the purchase without building the full model, since building it required the data itself.
Real estate wholesalers spent significant time and money researching potential sellers to build mailing lists. They sent hundreds to thousands of letters with very low conversion rates. Traditional methods relied on human analysis of a small number of data points to create broad lists, resulting in wasted marketing spend and missed opportunities for properties that would sell but were not yet on the market.
DataFlik was a lean team without the resources to build an internal machine learning team. They needed experienced ML engineers who could interview domain experts, design the model architecture, process massive datasets efficiently, and deliver a working baseline within a reasonable timeline.
NineTwoThree embedded with DataFlik's team, starting with a domain expert interview to identify the most predictive data points, then building a mini model that de-risked the full data purchase.
We interviewed a real estate prospecting expert to identify the most predictive data points, then built a mini ML model using sample data from vendors. This proved the approach worked before committing to the full 42 billion data point purchase.
72%
of organizations need 100,000+ labeled items for production ML confidence
We trained the full model on 42 billion data points to predict house sales, deal discounts, and investor movements. Scaling required ~20 AWS services and a 10TB data migration with zero downtime.
We continued working with DataFlik to refine the model as the real estate market shifts. The system adapts to target the owners most likely to sell, improving performance each month.
Monthly reports were manual, with 10% of customers needing custom fields. We built a web app that automated report generation and added a skip trace portal for a new revenue stream.
We grew the team from 1 data scientist to 8 ML engineers, 20 developers, QA, and designers. We hired and trained replacements, set up SDLC processes, and helped the founders prepare to raise funding before handing the project fully in-house.
NineTwoThree's model predicts motivated sellers with 85% accuracy, generating prioritized lists faster and more reliably than a human expert doing the same research manually.
DataFlik's users saw an average of 8 times better ROI versus other known seller lists. The AI-powered targeting allowed real estate wholesalers to focus marketing dollars on properties most likely to convert, reducing wasted spend on unqualified leads.
In 2022, DataFlik achieved a growth rate of 634% in monthly recurring revenue. The startup planned to release several new products and grow their team considerably, powered by the success of their ML-driven real estate platform.
NineTwoThree worked alongside DataFlik's team daily, iterating on the model and building new products on top of their ML-driven real estate platform.
.svg)
.webp)
The algorithm accurately predicts about 60%–70% of home sales and improves every month. NineTwoThree's team stands out for their knowledge and experience. They're well-integrated into the internal team, communicating well through Slack, Zoom, Google Meet, and monday.com.
Build a new platform, or add AI capabilities to existing software, we deliver solutions that work in real operational environments.
Meet with founders Andrew Amann & Pavel Kirillov
Custom AI implementation roadmap
No sales pressure
