Executive Summary
| Metric | Before Geofi | With Geofi |
|---|---|---|
| Properties screened / analyst / day | 30 | 400+ |
| Post-close rent variance vs. underwrite | ±8% | ±2.4% |
| Underwriting confidence intervals | Baseline | 60% narrower |
Institutional SFR returns are ultimately determined by selection — which properties you buy. Today, selection is constrained by analyst bandwidth: most firms can deeply underwrite only a fraction of their daily buy box.
Geofi removes that constraint with two tools:
- Forecast — tract-level rent growth predictions up to six months forward, with calibrated confidence intervals.
- Lens — automated underwriting for every listing in your target markets overnight. Full DCF, comps, and quantified uncertainty.
In a 90-day pilot, a top-20 operator increased analyst throughput from ~30 to 400+ properties screened per day. Post-close rent variance tightened from ±8% to ±2.4%. Underwriting confidence intervals narrowed by 60%.
The outcome: faster deployment, higher conviction, better properties. Same team.
The Problem
You're buying 150 homes this year across the Atlanta metro. Your returns will be determined by which 150.
Operations, financing, and hold strategy all matter — but they're downstream of selection. And selection is currently constrained by analyst bandwidth: perhaps 30 deeply underwritten properties per day against a buy box generating 10× that.
The best deals aren't being found. The best deals that your team has time to look at are being found. There's a difference, and it compounds.
Geofi: Forecast + Lens
We built two tools that work together:
- Forecast models rent growth at the Census-tract level—roughly 1,200 to 8,000 households per tract—updated weekly. The output is a distribution: you see both the prediction and the confidence interval around it.
- Lens runs automated underwriting on every listing in your target markets, refreshed daily. Each property gets a cap-rate estimate, uncertainty bands, and a pass/fail flag against your IC criteria.
Forecast feeds Lens. A property in a tract with +6% predicted growth and tight confidence is underwritten differently than the same property in an uncertain market. That's how selection improves: you're no longer choosing from the 30 deals you had time to model. You're choosing from the 400 that actually fit.
Forecast
Metro-level rent forecasts are necessary but insufficient. Selection within markets is key: a forecast like “Atlanta rent growth is +3.2%” doesn’t help investors decide which neighborhoods deserve capital.
Forecast operates at the Census-tract level, updated weekly. Below, we show rent growth for five tracts within the Atlanta MSA. Same metro, same macro environment—but the trajectories and confidence bands diverge sharply. This is the variance that determines which homes you buy and which homes to avoid.
The model captures labor demand, supply pipeline, affordability stress, and more—and expresses uncertainty honestly. What you see here is historical. With customers, we forecast years forward across every U.S. metro.
📊 Market Selection:
- Select multiple markets for comparative analysis
- Color-coded visualization for easy identification
🎯 Confidence Analysis:
- Toggle between 50% and 97.5% HDI bounds
- Visualize prediction uncertainty ranges
📈 Interactive Features:
- Hover for detailed data points and statistics
- Compare predicted vs. observed values
Forecasting Highlights
- Clayton County, Atlanta (Tract 13063-0089): +8.7% forecast, moderate bands. The standout performer. The model is heavily weighting 10-year employment growth and workforce participation—suggesting structural economic improvement, not a speculative spike. Rent growth and population trends also contribute. The 2024 observed value of 10.7% validated the model's bullish stance, and momentum continues into 2025.
- South Forsyth, Atlanta (Tract 13121-0234): +5.7% forecast, tight bands. High conviction. The model keys on rent growth, workforce commuting patterns, and steady population/household formation. Classic suburban growth story—demand outpacing supply in a high-amenity location. The tight bands reflect consistent, predictable fundamentals.
- Gwinnett industrial corridor, Atlanta (Tract 13135-0156): +4.7% forecast, moderate bands. Workforce dynamics dominate here—the model is putting significant weight on worker commute patterns (3-year growth) alongside rent appreciation and employment trends. The industrial-to-residential transition creates upside, but mixed signals warrant measured optimism rather than aggressive positioning.
- East Cobb, Atlanta (Tract 13067-0118): +3.5% forecast, moderate bands. Boring and predictable—exactly what yield-focused capital wants. The model sees unemployment improvement and stable population as the key drivers. Mature housing stock (median year structure built is a top feature) means limited new supply competition. Steady mid-single-digit returns with minimal downside risk.
- West End, Atlanta (Tract 13089-0412) : +2.9% forecast, moderate bands. The weakest forecast despite strong employment signals. The model is heavily weighting workforce participation and population, but the 10-year unemployment trend and household formation suggest the gentrification thesis has decelerated from the 6%+ observed in 2023–2024. The model won't overfit to recent appreciation—price accordingly.
Five tracts. One metro. Five completely different underwriting conversations.
Key Driver Evolution
Interpreting model predictions is just as crucial as forecasting accuracy. Our feature-importance explorer reveals how individual drivers shape price and rent changes year-over-year within each market.
Interpretability Highlights
By tracing driver strength year-over-year, investors can:
- Spot regime shifts: Detect when economic or demographic factors begin to dominate valuations.
- Validate intuition: Confirm that anticipated catalysts — e.g. rate cuts, zoning changes, employer relocations — manifest in the data.
- Manage risk: Reduce exposure to markets where prediction drivers are unstable or poorly understood.
- Prioritize diligence: Allocate research resources to drivers showing sudden importance spikes.
Lens
Lens was designed to answer the question: where is the best place to put capital?
With data at the national, regional, market, and hyper-local level, Lens trains an AI agent on your buy-box and preferences, paper-trades 100,000+ simulated portfolios, and surfaces personalized buy/hold/sell recommendations across the full market history. When ready, the AI agent goes live and properties in the active deal feed are screened, scored, and ranked to your investment preferences, maximizng risk-adjusted return.
In practice, Lens runs the same DCF math your analysts run—but across your entire buy box, overnight. The difference is coverage and consistency.
- Rent estimate from block-level comps—because a ZIP code in Atlanta can span tracts with 5 percentage points of forecast variance.
- CapEx model based on age, condition, and local cost indices.
- Rent trajectory pulled directly from Forecast—tract-specific, not metro-average.
- Year-1 cap rate with uncertainty bands.
- Pass/fail check against your IC criteria.
Each property is evaluated across 1,700+ variables and ranked to align with your buy-box and risk preferences.
Analyst Throughput → Capital Deployed
In a traditional workflow, an analyst can carefully underwrite 25–30 properties per day. With Lens, the entire market is scored overnight.
The goal isn't more analysis—it's better decisions. Lens automates screening and modeling so analysts can focus on what actually requires expertise: evaluating neighborhood trajectory, assessing renovation complexity, making the call on edge cases. The machine does the exhaustive work; the analyst focuses on the decisions that move capital.
Current workflow
An analyst reviews ~200 MLS listings, selects 30 for deeper modeling, spends the day in spreadsheets, and produces five recommendations. IC approves roughly one per week—not due to lack of opportunity, but because building conviction under time constraints is difficult.
- Recommendations/day~5
- Approvals/week~1
- Deals/month~4
- Deployed / analyst / month~$1.4M
With Lens
Each morning, the analyst starts with 400 properties already scored and ranked. Time shifts from building models to reviewing them—and because Lens surfaces deals with tighter uncertainty bands, IC moves faster.
- Deals reviewed deeply/day~20
- Approvals/week~5
- Deals/month~20
- Deployed / analyst / month~$7.0M
What actually changed
The bottleneck shifted. Previously, analysts spent ~80% of their time generating analysis and ~20% exercising judgment. With Lens, that ratio inverts—analysis is automated, judgment is amplified. And because each recommendation comes with calibrated confidence intervals, IC debates assumptions less and approves more.
- 5× capital deployed per analyst
- Same headcount
- Same buy box
- Higher conviction per decision
LP takeaway: A $500M fund targeting 12-month deployment typically needs 8–10 analysts at current throughput. With Lens, the same deployment pace requires 2. A fund that previously took 18 months to fully deploy can now close in under 4—filling allocations while conviction is high and competition is thin.
The Daily Queue
This is what the workflow inversion looks like in practice. Each morning, your analyst opens a ranked queue of properties already scored overnight. The question shifts from "what should I look at?" to "which of these should we pursue?"
| Price | Bd/Ba | Sq Ft | Year | Cap | Yield | Score | Sharpe | Verdict |
|---|---|---|---|---|---|---|---|---|
| $332,000 | 4/2 | 2,352 | 1988 | 4.5% | 6.1% | 98 | 1.54 | Buy |
| $310,000 | 3/2 | 1,586 | 1988 | 3.6% | 7.5% | 98 | 2.09 | Buy |
| $386,000 | 3/2 | 2,299 | 2003 | 4.3% | 5.7% | 97 | 1.77 | Buy |
| $200,500 | 3/2 | 1,444 | 1945 | 5.1% | 10.9% | 97 | 1.38 | Buy |
| $255,000 | 3/2 | 1,474 | 1984 | 4.0% | 8.6% | 96 | 2.24 | Buy |
| $270,000 | 4/3 | 2,862 | 2005 | 4.5% | 7.1% | 99 | 1.30 | Pass |
| $378,000 | 4/3 | 2,485 | 1986 | 4.0% | 5.8% | 99 | 0.44 | Pass |
| $290,000 | 4/3 | 2,154 | 1960 | 3.3% | 7.6% | 96 | 0.82 | Pass |
One Property, Two Verdicts
The power of Lens isn't just speed—it's discrimination. Two properties can look identical on surface metrics but diverge sharply on risk-adjusted returns.
Results
| Before | After (with Geofi) | |
|---|---|---|
| Properties screened / analyst / day | 30 | 400+ |
| Post-close rent vs. underwrite | ±8% | ±2.4% |
| Underwriting confidence intervals | Baseline | 60% narrower |
How Firms Adapt
Markets shift. Conditions change. The firms that perform are the ones that adjust quickly, see clearly, and execute without friction.
Forecast updates weekly, not quarterly. Lens rescans your markets overnight. When a neighborhood tips or a deal surfaces, you see it in the morning — not after close.
Tract-level awareness instead of ZIP averages. Distributions instead of point estimates. Calibrated predictions you can actually trust.
The workflow inverts. Analysts review scored opportunities instead of building spreadsheets. IC gets recommendations with assumptions explicit and uncertainty quantified.
The result: more capital deployed, into better properties, faster. Same team, less grind, higher conviction.
Limits
- Forecast use-cases. Prediction accuracy degrades the further out you look. The real value isn't perfect clarity -- it's adapting faster and smarter. Knowing when to step on the gas, when to brake, and why. Our models show their confidence and explain their reasoning. If we can't validate it, we don't publish it.
- Lens coverage. Lens handles the repeatable 85% of underwriting. The judgment calls — renovation complexity, deal dynamics — stay with your team. We build machines to amplify your process and your team.
- Error exists. We're wrong sometimes. The system learns from every decision, compounding knowledge and understanding over time. The claim is calibration, not clairvoyance.