AI Deal Scoring System Tested on 2K+ Huntsville Deals — Feedback?

Question

I built an automated system to analyze 2-4 unit deals at scale. Wanted to share the methodology and get feedback from experienced operators.

**The Process:**

1. Scrape Redfin for 2-4 unit listings under $220K (Huntsville, AL market)

2. Pull rent comps via RentCast API

3. Score each deal on:

- Cap rate (target: 8%+)

- Cash-on-cash return (target: 10%+)

- DSCR (minimum 1.25x)

- Price-to-rent ratio

4. Flag risks: deferred maintenance, high vacancy, overpriced comps

5. Output ranked list with conservative/optimistic scenarios

**Example Deal Analyzed:** - Duplex in Huntsville area - $185K | 2 units | 2bd/1ba each - Est. rent: $1,100/unit = $2,200/mo total (based on RentCast comps) - Cap rate: 9.8% | CoC: 12.4% (20% down, 7% rate) - Score: 82/100

**Question for the community:**

What metrics do you prioritize when analyzing small multifamily? Am I missing anything critical in my scoring model?

I'm testing this across multiple markets (Huntsville, Memphis, Indianapolis). If you're actively buying 2-4 units and want to see what the system surfaces in your market, happy to share examples.

Bo Smith · Answer

This is solid methodology. Youre thinking like an operator -- building a repeatable system instead of picking deals by gut feel. Your metrics are the right ones: cap rate, CoC return, DSCR. Thats what lenders care about and what actually pays out in the real world.

One blind spot Id flag: the RentCast comp data is usually optimistic. The rents youre finding in their comps might be for recently renovated units or managed buildings. A 2-unit in Huntsville that youre buying at $185k distressed probably wont command $1,100/unit once stabilized -- you might hit $900-950. The 82/100 score drops fast when you plug in conservative rent assumptions. Id recommend pulling last 6-12 months of actual 2-4 unit lease signings from property management companies in that submarket, not automated aggregators.

Also worth adding: DSCR as stated is 1.25x at origination, but in a rate-down scenario (if rates drop 200 bps), your margins compress or profits improve. And seasonal absorption matters -- when are your target properties actually selling in Huntsville? January flips move different than July.

What happens to your scoring when you run conservative rent ($900/unit) instead of the API comps?

Drew Sygit · Answer

How are you accounting for Neighborhood Class?

This will impact vacancy, rents, tenant pool performance, etc.

Andrew Foord · Answer

• Good question — neighborhood class is a first-order variable and it absolutely impacts the assumptions. I use conservative baseline reserves (8% vacancy, 8% management, $150/mo maintenance, $100/mo capex) and then adjust up based on neighborhood reality: C+ to B- neighborhoods (my target zone for first deals): Baseline holds or I haircut to 10% vacancy if turnover risk is elevatedRent validation: Rentometer + Zillow + actual comp listings, then I discount 5-10% for safety marginTenant screening: 620+ credit, 3x income-to-rent, landlord refs — and I assume higher turnover than A-class C or worse: I avoid these for first deals (can't absorb chaos while building systems)If I were buying there: 15%+ vacancy, stricter capex reserves, mandatory local PM
• Baseline holds or I haircut to 10% vacancy if turnover risk is elevated
• Rent validation: Rentometer + Zillow + actual comp listings, then I discount 5-10% for safety margin
• Tenant screening: 620+ credit, 3x income-to-rent, landlord refs — and I assume higher turnover than A-class
• I avoid these for first deals (can't absorb chaos while building systems)
• If I were buying there: 15%+ vacancy, stricter capex reserves, mandatory local PM
• A/B neighborhoods: Lower vacancy (6-7%), but purchase price compression kills cash flowI only pursue if creative financing unlocks the deal (seller finance, assumable debt, etc.) Bottom line: If a property only pencils with optimistic vacancy assumptions (e.g., 5% in a C neighborhood), I pass. The deal has to survive worst-case reality, not best-case projections.
• Lower vacancy (6-7%), but purchase price compression kills cash flow
• I only pursue if creative financing unlocks the deal (seller finance, assumable debt, etc.)

• Baseline holds or I haircut to 10% vacancy if turnover risk is elevated
• Rent validation: Rentometer + Zillow + actual comp listings, then I discount 5-10% for safety margin
• Tenant screening: 620+ credit, 3x income-to-rent, landlord refs — and I assume higher turnover than A-class

• I avoid these for first deals (can't absorb chaos while building systems)
• If I were buying there: 15%+ vacancy, stricter capex reserves, mandatory local PM

• Lower vacancy (6-7%), but purchase price compression kills cash flow
• I only pursue if creative financing unlocks the deal (seller finance, assumable debt, etc.)

Andrew Foord · Answer

Bo you nailed the weak point. RentCast is definitely ceiling not floor — I've been treating it as market rate and it's probably 10-15% optimistic on distressed stock. When I rerun that duplex at $900/unit the score drops to somewhere around 61/100 — still positive but no longer a green flag deal at that price point. Which honestly is the right answer.

The lease signing data from local PMs is a better source. I'm going to build that into the methodology as a manual verification step before anything gets flagged as a buy.

The seasonal absorption point is one I hadn't modeled at all. Do you have a sense of the Huntsville market cycle? When are motivated sellers actually moving inventory?

And the rate sensitivity layer is a good add — I'll run a +/- 200bps stress test on DSCR going forward. Appreciate the operator-level feedback.

Kenneth Garrett · Answer

@Andrew Foord

I use a different philosophy as I use the BRRRR method. I evaluate the purchase price + rehab + holding cost. I then use rentometer for rents and also use Zillow and apartments.com for a clear picture. Then calculate ARV. The goal of course is it to have no money stuck in. That is not always possible, but it is the goal. See example 2 unit:

$145K = Purchase

$40K = Rehab

$15K Holding Cost (utilities, taxes and private loan)

$200,000 Total Cost

$265,000 ARV

$2400 Rent

$265K x 80% = $212,000 actually borrow $200K (approximately 75% of ARV - Actual 75.4%

$1264 P&I

$300 Property Taxes

$100 Insurance

$1664 Total Rent is $2400. Add in vacancy, CapEx and maintenance 15% ($360) in general, but I already rehabbed it so Capex and maintenance are low, but I included it anyway.

Total cost = Rent $2400 - expenses $2024 cash flow today $376. Not a home run, but over time this will cash flow over $500 in approximately 4 years at a modest 2% increase each year. Remember tenants are paying your mortgage pay down, taxes, insurance, etc. Do this once or twice a year in five years it’s a nice flow plan. Not including appreciation which I consider the bonus. Great way to build generational wealth.

Andrew Foord · Answer

I ask about the markets you’re interested in because understanding your focus can help me align my strategies better. It’s valuable to know what’s performing well for experienced property managers like yourself!

Andrew Foord · Answer

Thanks for sharing, Drew!

I completely agree that having a strong handle on DOM is essential for informed decision-making, especially in the property management business. Using a standard metric as a filter is smart, and then diving deeper into the actual DOM is key.

Your approach highlights the importance of local knowledge, which can significantly impact rental strategies. I'll definitely keep this in mind when evaluating potential new markets.

Do you have any specific markets you’re currently focusing on?

Appreciate the insights!

Drew Sygit · Answer

Hmmm, why do you ask which markets I'm interested in?

Andrew Foord · Answer

Great points, Derek. You’re absolutely right that relying on active listings can distort the picture, especially when fast-leasing properties skip the scraping cycle.

I also pull closed lease comps from the MLS where available, as that data provides real market validated prices. The gap between asking and achieved rents in C or B- neighborhoods can be significant, and I factor that into my underwriting.

Neighborhood class is indeed tough for algorithms to capture accurately. Local agents often have insights that those data points miss. I make it a point to build relationships with local experts who can give me a clearer picture beyond just numbers.

Not every market has the density of MLS lease data, but where I can access it, I agree it’s a far better foundation for assessing potential performance.

Thanks for sharing your methodology! It’s always insightful to hear how others navigate these complexities.

Derek Hollis · Answer

The RentCast ceiling issue Bo flagged runs deeper than the percentage. The problem is structural. RentCast and most aggregators pull from active listings, not closed leases. What you're benchmarking against is inventory that hasn't moved yet, meaning you're modeling off asking prices on properties that sat long enough to get scraped. Fast-leasing properties in good condition rarely make it into that dataset. On distressed stock in C and B- neighborhoods, the gap between asking and actual achieved rent can be 20%+ in my market, and Nashville isn't exactly known for distressed rental inventory.

The fix I use when underwriting for myself or investor clients is pulling closed lease comps from the MLS rather than trusting any of the aggregators. That data shows what properties actually leased for in the last 90 days, not what someone is hoping to get. Not every market has dense MLS lease data but if yours does, it's a better floor than RentCast.

Drew's point on neighborhood class is the one I don't think any algorithm gets right. I've looked at Zillow's neighborhood classifications and they lag reality by years, don't distinguish between the good and bad blocks of the same zip, and won't tell you that the street two blocks over turns every 8 months while the one you're looking at has had the same tenants for three years. A local agent who works with investors will know that. Automated scoring won't. Worth building a "thin data" flag for markets where you don't trust your rent comps enough to let the score drive the decision.

AI Deal Scoring System Tested on 2K+ Huntsville Deals — Feedback?

AI Deal Scoring System Tested on 2K+ Huntsville Deals — Feedback?

Most Popular Reply

14 Replies