The most useful rule about AI tools for real estate investors is a restriction, not a permission: only build AI into the parts of the business that generate revenue. Everything else is a demo. Charles Blair, who has been a full-time investor for 35 years and runs Chucky Buy Lucky Houses out of Baltimore, estimates about 75% of his business is now touched by AI, and none of it went in because it was impressive.
He also stopped paying monthly for data platforms. His team built 13 in-house apps with Claude and ChatGPT, including a driving-for-dollars app and a bankruptcy app that surfaces relief-of-stay filings, and cut from six virtual assistants to two over roughly four years.
Below is the build-versus-buy decision that produced those tools, the headcount math behind it, and the underwriting thresholds and verification steps he still runs by hand before an offer goes out.
Key takeaways
- Apply one filter before adopting any AI tool: does it touch lead generation, outreach, screening, or offers? If it doesn’t touch revenue, skip it.
- Custom builds beat PropStream or Batch Leads subscriptions when your data need is narrower than what vendors sell — Blair’s bankruptcy app pulls relief-of-stay filings he says can’t be pulled the same way through PACER.
- Automation took Blair’s team from about six VAs to two over roughly four years while still closing three to five deals a month.
- Never treat AI output as final where numbers are involved. Blair’s go/stop/hold screening tool still gets every number and formula human-verified before an offer goes out.
- His MAO ranges: 60% to 70% of ARV on wholesale depending on jurisdiction, 80% on BRRRR, with subject-to and master lease as alternate structures and two to three exits identified per deal.
From the Real Estate Pros Show
This article draws on an interview with Charles Blair of Chucky Buy Lucky Houses / Mad Marketing Success on the Real Estate Pros Show, hosted by Meghan Escobar.
Where AI Tools for Real Estate Investors Actually Pay Off
The test is whether the tool touches revenue. Blair closes three to five deals a month across Maryland, DC and Virginia, plus wholesale partnerships in Houston, Ohio and Jacksonville, and says roughly 75% of the business now runs through AI in some form. None of it got there because it was interesting.
The failure mode he sees most often is what he calls the shiny new toy. An operator gets access to a model, generates something clever, and mistakes it for progress. His example: you can make an image of a man riding a pig down the street, and it will not make you a dollar.
Revenue-touching work in an acquisitions business is a short list, and it looks about the same in every shop:
- Lead generation and list building
- Seller and buyer outreach, plus follow-up sequences
- Deal screening and offer preparation
- Internal handoffs where deals stall between steps
Everything else can wait. The other half of his framing matters just as much: adoption is a dial, not a switch. You control how much of the business AI touches, and it does not have to touch all of it. A two-person acquisitions team that automates nothing but follow-up has still made the right call, because follow-up is where deals leak. Blair describes his own goal in exactly those terms — using systems and automation to stabilize the leakage of deals rather than to reinvent the whole operation at once.
Start with the one workflow that costs you closings today. Ship it. Then pick the next one.
Build vs Buy: Why He Stopped Renting Data Platforms
Blair’s team built 13 in-house apps for their own business using Claude and ChatGPT rather than subscribing to the platforms that sell similar functionality. Two examples show where the build makes sense.
The first is a driving-for-dollars app. A team member pulls up in front of a house, takes a picture, and the app captures the photo and pulls the property data behind it. Nothing exotic — it just does exactly what his process needs and nothing else.
The second is more instructive. They built an app that pulls bankruptcy data, specifically properties with relief-of-stay filings. Blair’s point is that this is not something you can pull the same way through PACER given how that system is set up, and it is not something you buy off the shelf from PropStream or Batch Leads. The lead source existed; the tool to work it did not.
That is the build-versus-buy line. Buy when your need is the same as everyone else’s, because a vendor has already amortized the cost across thousands of users. Build when your need is narrower or more specific than what vendors sell — a niche list nobody packages, a screening logic unique to your buy box, a workflow that only makes sense in your market.
Two honest caveats. Custom tools still need a data source, and you are responsible for its accuracy. And someone on your team has to own the thing after launch. The monthly subscription you cancelled included maintenance; your build does not.
If you’re going to bring AI into your business, what you have to bring it into is the part that allows you to generate more revenue. Get away from the shiny new toy. Find the actual tools you can use to generate profit, and if it doesn’t do that, move on and find another direction.
— Charles Blair, Chucky Buy Lucky Houses
Headcount Math: Six VAs Down to Two
Blair started moving toward AI about four years ago. Over that period his team went from roughly six virtual assistants to two, because automation absorbed a large share of what the VAs had been doing. Deal volume did not drop: he still runs three to five a month in his home market, plus co-wholesaling partnerships in three other markets where his company generates the lead, sends it to a local partner, and splits the deal.
The work that moves first is the work that is repetitive and rules-based. Pulling and formatting lists. Cleaning skip-trace returns. Data entry between systems. First-touch and follow-up sequences that run on a schedule regardless of what the seller says. Appointment confirmations. None of that requires judgment, which is exactly why it burned VA hours.
What does not move is anything requiring a decision or a real conversation. Talking to a motivated seller, negotiating a repair credit, handling a title problem, deciding whether a deal is worth a second look — those stay with people. The remaining headcount ends up handling exceptions and human contact, which is a better use of a good VA than list hygiene ever was.
Two practical notes before you cut anyone. First, sequence it the other way: automate the task, run it in parallel for a few weeks, then reassign the person. Second, going from six to two is a result, not a target. Blair got there over four years across an entire operation, not in a quarter.
Check But Verify: Keeping AI Out of Your Underwriting Errors
The single most common mistake Blair sees investors make with AI is treating the output as absolute. His rule is check but verify, and it applies hardest to anything numeric.
His team built a deal-screening tool that breaks a deal down and returns a verdict: go, stop, or hold up. It is useful precisely because it is fast and consistent, and it is dangerous for the same reason — a confident-looking output invites you to skip the step where you confirm it. So every number and every formula the tool produces gets verified by a human before an offer goes out.
Build the verification into the process rather than relying on discipline. A few things worth checking on every screened deal:
- ARV comps, pulled independently rather than accepted as stated
- The repair number against an actual scope, not a per-square-foot guess
- The arithmetic itself — confirm the formula the tool applied is the one you intended
- Any figure the model could not have sourced, which is where fabrication shows up
The right mental model is that the tool is a screener, not an underwriter. It exists to kill obvious no-deals quickly and flag the maybes, so your experienced people spend their attention on the short list. Blair has 30-plus years of pattern recognition behind that tool, and he still checks it. An investor with three years of comps in their head should check it harder.
The Underwriting Thresholds Behind the Automation
The automation sits on top of fixed numbers. Blair runs the maximum allowable offer formula on wholesale deals at anywhere from 60% to 70% of ARV, and which end he lands on depends on the jurisdiction. On a BRRRR he goes to 80%. Those spreads are not sentiment; different counties in the Maryland, DC and Virginia footprint carry different transfer costs, holding timelines and buyer appetite, and the offer has to absorb that.
Cash-price MAO is not the only structure on the table. He also writes subject-to deals and master lease deals, and those get evaluated one at a time based on the cash flow and the total dollars going into the transaction rather than a single percentage.
The discipline that ties it together is what he calls transaction engineering: every deal that crosses the table gets at least two to three exit strategies identified before he commits. A property that only works one way is a property that fails if that one way closes. He has had deals where the planned exit did not work and the deal survived because a second structure was already mapped.
His operating formula is blunter: action plus offers equals profit. If a task you perform today leads to an offer going out, repeat it tomorrow. If it does not, question why it is on the calendar. That is also the cleanest test for whether an AI tool belongs in your business — does it produce more offers, or just more output?
Lead Generation Stack: What AI Supplements, Not Replaces
The traditional channels still run underneath everything. Blair’s deals come from online ads, direct mail and cold calling, with AI built behind those channels rather than in place of them. He also gets a steady share of deals from students in his Maryland education community, which is a reminder that a real network is a lead source most investors under-build.
His framing on why the marketing side matters is direct: you are not an investor, you are a marketer. He built a marketing company alongside the investment business about eight years ago for that reason, and it is the engine that lets him co-wholesale in markets where he has no team on the ground. His company generates the lead, sends it to a partner in Houston, Ohio or Jacksonville, and splits the deal.
The funding reality is the constraint worth planning around. Blair has a private lender relationship going back more than 30 years, so his own deals get funded without friction. What he sees among newer investors in his region is different: if it isn’t a home run deal, it isn’t getting funded. Hard money is available, but the underwriting has tightened to the point where marginal deals do not clear.
The practical read for anyone scaling acquisitions right now: your AI stack should be pointed at generating more at-bats so you can afford to pass on everything that is merely acceptable. Volume of offers is what makes a strict buy box survivable.
Frequently asked questions
Is it realistic to build your own lead-generation software with AI instead of paying for PropStream or Batch Leads?
Yes, when your data need is narrower or more specific than what the vendors sell. Blair’s team built 13 in-house apps with Claude and ChatGPT, including a bankruptcy tool that pulls relief-of-stay filings, which he says isn’t available the same way through PACER or from the standard platforms.
Where the vendors still win is commodity data every investor uses. You are paying them to maintain the pipes and the accuracy. Build for the niche list nobody packages; buy for the list everyone needs.
Which tasks should a small investment team automate first with AI?
Start with repetitive, rules-based work that touches revenue: list pulling and formatting, data cleanup between systems, and first-touch plus follow-up sequences. Follow-up is usually the highest-return automation because that is where deals leak.
Leave anything requiring judgment or a real conversation with people. Seller negotiation, repair scoping, title problems and go/no-go decisions are not the place to start.
How do you stop AI from producing wrong numbers in your deal analysis?
Treat the output as a screen, never as a conclusion. Blair’s rule is check but verify: every number and every formula his go/stop/hold tool produces is confirmed by a human before an offer goes out.
Build the check into the workflow rather than trusting discipline. Independently pull comps, price repairs off an actual scope, and confirm the tool applied the formula you intended.
What MAO percentages does an experienced wholesaler actually use?
Blair runs 60% to 70% of ARV on wholesale deals, with the exact figure depending on the jurisdiction, and 80% on a BRRRR. Local transfer costs, timelines and end-buyer demand drive where in that range a given county lands.
He also writes subject-to and master lease deals, which get underwritten deal by deal on cash flow and total dollars in rather than a single percentage of value.
Does adding AI mean firing your virtual assistants?
Not immediately, and not all of them. Blair’s team went from about six VAs to two, but that happened over roughly four years as automation gradually absorbed the repetitive work, not as a single cut.
The safer sequence is to automate the task, run it alongside the person for a few weeks to confirm the output holds, then reassign that VA to exception handling and seller contact.
The bottom line
Pick the one workflow that is costing you deals right now, build or buy a fix for that single thing, and put a human verification step on any output that feeds a number into an offer. That is the whole sequence, and it works whether you end up with one tool or thirteen.
