Three AI applications are producing revenue for real estate investors right now: automated lead scoring inside the CRM, AI voice used to reactivate aged leads, and search visibility work that accounts for both Google and the LLMs. Almost everything else — nose-to-tail acquisitions bots, AI-generated comps, weekend agent-building projects — is still a distraction from the work that closes deals.
Carlos Zamora spent three years in acquisitions in Baltimore before a decade in the CRM side of the business, first with InvestorFuse and now with Carrot after the acquisition. He watches what investors actually deploy and what they abandon. His read on where the market landed after the last hype cycle, and the thresholds he uses to decide whether a tool earns its place, are below.
What follows: the scoring variables that determine call order, the lead-volume test for AI voice, the consent language problem people are getting jammed up over, and what Carrot’s data shows about where seller traffic is actually coming from.
Key takeaways
- Score leads on three things: response speed (under 24 hours beats several days beats no response), any mention of equity or distress, and source channel — direct mail and SEO leads outrank cold calling and expired listings.
- AI voice is a bad fit for a solo operator burning $400–500 PPC leads, and close to a no-brainer for a couple thousand aged CRM leads nobody was going to call anyway.
- If you run AI calling, your website consent form needs to explicitly cover text, email and AI calling. Have counsel review it.
- Google flags sites that dump 50–100 AI-written blog posts in a month. Carrot still recommends roughly one quality post per month, plus a fully completed Google Business Profile with reviews.
- Google searches from motivated house sellers are increasing, not collapsing — while ChatGPT shows up in the source code of Carrot inbound leads more every month.
From the Investor Fuel Show
This article draws on an interview with Carlos Zamora of Carrot on the Investor Fuel Show, hosted by Mike Hambright. Watch or listen to the full interview.
Where AI Actually Sits in an Acquisitions Business Right Now
The consensus has swung. Six months of Facebook posts predicting one-person real estate companies run entirely by agents gave way, by spring, to something more sober. Zamora watched even AI-voice enthusiasts pull back toward reality.
His warning to owners is about opportunity cost, not capability. You can spend nights and weekends building an agent that gets to 80 or 90 percent of what you need and never crosses the finish line — while the revenue-generating work in your business sits untouched.
“To what extent am I taking away my actual revenue-generating activities to work on this for hours at night?” is the question he thinks most operators skipped. The honest answer for a business development or acquisitions person is usually that it isn’t the best use of the time.
The second problem is structural. Mike Hambright, who grew the Investor Fuel team from 12 to 40 in eighteen months, is candid that his people are all on separate individual AI accounts that don’t talk to each other. That is not a CRM, not Slack, not Asana — it’s a dozen people dabbling in parallel with nothing compounding.
Carrot’s answer is a shared company brain everyone plugs into. Zamora uses it for things as ordinary as drafting copy in the CEO’s voice, and for something more interesting: the company keeps a Slack channel logging every customer win going back years, and the brain can pull three specific examples of an investor closing a deal from a given feature on request.
The distinction matters. An individual using ChatGPT gets faster. An organization with shared context and shared history gets an asset. Most investor teams have the first and none of the second.
AI Lead Scoring: The Variables That Decide Which Leads Get Called First
This is the application Zamora is most excited about, and it is the one with the shortest path to money. AI lead scoring ranks your database in real time so an acquisitions rep opens the CRM to a call order instead of an undifferentiated list.
The variables he names:
- Response speed. A seller who replies within 24 hours scores higher than one who takes several days, who scores higher than one not responding at all.
- Equity or distress signals. Any mention of either in the conversation pushes the score up.
- Source channel. A direct mail or SEO lead scores above a cold calling lead or an expired listing.
The workflow is where it gets useful. Send a mail piece, let the AI voice agent take the inbound conversation, and the call is transcribed and summarized automatically into the CRM — with the score updated before a human touches the record. Your acquisitions manager logs in that morning to leads already ranked by likelihood to close.
Carrot runs AI summaries and lead scoring natively and integrates with a voice tool for the conversation layer. Most competent CRMs are building some version of this.
The payoff Zamora emphasizes is not conversion rate — it’s morale. “If I was managing acquisitions people, I would want to keep their morale as high as possible.” A rep who spends the day on leads that are actually receptive burns out slower and earns more. A rep grinding through dead records quits.
Hambright’s framing: historically all leads look the same in the CRM unless someone manually tags them, so people treat them the same. They never were the same.
I can spend my nights and weekends building an agent that’s 80 to 90 percent of what I need, but I’m not going to get it across the finish line. To what extent am I taking away my actual revenue-generating activities to do that?
— Carlos Zamora, Carrot
When AI Voice Is Worth It and When It Is a Waste of Expensive Leads
Zamora’s threshold test has two inputs: lead volume and the team you already have to work leads.
The wrong application is obvious once stated. If you’re running PPC at $400 to $500 a lead and you’re a solo operator trying to get revenue in the door, handing those leads to an AI voice agent makes no sense. Those are the most expensive conversations you will have this month. You take them yourself.
The right application is equally clear. If you’ve been in business a few years and have a couple thousand leads sitting in the CRM that nobody is going to call, AI voice becomes close to a no-brainer. “It’s better to follow up with them with an AI voice, call them all, than to just let them sit there.” The comparison isn’t AI versus your best closer — it’s AI versus nothing.
On capability, Zamora is specific about where the line currently sits. Appointment setting is the sweet spot. Products marketed as nose-to-tail — lead to contract without a human — are what people want, because everyone wants one button. He doesn’t think that’s there yet.
On the B2B side Carrot uses a tool that can dial a couple thousand contacts at once and run the conversation in a recorded version of your own voice. The scale is real. The judgment is not.
His practical split: humans do the negotiating and the calling that matters, AI tees up the lead scores and clears the backlog. That allocation will shift as the models improve, but it is where the value sits today.
The Compliance Trap: AI Cold Calling and Consent Language
Zamora’s caution here is short and worth taking seriously: people are getting jammed up over AI cold calling. This is the fastest way to turn a productivity gain into a legal problem.
His practical fix is a form review. If you are making AI calls — including calls originating from inside your CRM — the consent language on your website needs to explicitly cover it. Not just “we may contact you.” The specific channels: text, email, and AI calling.
That matters most for multi-channel operators. If a lead comes in through a Carrot site, gets a text sequence, an email drip, and an AI voice follow-up, the consent box that lead checked needs to have authorized all three at the moment they checked it. Retrofitting consent after the fact is not a thing.
Two things to note. First, rules around automated and AI-generated calls are moving, and enforcement is uneven across states. Second, nothing here is legal advice and Zamora isn’t offering it as such — he’s flagging that the cheapest hour you can spend is having your own attorney read your opt-in language before you scale a dialer on top of it.
The asymmetry is what should motivate you. Updating a consent form costs almost nothing. Getting the letter costs a great deal more than any amount of aged-lead reactivation is going to earn you.
Seller Search Is Changing, But Google Still Drives the Leads
Carrot’s own data contradicts the assumption that LLMs are eating search. According to what Zamora sees, Google searches from motivated house sellers are increasing significantly — not declining. At the same time, ChatGPT shows up in the source code of inbound Carrot leads more every single month.
Both things are true at once because the overall volume of seller search is up. And there is a demographic reality behind it: if your target is a 50-plus seller, they are probably not prompting Claude. They are typing into Google, possibly into Google’s AI mode.
The tactical implications:
- Don’t mass-publish AI content. Google was flagging sites that pushed 50 to 100 AI-written blog posts in a month as probably not legitimate. Carrot’s recommendation remains roughly one quality post per month.
- Content depth still converts. Carrot’s content-heavy strategy exists because real estate, financial services and health are the categories where people want to be exhaustively informed before deciding. A house is the biggest asset 99 percent of sellers will ever own. This is not an add-to-cart purchase.
- Fix your Google Business Profile first. Hours, location, site link, and reviews. Zamora calls this the lowest-hanging fruit most investors still haven’t dialed in, and he sees people closing deals off it alone.
Use AI to find what keywords sellers are searching. Don’t use it to carpet-bomb your own domain.
What AI Still Shouldn’t Own: Comps, Offers and the Seller Relationship
Zamora expects AI deal analysis — comping and offer creation — to be the next product wave, and he’s also clear that plenty of investors are apprehensive about it. That apprehension is well-founded.
Hambright’s history explains why. When Zestimates first appeared, he refused to price off them and still runs MLS comps two decades later. Recently he listed two rental properties, both significantly above their Zillow values, and both moved. His read: if you rely too heavily on an automated number and buy one wrong, you get beat up. Sell one wrong and you leave money on the table. Neither error is recoverable by better software.
Where AI earns its place in the relationship is memory, not valuation. Twenty years ago investors were taught to note that a seller likes cats or follows the Cowboys, drop it in the CRM, and surface it on the fifteenth call a year later. That was a static note someone had to type and usually didn’t.
Now transcription and AI notes capture it automatically, along with exactly where the last conversation ended. Zamora’s point is that this is a differentiator precisely because most sellers are talking to several buyers, and most channels require multiple follow-ups before anything closes. The operator who recalls something personal on call six stands out from four competitors reading a script.
The pattern across all of it: AI handles the recall, the ranking and the backlog. Humans handle the number and the relationship.
Frequently asked questions
What data points should an AI lead scoring model use for motivated seller leads?
Three inputs cover most of the signal: how fast the seller responds, whether the conversation contains any mention of equity or distress, and which marketing channel produced the lead. A seller replying within 24 hours scores above one taking several days, who scores above a non-responder. Direct mail and SEO leads score above cold calling and expired listings.
Feed those from a transcribed conversation so the summary and score are written into the CRM before an acquisitions rep opens the record.
At what point does AI voice calling make financial sense for a wholesaling business?
When you have a large backlog of aged leads and not enough people to call them. Carlos Zamora’s threshold is roughly a couple thousand untouched leads in the CRM — at that point AI voice is competing with nothing, not with your best closer.
It does not make sense for a solo operator running $400 to $500 PPC leads who needs revenue in the door. Those conversations are too expensive to delegate to a bot.
Do I need to disclose AI calling in my website’s consent form?
If you are making AI-generated calls to leads that came through your site, your consent language should explicitly cover it — along with text and email if you use those channels. This is especially important for multi-channel campaigns where a single opt-in has to authorize several contact methods.
Investors have run into trouble over AI cold calling. Have your own attorney review your opt-in language before scaling a dialer on top of it; nothing here substitutes for legal counsel.
How many blog posts per month should a real estate investor publish without getting flagged by Google?
Roughly one quality post per month is what Carrot still recommends. Sites that published 50 to 100 AI-written posts in a single month were getting flagged by Google as probably not legitimate.
Real estate content also needs real depth to convert, because a house is the largest asset most sellers will ever handle. Use AI to identify which keywords sellers are searching, then write something substantive for each one.
Are motivated sellers actually finding investors through ChatGPT instead of Google?
Some are, but not instead of Google. Carrot sees ChatGPT appearing in the source code of inbound leads more every month, while Google searches from motivated house sellers are increasing significantly at the same time.
Total seller search volume is growing. If your target demographic skews 50-plus, most of them are still on Google — including Google’s AI mode — so your Google Business Profile and organic content remain the priority.
The bottom line
Pick the one application that matches your current constraint and deploy only that. If your reps are working an undifferentiated lead list, turn on scoring. If you have thousands of dead leads and no one to call them, run AI voice at the backlog. If neither is your bottleneck, spend the afternoon completing your Google Business Profile and asking for reviews instead — and stop building agents at midnight.
