An AI agent that finds leads by reading every Google review

Full walkthrough

This is a case study of a project I did for one of my clients. He's a co-founder of a US-based SaaS company that builds solutions for hotel renovation contractors, and he wanted to expand his offering with AI. In this video I'll demo how the AI finds leads for hotel renovation contractors, show what it found, how many results and at what cost, and share a small trick I used to make the AI more trustworthy for this specific job.

The manual process

If you're a hotel renovation contractor in Los Angeles, an ideal lead looks something like a small, two-star motel that needs a renovation - and, ideally, needs it urgently. To find leads like that today, you'd usually hire someone to search by hand, and the process looks like this. You search for two-star motels in LA, open one, and read its Google reviews looking for signals of outdated, worn furniture. You read through them one by one, and if enough reviews point to old furniture, you mark it as a potential lead.

Then you check the public permit records - if the last permit is from, say, 1999, it probably hasn't been renovated recently, which supports the lead. You also want to know if it's an independent motel or part of a chain, since independents make better leads, and that isn't always stated explicitly. Doing all of this by hand takes a long time to surface even a handful of qualified leads - and by the time you've processed it all, another contractor may have already reached out.

The AI solution

Here's the application I built for the client. The first step is simple: enter your search criteria - an address, a radius (say, 10 miles), and a few other settings - and start the search. The AI agent kicks off and you can watch every step it takes.

It goes to Google Maps, looks within the radius, and locates the motels. Then it classifies ownership - reading the information about each motel to decide whether it's independent or part of a chain. Next, and this is the key part, it goes out to the web and reads through the full set of reviews for each motel. That matters: if you search reviews by keyword for "old furniture," you'll miss a review written in Spanish, or one that just says "old," and you'd misqualify the lead. The AI reads all of the reviews with full context and uses that to judge whether the lead is qualified. Finally it computes a score, and you get a ranked result.

The trust trick

Here's the trick I wanted to show. Think of a student who hands in a math homework with only the answers - "a, b, c." The teacher asks, "How did you get there? What was your reasoning?" If the student can't explain it, they didn't really understand it. Same with AI. If I just ask for a score and it says "0.7," that looks suspicious and I get the urge to go double-check it myself.

So, like the teacher asking the student to explain, I have the agent justify its score. For one property it flagged multiple reviews about furniture and room condition - and it quotes the actual Google review. When I copy that quote into the reviews search, there's a real review that says the bunk beds were pretty shaky. The point of the trick is to make the AI explain why it decided what it decided. Reading that explanation increases my trust: I can see it actually did the work and really read the reviews.

Room to grow, and why ChatGPT can't just do this

This version doesn't check the permit records yet. After I finished it, I handed off to the client's team to extend it - for example, searching Facebook groups where a motel owner might post "I own this motel in LA and I need a furniture renovation." If that was posted five minutes ago, it's a hot lead you can reach out to right away. Lots of room to grow.

One last thing that's interesting. This agent actually goes live and reads all the Google reviews. What if I just ask ChatGPT to do it instead? I copy the Google Maps address, go to ChatGPT, and ask it to read all the reviews - and it tells me it can't open reviews from a link, and asks me to copy and paste them in. That's the whole point: I don't want to copy-paste, I want the AI to do the job, which is exactly what this solution does. Hopefully, even if you're in a completely different industry, this gives you a concrete example of a real AI application you can adapt to your own work.