We charged 5k for an AI system that's made our client 490k in commission

A real estate client of ours in Canada had 60k+ leads sitting in his CRM, some going back to 2017. He'd paid for every one of them at some point. He followed up a few times like everyone does and then moved on to the fresh ones. I think every agent has a version of this. The leads that don't convert in the first couple of weeks go into the CRM and basically never come out. For context we run an automation agency and have been at it 8 years. He came to us knowing there were deals buried in there, and there was no way he was paying ISAs to call 60k people. What we built is a text AI and a voice AI. They both have the same job, which is finding out where each person is at right now. Plenty have already bought, and a lot are stuck on "maybe next spring." The first text was easy to say no to on purpose. "Hey {first name}, it's {agent} with {brokerage}. Are you still thinking about a place in {area} or did plans change?" Every conversation gets summarized and tagged in a dashboard, and anyone who sounds ready gets flagged for him to call. Before anyone asks about consent, we only contacted people whose express consent he had on file from his lead forms, and anyone who said stop was out of the system the same day. The voice AI also says it's an AI in the first few seconds of every call. Here's roughly what came back over about seven months. About 11k of the numbers were dead or fake, and after also pulling anyone without consent on file we had about 44k left to contact. Around 9,100 replied or picked up and actually talked. About 3,200 had already bought with someone else and 1,700 said they were planning to move in the next year. Only 480 were ready to talk right now. That's turned into 210 appointments and 34 closed deals so far, with another 15 in progress. Those 34 are about 490k in commission for him. The "already bought" group is the one I almost threw away lol. They're homeowners now though, which makes them his next listings. So they went into their own list with the purchase date and the area. The "maybe next year" people are the ones agents lose track of. Someone tells you "spring" in October and by March nobody remembers. The AI asks when and the system books the follow-up for then. So whoever said spring gets a message in spring. I think keeping the AI's job small is a big part of why this worked. Its whole job was finding out where someone was at, and everything after that was him. People are a lot more relaxed talking to something that's just checking in. On cost, working through that list with ISAs would have meant four full-timers for about six months. We charged him 5k for the build, which was way too cheap in hindsight tbh. Add around 6k in texting and call costs and the whole thing came to roughly what one ISA would cost for two months. Plenty went sideways too. In the first week the voice AI called a woman who'd bought a house through him two years earlier and asked if she was still looking. His CRM had her twice. After that we matched every lead against his closed deals before anyone got contacted. A few hundred people also replied in French and the first version kept answering in English, which went about how you'd expect. It switches now. Anyway, if you've got a few thousand leads you stopped calling, that's probably the cheapest pipeline you have. You already paid for it. TLDR: a client in Canada had 60k+ dead leads in his CRM. We built a text AI and a voice AI to find out where each person was at and flag the ready ones for him, and it's turned into 34 closed deals so far. Keep the "already bought" and "maybe next year" people, since they're your next listings and your next buyers.