A knowledge-graph audit of the Cloze help center, demo transcripts, and MCP docs.

Where I think this is going.

Not a roadmap. Just the bets I'd make. You're probably already thinking about some of these, and that's kind of the point: so am I.

1. Reach you rent is breaking. Reach you own is the business. Lead referral fees routinely run 30 to 40% of a commission, Zillow's are now the subject of a federal class action, and just weeks ago Chicago agents watched roughly 43,000 listings vanish from Zillow overnight when MRED cut the listing feed. Meanwhile the social and email feeds are filling with AI content and attention keeps retreating into group chats and small lists. When rented reach gets this expensive and this fragile, the only thing an agent truly owns is their relationships. Cloze already built the rooms this future needs: Segments, Stages, Tags, and Audiences let an agent reach into them and make each message personal. That isn't drip marketing. It's being personally present at scale, which has been the Cloze ethos all along. That's the shift: the CRM becomes less a system of record and more a context engine, the place an agent reaches out from, not just the place they look things up.
2. The architecture bet is about to pay off, again. Apple and Google are moving serious AI onto the device, especially on phones. When AI does run in a CRM today, it almost always runs in the cloud: every insight a round trip with someone's client data in it and a metered bill attached. As that compute moves on-device, the architecture that already keeps each user's data in its own isolated store is the one positioned to meet it. Cloze made that bet over a decade ago, while everyone else pooled tenants into one shared database. That isolation means each agent's world is already a self-contained unit, the shape an on-device model can reason over without untangling one customer from the rest. Run more of the intelligence right there and the privacy story sharpens from “your data doesn't train anyone or get sold” to “the AI can work without it ever leaving the device,” a real chunk of inference cost comes off the metered cloud, and the everyday actions get fast.
3. Make the records the brokerage owns worth owning. So much agent data is half a contact: a name and an old email, or a spouse's phone on the wrong record. The Connected Brokerage already names the orphaned-client problem. The next step is making those records worth reassigning: opt-in enrichment at the brokerage level, run on contacts the brokerage already owns, before an orphaned client is reassigned or a site lead is routed. The agent's data stays the agent's. The brokerage's data starts pulling its weight. And generating revenue.
4. A brokerage-aware adoption assistant. I built a public AI that web-searches a visitor's company and answers in context. The same pattern could give each brokerage a first-layer assistant that knows their config and their plan, so an agent gets “yes, you have that, here's how” instead of a help article. This one I can show you working today.