There's a version of AI in real estate that genuinely helps, and a version that quietly damages the relationships it was bought to protect. The difference between them has very little to do with how advanced the technology is. It's about what you let it decide.
#Two buckets
More or less everything a CRM does with AI falls into one of two categories.
The first is recall and preparation. Remembering that a client's lease ends in March. Drafting a reply to an email you haven't read yet. Pulling every note from the last eighteen months before you get on a call. Flagging which relationships have gone quiet.
The second is judgement and timing. Deciding whether the message goes out at all. Picking the moment. Reading whether somebody actually wants to hear from you this week.
AI is overwhelmingly better than a busy human at the first bucket. At the second it is reliably, and occasionally catastrophically, worse.
#Why the second bucket goes wrong
An automated system has no idea what happened yesterday.
It doesn't know your client's father died on Tuesday. It doesn't know the deal fell through this morning, or that the couple you've been working with have just separated. It knows the cadence says day fourteen, and on day fourteen it will send.
A cheerful check-in during a family emergency isn't thoughtful. It's tone-deaf, and it gets remembered for a lot longer than any of the well-timed messages that came before it.
That asymmetry is the whole point. Consistent follow-up earns trust slowly. One badly-timed automated message can spend most of it at once.
#The actual tell
When somebody says they can always tell when a message is automated, they're rarely reacting to the writing. Generic copy is a mild irritation and most people barely register it.
What they're reacting to is timing that ignores context. The message that arrived as though nothing had happened, because as far as the system was concerned, nothing had.
#Where the line sits
A split that holds up in practice.
Let it handle:
- Remembering dates, anniversaries and lease expirations
- Drafting first-pass replies in your voice
- Summarising a long thread before a call
- Flagging relationships with no contact in ninety days
- Logging what happened and when
- Answering the calls you genuinely can't take
Keep for yourself:
- Whether a message goes out at all
- Anything at all after bad news
- Price conversations and negotiation strategy
- The first contact after a long silence
- Anything involving conflict, disappointment, or a mistake you made
The pattern underneath is fairly simple. Automate the remembering, keep the deciding.
Technology should make you more present, not more prolific.
#The confirmation step is the design
The most important feature in an AI-assisted CRM is a boring one. It asks before it acts.
A system that drafts a message and then waits is a system you can hand your entire client list to. A system that sends on its own is one you have to audit constantly, and auditing eats more attention than writing the messages yourself would have.
This is also why irreversible actions deserve a different standard from reversible ones. Getting a draft wrong costs you ten seconds. Sending the wrong message to two hundred past clients costs you relationships you spent a decade building.
#What good looks like
It's unglamorous to describe. Every morning there's a short list: who to reach out to, why, and a draft for each one. You read four of them, adjust two, skip one because you know something the system doesn't, and send the rest.
The skipping matters more than it sounds. A system that makes skipping easy is one that respects your judgement. A system that makes it hard, or that has already sent, has quietly replaced it.
That's the test worth applying to anything in this category. Not how much can it do without me, but how easy does it make it for me to say no.