Your First Conversation With a Client Is No Longer Yours

When I worked as a real estate broker, I had one permanent problem: leads piled up unanswered.

Thirty to forty of them arrived during a single duty shift. Each one had to be called, logged in the CRM, and the outcome of the conversation recorded. On top of that sat the tasks left over from previous shifts.

Here is how it collapsed. A task is scheduled for ten in the morning. I arrive at the office, a client calls, and I spend an hour on the phone. My schedule has moved by an hour, and in that hour several tasks have gone overdue. While I clear them, new ones pile up for exactly the same reason they did yesterday. Then you start to drown.

Sometimes a client would fly in and three days would disappear from my calendar — I was driving him around viewings. Three days out of the office meant a hundred and sixty overdue tasks.

Note what causes the collapse. It is not laziness. It is success. The better I worked with one client, the more completely I lost all the others.

And then comes the moment when you simply abandon those tasks and those people. Not because you decided to. Because you no longer remember what you discussed or where you left off — he called while you were driving, told you something new, and it was gone. You come back to the office and find two hundred overdue leads, with fresh ones still arriving.

This produces no results. None at all.

One deal out of a hundred conversations

In my own practice it looked like this: out of a hundred people I messaged and spoke to, one reached a transaction.

The ratio itself never bothered me. The problem is that your attention is spread evenly across all hundred. You cannot know in advance which conversation is the one that matters, so each of them gets an hour of your time. And you only have so many hours.

This is where the machine belongs. Not to reply faster and around the clock — that is a pleasant side effect people talk about far too much. It belongs there to filter.

An agent that sells loses the client

I built the first version of my agent for incoming leads not in real estate, but in my other business — a two-day intensive where people build working AI tools of their own. The mechanics transfer directly, so let me describe them honestly, including the part that broke.

The agent handled incoming messages on Instagram. The first thing I saw in those conversations was this: people asked for the price in their very first message, and then disappeared.

The agent answered them honestly. And that answer ended the conversation.

Someone who opens with the price is not yet a client. They have no context in which that number means anything. Any figure named at that moment sounds either expensive or cheap, and both readings close the dialogue.

So I turned the agent around. It does not name the price in writing at all, and the goal of the exchange is to schedule a call. Not to sell. To hand the person over.

Anyone who sells something people think about before buying will recognise this instantly. "How much is it?" as an opening line is the same conversation in every industry.

An instruction is not a safeguard

Then came the mistake that made me want to write this.

I wrote a rule for the agent: never state the price. Clients extracted it anyway. They simply kept pressing, and the agent gave in — because the price was sitting in its knowledge base.

I had to go in by hand and delete the price from the knowledge base.

The conclusion was not obvious, which is exactly why it is worth something. A prohibition does not hold if the forbidden thing is stored inside the agent. A "do not say this" rule survives until the first persistent person on the other end. Only one approach works reliably: remove the data. An agent cannot say what it does not know.

For anyone putting a machine in front of their customers, this means something very practical. Everything you handed your assistant "just in case" — price lists, discount ranges, internal notes, what your supplier actually charges you, the real position of the person you are negotiating for — it will eventually say to a customer. Not because it malfunctioned, but because someone asked insistently. A system like this has to be designed around what it must not know.

Assisted, not autonomous

There is a fashionable image: the agent works through the night and a finished result is waiting for you in the morning. It is an attractive picture, and it describes something other than what most companies need today.

Mine works differently. I take the contact straight out of the conversation: I write a single message into that thread and the agent goes quiet. From there I continue myself.

Switching costs one message, not a configuration change. That is the whole difference between an assistant and a replacement.

What cannot be delegated

Some buyers absolutely need to speak to a living person. Others do not. It depends on the customer, and the range is enormous.

So: deciding on the buyer's behalf whether he needs a human is not the agent's call. Its job is to find that out and pass it on. It never takes over the personal conversation itself when the customer wants one.

And when the customer does not want one, and the agent is genuinely good — it reads the request, it knows what to establish and how to speak — then almost everything can be handed to it. The ceiling here is not technological. It is set by how ready a particular buyer is to accept a decision with no human behind it.

What I am not claiming

I have been doing this for a few months, so let me be straightforward about where the machine loses.

It answers only within its knowledge base, and mine is incomparably wider. That base has to be extended and retrained constantly, and it will stay that way for a long time. Anyone expecting an agent to work out of the box will be disappointed quickly.

And I will not claim that better-quality people now reach me. I started recently and I have not measured that. The one result I am prepared to stand behind is this: the call is scheduled and the contact is captured without my involvement.

For someone who once had two hundred overdue leads waiting at the office, that is enough to believe the direction is right.

Where to start

The first question here is not about tooling. It is about you: which part of your conversations should never have reached you in the first place?

The answer determines what is worth automating — with your own hands or with a ready-made service, that is a matter of taste and budget. But for as long as you answer everyone yourself, every successful customer will keep costing you all the others.