Every sales leader has met Don Draper. The star closer who carries the whole deal in his head — the budget, the real objection, the spouse who actually signs off — and writes none of it down. While he’s winning, nobody minds. Then he goes on holiday, or he leaves, and the deal walks out with him, because it never lived anywhere but his memory.
That is the CRM problem in one person, but it isn’t really about the one star. Walk into most firms I’ve audited and the CRM is a graveyard: half-filled cards, deals parked in stages they left weeks ago, contacts with no notes. The team fills it in last, if at all — updating the CRM is the tax you pay after the selling is done, and the selling never stops.
What the empty CRM hides
Empty fields are cheap; the blind spot they create is not. When the CRM lags reality, the people who most need to read the pipeline — founder, CEO, head of sales — are steering on yesterday’s weather report.
That gets expensive at the edges of the business. A regional office can drift for a quarter or two, quietly under water, and leadership catches it late, because the numbers they steer by are a reconstruction rather than a live feed. By the time the picture is sharp enough to act on, you have already paid for the lag in salaries, rent, and the moves you didn’t make.
It’s the difference between flying with a window and flying on instruments — except the instruments are three weeks behind. You still land the plane. You just learn about the mountain later than you’d like.
We stopped fixing discipline with more discipline
Reminders, mandatory fields, “CRM hygiene” Fridays — they all fight human nature and lose. The bottleneck was never the rep; it was the manual step.
So we put the work where the information already lives: the chats. In the markets we work — Dubai, Cyprus, Bangkok — the deals get done in WhatsApp and Telegram, not in the CRM. We let an agent read those and keep the record current on its own.
How it works
Under the hood it is simpler than it sounds. Think of a diligent night clerk who reads every conversation and files it correctly before morning — the Elves and the Shoemaker, if the elves had an API.
The one part worth slowing down on is the memory. The first time the agent meets a lead it does the detective work — match the number, read the open deals, figure out who this is. It stores the answer. The second time, it already knows the thread, so it stops re-asking and starts anticipating. That is the difference between a macro and an assistant.
Where this goes next
Once the agent lives on the chats, the CRM stops being a form to fill and becomes something you talk to. Two directions we’re building toward:
- Voice, in the field. A rep at a showing or a property expo doesn’t open a CRM. They send a voice note to a Telegram bot — “met the Kim family, two kids, wants sea view under 1.5, follow up Thursday” — and it lands on the lead, transcribed and structured. The notes get richer because capturing them costs nothing.
- Errands, not just records. The same agent that files the lead can start handling the busywork around it: assemble a PDF or web proposal, shortlist properties by the buyer’s criteria, or pull together a region primer — the tax-and-residency basics for a first-time Cyprus buyer, say — so the rep walks in prepared.
This is no longer a lab demo. McKinsey’s 2026 review of agentic AI in real estate reports firms capturing incremental home sales from after-hours agents that kept engaging buyers around the clock [1]. The advantage is plain: the work stopped waiting for a human to have a free minute.
How to start — pick your depth
You don’t need our stack to feel the difference. Two versions, depending on how far you want to go.
The light version — Claude or ChatGPT, this afternoon. Make a small project and put your CRM’s field names in the instructions: extract name, phone, budget, stage, next step. At the end of the day, paste in a chat thread and let it hand back a clean, structured block you drop into the CRM. It isn’t automatic, but it turns ten minutes of retyping into ten seconds of pasting, and it forces the fields to be consistent. Teams I’ve handed this to feel the difference within a week.
The full version — for the technical. Connect an agent to the WhatsApp or Telegram account through an MCP, give it read access to your CRM’s API and a small memory store, and run the loop above on every new message. Start read-only — the agent drafts, a human approves — until you trust it, then let it write directly. The unlock is the memory: once it remembers who’s who, the whole thing shifts from typing to thinking.
A tidy CRM is a side effect
The real product is shared sight: the founder, the head of sales, and the regional manager all reading the same live picture — so the call to double down on Dubai or close the London desk gets made on this quarter’s data, not last quarter’s ghost.
[1] — McKinsey, How agentic AI can reshape real estate’s operating model (2026): https://www.mckinsey.com/industries/real-estate/our-insights/how-agentic-ai-can-reshape-real-estates-operating-model.